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>
16 #include <linux/iversion.h>
19 #include "mds_client.h"
21 #include <linux/ceph/decode.h>
24 * Ceph inode operations
26 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
27 * setattr, etc.), xattr helpers, and helpers for assimilating
28 * metadata returned by the MDS into our cache.
30 * Also define helpers for doing asynchronous writeback, invalidation,
31 * and truncation for the benefit of those who can't afford to block
32 * (typically because they are in the message handler path).
35 static const struct inode_operations ceph_symlink_iops;
37 static void ceph_inode_work(struct work_struct *work);
40 * find or create an inode, given the ceph ino number
42 static int ceph_set_ino_cb(struct inode *inode, void *data)
44 struct ceph_inode_info *ci = ceph_inode(inode);
45 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
47 ci->i_vino = *(struct ceph_vino *)data;
48 inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
49 inode_set_iversion_raw(inode, 0);
50 percpu_counter_inc(&mdsc->metric.total_inodes);
55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
59 if (ceph_vino_is_reserved(vino))
60 return ERR_PTR(-EREMOTEIO);
62 inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
63 ceph_set_ino_cb, &vino);
65 return ERR_PTR(-ENOMEM);
67 dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
68 ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
73 * get/constuct snapdir inode for a given directory
75 struct inode *ceph_get_snapdir(struct inode *parent)
77 struct ceph_vino vino = {
78 .ino = ceph_ino(parent),
81 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
82 struct ceph_inode_info *ci = ceph_inode(inode);
84 BUG_ON(!S_ISDIR(parent->i_mode));
87 inode->i_mode = parent->i_mode;
88 inode->i_uid = parent->i_uid;
89 inode->i_gid = parent->i_gid;
90 inode->i_mtime = parent->i_mtime;
91 inode->i_ctime = parent->i_ctime;
92 inode->i_atime = parent->i_atime;
94 ci->i_btime = ceph_inode(parent)->i_btime;
96 if (inode->i_state & I_NEW) {
97 inode->i_op = &ceph_snapdir_iops;
98 inode->i_fop = &ceph_snapdir_fops;
99 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
100 unlock_new_inode(inode);
106 const struct inode_operations ceph_file_iops = {
107 .permission = ceph_permission,
108 .setattr = ceph_setattr,
109 .getattr = ceph_getattr,
110 .listxattr = ceph_listxattr,
111 .get_acl = ceph_get_acl,
112 .set_acl = ceph_set_acl,
117 * We use a 'frag tree' to keep track of the MDS's directory fragments
118 * for a given inode (usually there is just a single fragment). We
119 * need to know when a child frag is delegated to a new MDS, or when
120 * it is flagged as replicated, so we can direct our requests
125 * find/create a frag in the tree
127 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
131 struct rb_node *parent = NULL;
132 struct ceph_inode_frag *frag;
135 p = &ci->i_fragtree.rb_node;
138 frag = rb_entry(parent, struct ceph_inode_frag, node);
139 c = ceph_frag_compare(f, frag->frag);
148 frag = kmalloc(sizeof(*frag), GFP_NOFS);
150 return ERR_PTR(-ENOMEM);
157 rb_link_node(&frag->node, parent, p);
158 rb_insert_color(&frag->node, &ci->i_fragtree);
160 dout("get_or_create_frag added %llx.%llx frag %x\n",
161 ceph_vinop(&ci->vfs_inode), f);
166 * find a specific frag @f
168 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
170 struct rb_node *n = ci->i_fragtree.rb_node;
173 struct ceph_inode_frag *frag =
174 rb_entry(n, struct ceph_inode_frag, node);
175 int c = ceph_frag_compare(f, frag->frag);
187 * Choose frag containing the given value @v. If @pfrag is
188 * specified, copy the frag delegation info to the caller if
191 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
192 struct ceph_inode_frag *pfrag, int *found)
194 u32 t = ceph_frag_make(0, 0);
195 struct ceph_inode_frag *frag;
203 WARN_ON(!ceph_frag_contains_value(t, v));
204 frag = __ceph_find_frag(ci, t);
206 break; /* t is a leaf */
207 if (frag->split_by == 0) {
209 memcpy(pfrag, frag, sizeof(*pfrag));
216 nway = 1 << frag->split_by;
217 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
218 frag->split_by, nway);
219 for (i = 0; i < nway; i++) {
220 n = ceph_frag_make_child(t, frag->split_by, i);
221 if (ceph_frag_contains_value(n, v)) {
228 dout("choose_frag(%x) = %x\n", v, t);
233 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
234 struct ceph_inode_frag *pfrag, int *found)
237 mutex_lock(&ci->i_fragtree_mutex);
238 ret = __ceph_choose_frag(ci, v, pfrag, found);
239 mutex_unlock(&ci->i_fragtree_mutex);
244 * Process dirfrag (delegation) info from the mds. Include leaf
245 * fragment in tree ONLY if ndist > 0. Otherwise, only
246 * branches/splits are included in i_fragtree)
248 static int ceph_fill_dirfrag(struct inode *inode,
249 struct ceph_mds_reply_dirfrag *dirinfo)
251 struct ceph_inode_info *ci = ceph_inode(inode);
252 struct ceph_inode_frag *frag;
253 u32 id = le32_to_cpu(dirinfo->frag);
254 int mds = le32_to_cpu(dirinfo->auth);
255 int ndist = le32_to_cpu(dirinfo->ndist);
260 spin_lock(&ci->i_ceph_lock);
262 diri_auth = ci->i_auth_cap->mds;
263 spin_unlock(&ci->i_ceph_lock);
265 if (mds == -1) /* CDIR_AUTH_PARENT */
268 mutex_lock(&ci->i_fragtree_mutex);
269 if (ndist == 0 && mds == diri_auth) {
270 /* no delegation info needed. */
271 frag = __ceph_find_frag(ci, id);
274 if (frag->split_by == 0) {
275 /* tree leaf, remove */
276 dout("fill_dirfrag removed %llx.%llx frag %x"
277 " (no ref)\n", ceph_vinop(inode), id);
278 rb_erase(&frag->node, &ci->i_fragtree);
281 /* tree branch, keep and clear */
282 dout("fill_dirfrag cleared %llx.%llx frag %x"
283 " referral\n", ceph_vinop(inode), id);
291 /* find/add this frag to store mds delegation info */
292 frag = __get_or_create_frag(ci, id);
294 /* this is not the end of the world; we can continue
295 with bad/inaccurate delegation info */
296 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
297 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
303 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
304 for (i = 0; i < frag->ndist; i++)
305 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
306 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
307 ceph_vinop(inode), frag->frag, frag->ndist);
310 mutex_unlock(&ci->i_fragtree_mutex);
314 static int frag_tree_split_cmp(const void *l, const void *r)
316 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
317 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
318 return ceph_frag_compare(le32_to_cpu(ls->frag),
319 le32_to_cpu(rs->frag));
322 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
325 return f == ceph_frag_make(0, 0);
326 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
328 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
331 static int ceph_fill_fragtree(struct inode *inode,
332 struct ceph_frag_tree_head *fragtree,
333 struct ceph_mds_reply_dirfrag *dirinfo)
335 struct ceph_inode_info *ci = ceph_inode(inode);
336 struct ceph_inode_frag *frag, *prev_frag = NULL;
337 struct rb_node *rb_node;
338 unsigned i, split_by, nsplits;
342 mutex_lock(&ci->i_fragtree_mutex);
343 nsplits = le32_to_cpu(fragtree->nsplits);
344 if (nsplits != ci->i_fragtree_nsplits) {
346 } else if (nsplits) {
347 i = prandom_u32() % nsplits;
348 id = le32_to_cpu(fragtree->splits[i].frag);
349 if (!__ceph_find_frag(ci, id))
351 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
352 rb_node = rb_first(&ci->i_fragtree);
353 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
354 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
357 if (!update && dirinfo) {
358 id = le32_to_cpu(dirinfo->frag);
359 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
366 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
367 frag_tree_split_cmp, NULL);
370 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
371 rb_node = rb_first(&ci->i_fragtree);
372 for (i = 0; i < nsplits; i++) {
373 id = le32_to_cpu(fragtree->splits[i].frag);
374 split_by = le32_to_cpu(fragtree->splits[i].by);
375 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
376 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
377 "frag %x split by %d\n", ceph_vinop(inode),
378 i, nsplits, id, split_by);
383 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
384 if (ceph_frag_compare(frag->frag, id) >= 0) {
385 if (frag->frag != id)
388 rb_node = rb_next(rb_node);
391 rb_node = rb_next(rb_node);
392 /* delete stale split/leaf node */
393 if (frag->split_by > 0 ||
394 !is_frag_child(frag->frag, prev_frag)) {
395 rb_erase(&frag->node, &ci->i_fragtree);
396 if (frag->split_by > 0)
397 ci->i_fragtree_nsplits--;
403 frag = __get_or_create_frag(ci, id);
407 if (frag->split_by == 0)
408 ci->i_fragtree_nsplits++;
409 frag->split_by = split_by;
410 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
414 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
415 rb_node = rb_next(rb_node);
416 /* delete stale split/leaf node */
417 if (frag->split_by > 0 ||
418 !is_frag_child(frag->frag, prev_frag)) {
419 rb_erase(&frag->node, &ci->i_fragtree);
420 if (frag->split_by > 0)
421 ci->i_fragtree_nsplits--;
426 mutex_unlock(&ci->i_fragtree_mutex);
431 * initialize a newly allocated inode.
433 struct inode *ceph_alloc_inode(struct super_block *sb)
435 struct ceph_inode_info *ci;
438 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
442 dout("alloc_inode %p\n", &ci->vfs_inode);
444 spin_lock_init(&ci->i_ceph_lock);
447 ci->i_inline_version = 0;
448 ci->i_time_warp_seq = 0;
449 ci->i_ceph_flags = 0;
450 atomic64_set(&ci->i_ordered_count, 1);
451 atomic64_set(&ci->i_release_count, 1);
452 atomic64_set(&ci->i_complete_seq[0], 0);
453 atomic64_set(&ci->i_complete_seq[1], 0);
454 ci->i_symlink = NULL;
459 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
460 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
461 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
463 ci->i_fragtree = RB_ROOT;
464 mutex_init(&ci->i_fragtree_mutex);
466 ci->i_xattrs.blob = NULL;
467 ci->i_xattrs.prealloc_blob = NULL;
468 ci->i_xattrs.dirty = false;
469 ci->i_xattrs.index = RB_ROOT;
470 ci->i_xattrs.count = 0;
471 ci->i_xattrs.names_size = 0;
472 ci->i_xattrs.vals_size = 0;
473 ci->i_xattrs.version = 0;
474 ci->i_xattrs.index_version = 0;
476 ci->i_caps = RB_ROOT;
477 ci->i_auth_cap = NULL;
478 ci->i_dirty_caps = 0;
479 ci->i_flushing_caps = 0;
480 INIT_LIST_HEAD(&ci->i_dirty_item);
481 INIT_LIST_HEAD(&ci->i_flushing_item);
482 ci->i_prealloc_cap_flush = NULL;
483 INIT_LIST_HEAD(&ci->i_cap_flush_list);
484 init_waitqueue_head(&ci->i_cap_wq);
485 ci->i_hold_caps_max = 0;
486 INIT_LIST_HEAD(&ci->i_cap_delay_list);
487 INIT_LIST_HEAD(&ci->i_cap_snaps);
488 ci->i_head_snapc = NULL;
491 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
492 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
493 ci->i_nr_by_mode[i] = 0;
495 mutex_init(&ci->i_truncate_mutex);
496 ci->i_truncate_seq = 0;
497 ci->i_truncate_size = 0;
498 ci->i_truncate_pending = 0;
501 ci->i_reported_size = 0;
502 ci->i_wanted_max_size = 0;
503 ci->i_requested_max_size = 0;
507 ci->i_rdcache_ref = 0;
511 ci->i_wrbuffer_ref = 0;
512 ci->i_wrbuffer_ref_head = 0;
513 atomic_set(&ci->i_filelock_ref, 0);
514 atomic_set(&ci->i_shared_gen, 1);
515 ci->i_rdcache_gen = 0;
516 ci->i_rdcache_revoking = 0;
518 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
519 INIT_LIST_HEAD(&ci->i_unsafe_iops);
520 spin_lock_init(&ci->i_unsafe_lock);
522 ci->i_snap_realm = NULL;
523 INIT_LIST_HEAD(&ci->i_snap_realm_item);
524 INIT_LIST_HEAD(&ci->i_snap_flush_item);
526 INIT_WORK(&ci->i_work, ceph_inode_work);
528 memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
530 ceph_fscache_inode_init(ci);
532 return &ci->vfs_inode;
535 void ceph_free_inode(struct inode *inode)
537 struct ceph_inode_info *ci = ceph_inode(inode);
539 kfree(ci->i_symlink);
540 kmem_cache_free(ceph_inode_cachep, ci);
543 void ceph_evict_inode(struct inode *inode)
545 struct ceph_inode_info *ci = ceph_inode(inode);
546 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
547 struct ceph_inode_frag *frag;
550 dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
552 percpu_counter_dec(&mdsc->metric.total_inodes);
554 truncate_inode_pages_final(&inode->i_data);
557 ceph_fscache_unregister_inode_cookie(ci);
559 __ceph_remove_caps(ci);
561 if (__ceph_has_any_quota(ci))
562 ceph_adjust_quota_realms_count(inode, false);
565 * we may still have a snap_realm reference if there are stray
566 * caps in i_snap_caps.
568 if (ci->i_snap_realm) {
569 if (ceph_snap(inode) == CEPH_NOSNAP) {
570 struct ceph_snap_realm *realm = ci->i_snap_realm;
571 dout(" dropping residual ref to snap realm %p\n",
573 spin_lock(&realm->inodes_with_caps_lock);
574 list_del_init(&ci->i_snap_realm_item);
575 ci->i_snap_realm = NULL;
576 if (realm->ino == ci->i_vino.ino)
578 spin_unlock(&realm->inodes_with_caps_lock);
579 ceph_put_snap_realm(mdsc, realm);
581 ceph_put_snapid_map(mdsc, ci->i_snapid_map);
582 ci->i_snap_realm = NULL;
586 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
587 frag = rb_entry(n, struct ceph_inode_frag, node);
588 rb_erase(n, &ci->i_fragtree);
591 ci->i_fragtree_nsplits = 0;
593 __ceph_destroy_xattrs(ci);
594 if (ci->i_xattrs.blob)
595 ceph_buffer_put(ci->i_xattrs.blob);
596 if (ci->i_xattrs.prealloc_blob)
597 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
599 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
600 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
603 static inline blkcnt_t calc_inode_blocks(u64 size)
605 return (size + (1<<9) - 1) >> 9;
609 * Helpers to fill in size, ctime, mtime, and atime. We have to be
610 * careful because either the client or MDS may have more up to date
611 * info, depending on which capabilities are held, and whether
612 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
613 * and size are monotonically increasing, except when utimes() or
614 * truncate() increments the corresponding _seq values.)
616 int ceph_fill_file_size(struct inode *inode, int issued,
617 u32 truncate_seq, u64 truncate_size, u64 size)
619 struct ceph_inode_info *ci = ceph_inode(inode);
622 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
623 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
624 dout("size %lld -> %llu\n", inode->i_size, size);
625 if (size > 0 && S_ISDIR(inode->i_mode)) {
626 pr_err("fill_file_size non-zero size for directory\n");
629 i_size_write(inode, size);
630 inode->i_blocks = calc_inode_blocks(size);
631 ci->i_reported_size = size;
632 if (truncate_seq != ci->i_truncate_seq) {
633 dout("truncate_seq %u -> %u\n",
634 ci->i_truncate_seq, truncate_seq);
635 ci->i_truncate_seq = truncate_seq;
637 /* the MDS should have revoked these caps */
638 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
641 CEPH_CAP_FILE_LAZYIO));
643 * If we hold relevant caps, or in the case where we're
644 * not the only client referencing this file and we
645 * don't hold those caps, then we need to check whether
646 * the file is either opened or mmaped
648 if ((issued & (CEPH_CAP_FILE_CACHE|
649 CEPH_CAP_FILE_BUFFER)) ||
650 mapping_mapped(inode->i_mapping) ||
651 __ceph_is_file_opened(ci)) {
652 ci->i_truncate_pending++;
657 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
658 ci->i_truncate_size != truncate_size) {
659 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
661 ci->i_truncate_size = truncate_size;
665 ceph_fscache_invalidate(inode);
670 void ceph_fill_file_time(struct inode *inode, int issued,
671 u64 time_warp_seq, struct timespec64 *ctime,
672 struct timespec64 *mtime, struct timespec64 *atime)
674 struct ceph_inode_info *ci = ceph_inode(inode);
677 if (issued & (CEPH_CAP_FILE_EXCL|
679 CEPH_CAP_FILE_BUFFER|
681 CEPH_CAP_XATTR_EXCL)) {
682 if (ci->i_version == 0 ||
683 timespec64_compare(ctime, &inode->i_ctime) > 0) {
684 dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
685 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
686 ctime->tv_sec, ctime->tv_nsec);
687 inode->i_ctime = *ctime;
689 if (ci->i_version == 0 ||
690 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
691 /* the MDS did a utimes() */
692 dout("mtime %lld.%09ld -> %lld.%09ld "
694 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
695 mtime->tv_sec, mtime->tv_nsec,
696 ci->i_time_warp_seq, (int)time_warp_seq);
698 inode->i_mtime = *mtime;
699 inode->i_atime = *atime;
700 ci->i_time_warp_seq = time_warp_seq;
701 } else if (time_warp_seq == ci->i_time_warp_seq) {
702 /* nobody did utimes(); take the max */
703 if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
704 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
705 inode->i_mtime.tv_sec,
706 inode->i_mtime.tv_nsec,
707 mtime->tv_sec, mtime->tv_nsec);
708 inode->i_mtime = *mtime;
710 if (timespec64_compare(atime, &inode->i_atime) > 0) {
711 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
712 inode->i_atime.tv_sec,
713 inode->i_atime.tv_nsec,
714 atime->tv_sec, atime->tv_nsec);
715 inode->i_atime = *atime;
717 } else if (issued & CEPH_CAP_FILE_EXCL) {
718 /* we did a utimes(); ignore mds values */
723 /* we have no write|excl caps; whatever the MDS says is true */
724 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
725 inode->i_ctime = *ctime;
726 inode->i_mtime = *mtime;
727 inode->i_atime = *atime;
728 ci->i_time_warp_seq = time_warp_seq;
733 if (warn) /* time_warp_seq shouldn't go backwards */
734 dout("%p mds time_warp_seq %llu < %u\n",
735 inode, time_warp_seq, ci->i_time_warp_seq);
739 * Populate an inode based on info from mds. May be called on new or
742 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
743 struct ceph_mds_reply_info_in *iinfo,
744 struct ceph_mds_reply_dirfrag *dirinfo,
745 struct ceph_mds_session *session, int cap_fmode,
746 struct ceph_cap_reservation *caps_reservation)
748 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
749 struct ceph_mds_reply_inode *info = iinfo->in;
750 struct ceph_inode_info *ci = ceph_inode(inode);
751 int issued, new_issued, info_caps;
752 struct timespec64 mtime, atime, ctime;
753 struct ceph_buffer *xattr_blob = NULL;
754 struct ceph_buffer *old_blob = NULL;
755 struct ceph_string *pool_ns = NULL;
756 struct ceph_cap *new_cap = NULL;
759 bool queue_trunc = false;
760 bool new_version = false;
761 bool fill_inline = false;
763 lockdep_assert_held(&mdsc->snap_rwsem);
765 dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
766 inode, ceph_vinop(inode), le64_to_cpu(info->version),
769 info_caps = le32_to_cpu(info->cap.caps);
771 /* prealloc new cap struct */
772 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
773 new_cap = ceph_get_cap(mdsc, caps_reservation);
779 * prealloc xattr data, if it looks like we'll need it. only
780 * if len > 4 (meaning there are actually xattrs; the first 4
781 * bytes are the xattr count).
783 if (iinfo->xattr_len > 4) {
784 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
786 pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
790 if (iinfo->pool_ns_len > 0)
791 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
794 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
795 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
797 spin_lock(&ci->i_ceph_lock);
800 * provided version will be odd if inode value is projected,
801 * even if stable. skip the update if we have newer stable
802 * info (ours>=theirs, e.g. due to racing mds replies), unless
803 * we are getting projected (unstable) info (in which case the
804 * version is odd, and we want ours>theirs).
810 if (ci->i_version == 0 ||
811 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
812 le64_to_cpu(info->version) > (ci->i_version & ~1)))
815 /* Update change_attribute */
816 inode_set_max_iversion_raw(inode, iinfo->change_attr);
818 __ceph_caps_issued(ci, &issued);
819 issued |= __ceph_caps_dirty(ci);
820 new_issued = ~issued & info_caps;
823 inode->i_rdev = le32_to_cpu(info->rdev);
824 /* directories have fl_stripe_unit set to zero */
825 if (le32_to_cpu(info->layout.fl_stripe_unit))
827 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
829 inode->i_blkbits = CEPH_BLOCK_SHIFT;
831 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
833 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
834 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
835 inode->i_mode = le32_to_cpu(info->mode);
836 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
837 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
838 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
839 from_kuid(&init_user_ns, inode->i_uid),
840 from_kgid(&init_user_ns, inode->i_gid));
841 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
842 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
845 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
846 (issued & CEPH_CAP_LINK_EXCL) == 0)
847 set_nlink(inode, le32_to_cpu(info->nlink));
849 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
850 /* be careful with mtime, atime, size */
851 ceph_decode_timespec64(&atime, &info->atime);
852 ceph_decode_timespec64(&mtime, &info->mtime);
853 ceph_decode_timespec64(&ctime, &info->ctime);
854 ceph_fill_file_time(inode, issued,
855 le32_to_cpu(info->time_warp_seq),
856 &ctime, &mtime, &atime);
859 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
860 ci->i_files = le64_to_cpu(info->files);
861 ci->i_subdirs = le64_to_cpu(info->subdirs);
865 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
866 s64 old_pool = ci->i_layout.pool_id;
867 struct ceph_string *old_ns;
869 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
870 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
871 lockdep_is_held(&ci->i_ceph_lock));
872 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
874 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
875 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
879 queue_trunc = ceph_fill_file_size(inode, issued,
880 le32_to_cpu(info->truncate_seq),
881 le64_to_cpu(info->truncate_size),
882 le64_to_cpu(info->size));
883 /* only update max_size on auth cap */
884 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
885 ci->i_max_size != le64_to_cpu(info->max_size)) {
886 dout("max_size %lld -> %llu\n", ci->i_max_size,
887 le64_to_cpu(info->max_size));
888 ci->i_max_size = le64_to_cpu(info->max_size);
892 /* layout and rstat are not tracked by capability, update them if
893 * the inode info is from auth mds */
894 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
895 if (S_ISDIR(inode->i_mode)) {
896 ci->i_dir_layout = iinfo->dir_layout;
897 ci->i_rbytes = le64_to_cpu(info->rbytes);
898 ci->i_rfiles = le64_to_cpu(info->rfiles);
899 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
900 ci->i_dir_pin = iinfo->dir_pin;
901 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
906 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
907 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
908 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
909 if (ci->i_xattrs.blob)
910 old_blob = ci->i_xattrs.blob;
911 ci->i_xattrs.blob = xattr_blob;
913 memcpy(ci->i_xattrs.blob->vec.iov_base,
914 iinfo->xattr_data, iinfo->xattr_len);
915 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
916 ceph_forget_all_cached_acls(inode);
917 ceph_security_invalidate_secctx(inode);
921 /* finally update i_version */
922 if (le64_to_cpu(info->version) > ci->i_version)
923 ci->i_version = le64_to_cpu(info->version);
925 inode->i_mapping->a_ops = &ceph_aops;
927 switch (inode->i_mode & S_IFMT) {
932 inode->i_blkbits = PAGE_SHIFT;
933 init_special_inode(inode, inode->i_mode, inode->i_rdev);
934 inode->i_op = &ceph_file_iops;
937 inode->i_op = &ceph_file_iops;
938 inode->i_fop = &ceph_file_fops;
941 inode->i_op = &ceph_symlink_iops;
942 if (!ci->i_symlink) {
943 u32 symlen = iinfo->symlink_len;
946 spin_unlock(&ci->i_ceph_lock);
948 if (symlen != i_size_read(inode)) {
949 pr_err("%s %llx.%llx BAD symlink "
950 "size %lld\n", __func__,
953 i_size_write(inode, symlen);
954 inode->i_blocks = calc_inode_blocks(symlen);
958 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
962 spin_lock(&ci->i_ceph_lock);
966 kfree(sym); /* lost a race */
968 inode->i_link = ci->i_symlink;
971 inode->i_op = &ceph_dir_iops;
972 inode->i_fop = &ceph_dir_fops;
975 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
976 ceph_vinop(inode), inode->i_mode);
979 /* were we issued a capability? */
981 if (ceph_snap(inode) == CEPH_NOSNAP) {
982 ceph_add_cap(inode, session,
983 le64_to_cpu(info->cap.cap_id),
985 le32_to_cpu(info->cap.wanted),
986 le32_to_cpu(info->cap.seq),
987 le32_to_cpu(info->cap.mseq),
988 le64_to_cpu(info->cap.realm),
989 info->cap.flags, &new_cap);
991 /* set dir completion flag? */
992 if (S_ISDIR(inode->i_mode) &&
993 ci->i_files == 0 && ci->i_subdirs == 0 &&
994 (info_caps & CEPH_CAP_FILE_SHARED) &&
995 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
996 !__ceph_dir_is_complete(ci)) {
997 dout(" marking %p complete (empty)\n", inode);
998 i_size_write(inode, 0);
999 __ceph_dir_set_complete(ci,
1000 atomic64_read(&ci->i_release_count),
1001 atomic64_read(&ci->i_ordered_count));
1006 dout(" %p got snap_caps %s\n", inode,
1007 ceph_cap_string(info_caps));
1008 ci->i_snap_caps |= info_caps;
1012 if (iinfo->inline_version > 0 &&
1013 iinfo->inline_version >= ci->i_inline_version) {
1014 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1015 ci->i_inline_version = iinfo->inline_version;
1016 if (ci->i_inline_version != CEPH_INLINE_NONE &&
1017 (locked_page || (info_caps & cache_caps)))
1021 if (cap_fmode >= 0) {
1023 pr_warn("mds issued no caps on %llx.%llx\n",
1025 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1028 spin_unlock(&ci->i_ceph_lock);
1031 ceph_fill_inline_data(inode, locked_page,
1032 iinfo->inline_data, iinfo->inline_len);
1035 wake_up_all(&ci->i_cap_wq);
1037 /* queue truncate if we saw i_size decrease */
1039 ceph_queue_vmtruncate(inode);
1041 /* populate frag tree */
1042 if (S_ISDIR(inode->i_mode))
1043 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1045 /* update delegation info? */
1047 ceph_fill_dirfrag(inode, dirinfo);
1052 ceph_put_cap(mdsc, new_cap);
1053 ceph_buffer_put(old_blob);
1054 ceph_buffer_put(xattr_blob);
1055 ceph_put_string(pool_ns);
1060 * caller should hold session s_mutex and dentry->d_lock.
1062 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1063 struct ceph_mds_reply_lease *lease,
1064 struct ceph_mds_session *session,
1065 unsigned long from_time,
1066 struct ceph_mds_session **old_lease_session)
1068 struct ceph_dentry_info *di = ceph_dentry(dentry);
1069 unsigned mask = le16_to_cpu(lease->mask);
1070 long unsigned duration = le32_to_cpu(lease->duration_ms);
1071 long unsigned ttl = from_time + (duration * HZ) / 1000;
1072 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1074 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1075 dentry, duration, ttl);
1077 /* only track leases on regular dentries */
1078 if (ceph_snap(dir) != CEPH_NOSNAP)
1081 if (mask & CEPH_LEASE_PRIMARY_LINK)
1082 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1084 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1086 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1087 if (!(mask & CEPH_LEASE_VALID)) {
1088 __ceph_dentry_dir_lease_touch(di);
1092 if (di->lease_gen == session->s_cap_gen &&
1093 time_before(ttl, di->time))
1094 return; /* we already have a newer lease. */
1096 if (di->lease_session && di->lease_session != session) {
1097 *old_lease_session = di->lease_session;
1098 di->lease_session = NULL;
1101 if (!di->lease_session)
1102 di->lease_session = ceph_get_mds_session(session);
1103 di->lease_gen = session->s_cap_gen;
1104 di->lease_seq = le32_to_cpu(lease->seq);
1105 di->lease_renew_after = half_ttl;
1106 di->lease_renew_from = 0;
1109 __ceph_dentry_lease_touch(di);
1112 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1113 struct ceph_mds_reply_lease *lease,
1114 struct ceph_mds_session *session,
1115 unsigned long from_time)
1117 struct ceph_mds_session *old_lease_session = NULL;
1118 spin_lock(&dentry->d_lock);
1119 __update_dentry_lease(dir, dentry, lease, session, from_time,
1120 &old_lease_session);
1121 spin_unlock(&dentry->d_lock);
1122 ceph_put_mds_session(old_lease_session);
1126 * update dentry lease without having parent inode locked
1128 static void update_dentry_lease_careful(struct dentry *dentry,
1129 struct ceph_mds_reply_lease *lease,
1130 struct ceph_mds_session *session,
1131 unsigned long from_time,
1132 char *dname, u32 dname_len,
1133 struct ceph_vino *pdvino,
1134 struct ceph_vino *ptvino)
1138 struct ceph_mds_session *old_lease_session = NULL;
1140 spin_lock(&dentry->d_lock);
1141 /* make sure dentry's name matches target */
1142 if (dentry->d_name.len != dname_len ||
1143 memcmp(dentry->d_name.name, dname, dname_len))
1146 dir = d_inode(dentry->d_parent);
1147 /* make sure parent matches dvino */
1148 if (!ceph_ino_compare(dir, pdvino))
1151 /* make sure dentry's inode matches target. NULL ptvino means that
1152 * we expect a negative dentry */
1154 if (d_really_is_negative(dentry))
1156 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1159 if (d_really_is_positive(dentry))
1163 __update_dentry_lease(dir, dentry, lease, session,
1164 from_time, &old_lease_session);
1166 spin_unlock(&dentry->d_lock);
1167 ceph_put_mds_session(old_lease_session);
1171 * splice a dentry to an inode.
1172 * caller must hold directory i_mutex for this to be safe.
1174 static int splice_dentry(struct dentry **pdn, struct inode *in)
1176 struct dentry *dn = *pdn;
1177 struct dentry *realdn;
1179 BUG_ON(d_inode(dn));
1181 if (S_ISDIR(in->i_mode)) {
1182 /* If inode is directory, d_splice_alias() below will remove
1183 * 'realdn' from its origin parent. We need to ensure that
1184 * origin parent's readdir cache will not reference 'realdn'
1186 realdn = d_find_any_alias(in);
1188 struct ceph_dentry_info *di = ceph_dentry(realdn);
1189 spin_lock(&realdn->d_lock);
1191 realdn->d_op->d_prune(realdn);
1194 di->lease_shared_gen = 0;
1197 spin_unlock(&realdn->d_lock);
1202 /* dn must be unhashed */
1203 if (!d_unhashed(dn))
1205 realdn = d_splice_alias(in, dn);
1206 if (IS_ERR(realdn)) {
1207 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1208 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1209 return PTR_ERR(realdn);
1213 dout("dn %p (%d) spliced with %p (%d) "
1214 "inode %p ino %llx.%llx\n",
1216 realdn, d_count(realdn),
1217 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1221 BUG_ON(!ceph_dentry(dn));
1222 dout("dn %p attached to %p ino %llx.%llx\n",
1223 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1229 * Incorporate results into the local cache. This is either just
1230 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1233 * A reply may contain
1234 * a directory inode along with a dentry.
1235 * and/or a target inode
1237 * Called with snap_rwsem (read).
1239 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1241 struct ceph_mds_session *session = req->r_session;
1242 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1243 struct inode *in = NULL;
1244 struct ceph_vino tvino, dvino;
1245 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1248 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1249 rinfo->head->is_dentry, rinfo->head->is_target);
1251 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1252 dout("fill_trace reply is empty!\n");
1253 if (rinfo->head->result == 0 && req->r_parent)
1254 ceph_invalidate_dir_request(req);
1258 if (rinfo->head->is_dentry) {
1259 struct inode *dir = req->r_parent;
1262 err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1263 rinfo->dirfrag, session, -1,
1264 &req->r_caps_reservation);
1271 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1272 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1273 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1275 struct dentry *dn, *parent;
1277 BUG_ON(!rinfo->head->is_target);
1278 BUG_ON(req->r_dentry);
1280 parent = d_find_any_alias(dir);
1283 dname.name = rinfo->dname;
1284 dname.len = rinfo->dname_len;
1285 dname.hash = full_name_hash(parent, dname.name, dname.len);
1286 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1287 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1289 dn = d_lookup(parent, &dname);
1290 dout("d_lookup on parent=%p name=%.*s got %p\n",
1291 parent, dname.len, dname.name, dn);
1294 dn = d_alloc(parent, &dname);
1295 dout("d_alloc %p '%.*s' = %p\n", parent,
1296 dname.len, dname.name, dn);
1303 } else if (d_really_is_positive(dn) &&
1304 (ceph_ino(d_inode(dn)) != tvino.ino ||
1305 ceph_snap(d_inode(dn)) != tvino.snap)) {
1306 dout(" dn %p points to wrong inode %p\n",
1308 ceph_dir_clear_ordered(dir);
1319 if (rinfo->head->is_target) {
1320 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1321 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1323 in = ceph_get_inode(sb, tvino);
1329 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1331 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1332 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1333 rinfo->head->result == 0) ? req->r_fmode : -1,
1334 &req->r_caps_reservation);
1336 pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1337 in, ceph_vinop(in));
1338 if (in->i_state & I_NEW)
1339 discard_new_inode(in);
1344 req->r_target_inode = in;
1345 if (in->i_state & I_NEW)
1346 unlock_new_inode(in);
1350 * ignore null lease/binding on snapdir ENOENT, or else we
1351 * will have trouble splicing in the virtual snapdir later
1353 if (rinfo->head->is_dentry &&
1354 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1355 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1356 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1357 fsc->mount_options->snapdir_name,
1358 req->r_dentry->d_name.len))) {
1360 * lookup link rename : null -> possibly existing inode
1361 * mknod symlink mkdir : null -> new inode
1362 * unlink : linked -> null
1364 struct inode *dir = req->r_parent;
1365 struct dentry *dn = req->r_dentry;
1366 bool have_dir_cap, have_lease;
1370 BUG_ON(d_inode(dn->d_parent) != dir);
1372 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1373 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1375 BUG_ON(ceph_ino(dir) != dvino.ino);
1376 BUG_ON(ceph_snap(dir) != dvino.snap);
1378 /* do we have a lease on the whole dir? */
1380 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1381 CEPH_CAP_FILE_SHARED);
1383 /* do we have a dn lease? */
1384 have_lease = have_dir_cap ||
1385 le32_to_cpu(rinfo->dlease->duration_ms);
1387 dout("fill_trace no dentry lease or dir cap\n");
1390 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1391 struct inode *olddir = req->r_old_dentry_dir;
1394 dout(" src %p '%pd' dst %p '%pd'\n",
1398 dout("fill_trace doing d_move %p -> %p\n",
1399 req->r_old_dentry, dn);
1401 /* d_move screws up sibling dentries' offsets */
1402 ceph_dir_clear_ordered(dir);
1403 ceph_dir_clear_ordered(olddir);
1405 d_move(req->r_old_dentry, dn);
1406 dout(" src %p '%pd' dst %p '%pd'\n",
1411 /* ensure target dentry is invalidated, despite
1412 rehashing bug in vfs_rename_dir */
1413 ceph_invalidate_dentry_lease(dn);
1415 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1416 ceph_dentry(req->r_old_dentry)->offset);
1418 /* swap r_dentry and r_old_dentry in case that
1419 * splice_dentry() gets called later. This is safe
1420 * because no other place will use them */
1421 req->r_dentry = req->r_old_dentry;
1422 req->r_old_dentry = dn;
1427 if (!rinfo->head->is_target) {
1428 dout("fill_trace null dentry\n");
1429 if (d_really_is_positive(dn)) {
1430 dout("d_delete %p\n", dn);
1431 ceph_dir_clear_ordered(dir);
1433 } else if (have_lease) {
1436 update_dentry_lease(dir, dn,
1437 rinfo->dlease, session,
1438 req->r_request_started);
1443 /* attach proper inode */
1444 if (d_really_is_negative(dn)) {
1445 ceph_dir_clear_ordered(dir);
1447 err = splice_dentry(&req->r_dentry, in);
1450 dn = req->r_dentry; /* may have spliced */
1451 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1452 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1453 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1460 update_dentry_lease(dir, dn,
1461 rinfo->dlease, session,
1462 req->r_request_started);
1464 dout(" final dn %p\n", dn);
1465 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1466 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1467 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1468 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1469 struct inode *dir = req->r_parent;
1471 /* fill out a snapdir LOOKUPSNAP dentry */
1473 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1474 BUG_ON(!req->r_dentry);
1475 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1476 ceph_dir_clear_ordered(dir);
1478 err = splice_dentry(&req->r_dentry, in);
1481 } else if (rinfo->head->is_dentry && req->r_dentry) {
1482 /* parent inode is not locked, be carefull */
1483 struct ceph_vino *ptvino = NULL;
1484 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1485 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1486 if (rinfo->head->is_target) {
1487 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1488 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1491 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1492 session, req->r_request_started,
1493 rinfo->dname, rinfo->dname_len,
1497 dout("fill_trace done err=%d\n", err);
1502 * Prepopulate our cache with readdir results, leases, etc.
1504 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1505 struct ceph_mds_session *session)
1507 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1510 for (i = 0; i < rinfo->dir_nr; i++) {
1511 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1512 struct ceph_vino vino;
1516 vino.ino = le64_to_cpu(rde->inode.in->ino);
1517 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1519 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1522 dout("new_inode badness got %d\n", err);
1525 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1526 -1, &req->r_caps_reservation);
1528 pr_err("ceph_fill_inode badness on %p got %d\n",
1531 if (in->i_state & I_NEW) {
1533 discard_new_inode(in);
1535 } else if (in->i_state & I_NEW) {
1536 unlock_new_inode(in);
1539 /* avoid calling iput_final() in mds dispatch threads */
1540 ceph_async_iput(in);
1546 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1550 put_page(ctl->page);
1555 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1556 struct ceph_readdir_cache_control *ctl,
1557 struct ceph_mds_request *req)
1559 struct ceph_inode_info *ci = ceph_inode(dir);
1560 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1561 unsigned idx = ctl->index % nsize;
1562 pgoff_t pgoff = ctl->index / nsize;
1564 if (!ctl->page || pgoff != page_index(ctl->page)) {
1565 ceph_readdir_cache_release(ctl);
1567 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1569 ctl->page = find_lock_page(&dir->i_data, pgoff);
1572 return idx == 0 ? -ENOMEM : 0;
1574 /* reading/filling the cache are serialized by
1575 * i_mutex, no need to use page lock */
1576 unlock_page(ctl->page);
1577 ctl->dentries = kmap(ctl->page);
1579 memset(ctl->dentries, 0, PAGE_SIZE);
1582 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1583 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1584 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1585 ctl->dentries[idx] = dn;
1588 dout("disable readdir cache\n");
1594 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1595 struct ceph_mds_session *session)
1597 struct dentry *parent = req->r_dentry;
1598 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1599 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1603 int err = 0, skipped = 0, ret, i;
1604 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1605 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1608 struct ceph_readdir_cache_control cache_ctl = {};
1610 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1611 return readdir_prepopulate_inodes_only(req, session);
1613 if (rinfo->hash_order) {
1615 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1617 strlen(req->r_path2));
1618 last_hash = ceph_frag_value(last_hash);
1619 } else if (rinfo->offset_hash) {
1620 /* mds understands offset_hash */
1621 WARN_ON_ONCE(req->r_readdir_offset != 2);
1622 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1626 if (rinfo->dir_dir &&
1627 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1628 dout("readdir_prepopulate got new frag %x -> %x\n",
1629 frag, le32_to_cpu(rinfo->dir_dir->frag));
1630 frag = le32_to_cpu(rinfo->dir_dir->frag);
1631 if (!rinfo->hash_order)
1632 req->r_readdir_offset = 2;
1635 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1636 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1637 rinfo->dir_nr, parent);
1639 dout("readdir_prepopulate %d items under dn %p\n",
1640 rinfo->dir_nr, parent);
1642 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1644 if (ceph_frag_is_leftmost(frag) &&
1645 req->r_readdir_offset == 2 &&
1646 !(rinfo->hash_order && last_hash)) {
1647 /* note dir version at start of readdir so we can
1648 * tell if any dentries get dropped */
1649 req->r_dir_release_cnt =
1650 atomic64_read(&ci->i_release_count);
1651 req->r_dir_ordered_cnt =
1652 atomic64_read(&ci->i_ordered_count);
1653 req->r_readdir_cache_idx = 0;
1657 cache_ctl.index = req->r_readdir_cache_idx;
1658 fpos_offset = req->r_readdir_offset;
1660 /* FIXME: release caps/leases if error occurs */
1661 for (i = 0; i < rinfo->dir_nr; i++) {
1662 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1663 struct ceph_vino tvino;
1665 dname.name = rde->name;
1666 dname.len = rde->name_len;
1667 dname.hash = full_name_hash(parent, dname.name, dname.len);
1669 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1670 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1672 if (rinfo->hash_order) {
1673 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1674 rde->name, rde->name_len);
1675 hash = ceph_frag_value(hash);
1676 if (hash != last_hash)
1679 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1681 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1685 dn = d_lookup(parent, &dname);
1686 dout("d_lookup on parent=%p name=%.*s got %p\n",
1687 parent, dname.len, dname.name, dn);
1690 dn = d_alloc(parent, &dname);
1691 dout("d_alloc %p '%.*s' = %p\n", parent,
1692 dname.len, dname.name, dn);
1694 dout("d_alloc badness\n");
1698 } else if (d_really_is_positive(dn) &&
1699 (ceph_ino(d_inode(dn)) != tvino.ino ||
1700 ceph_snap(d_inode(dn)) != tvino.snap)) {
1701 struct ceph_dentry_info *di = ceph_dentry(dn);
1702 dout(" dn %p points to wrong inode %p\n",
1705 spin_lock(&dn->d_lock);
1706 if (di->offset > 0 &&
1707 di->lease_shared_gen ==
1708 atomic_read(&ci->i_shared_gen)) {
1709 __ceph_dir_clear_ordered(ci);
1712 spin_unlock(&dn->d_lock);
1720 if (d_really_is_positive(dn)) {
1723 in = ceph_get_inode(parent->d_sb, tvino);
1725 dout("new_inode badness\n");
1733 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1734 -1, &req->r_caps_reservation);
1736 pr_err("ceph_fill_inode badness on %p\n", in);
1737 if (d_really_is_negative(dn)) {
1738 /* avoid calling iput_final() in mds
1739 * dispatch threads */
1740 if (in->i_state & I_NEW) {
1742 discard_new_inode(in);
1744 ceph_async_iput(in);
1750 if (in->i_state & I_NEW)
1751 unlock_new_inode(in);
1753 if (d_really_is_negative(dn)) {
1754 if (ceph_security_xattr_deadlock(in)) {
1755 dout(" skip splicing dn %p to inode %p"
1756 " (security xattr deadlock)\n", dn, in);
1757 ceph_async_iput(in);
1762 err = splice_dentry(&dn, in);
1767 ceph_dentry(dn)->offset = rde->offset;
1769 update_dentry_lease(d_inode(parent), dn,
1770 rde->lease, req->r_session,
1771 req->r_request_started);
1773 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1774 ret = fill_readdir_cache(d_inode(parent), dn,
1783 if (err == 0 && skipped == 0) {
1784 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1785 req->r_readdir_cache_idx = cache_ctl.index;
1787 ceph_readdir_cache_release(&cache_ctl);
1788 dout("readdir_prepopulate done\n");
1792 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1794 struct ceph_inode_info *ci = ceph_inode(inode);
1797 spin_lock(&ci->i_ceph_lock);
1798 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1799 i_size_write(inode, size);
1800 inode->i_blocks = calc_inode_blocks(size);
1802 ret = __ceph_should_report_size(ci);
1804 spin_unlock(&ci->i_ceph_lock);
1809 * Put reference to inode, but avoid calling iput_final() in current thread.
1810 * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1813 void ceph_async_iput(struct inode *inode)
1818 if (atomic_add_unless(&inode->i_count, -1, 1))
1820 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1821 &ceph_inode(inode)->i_work))
1823 /* queue work failed, i_count must be at least 2 */
1828 * Write back inode data in a worker thread. (This can't be done
1829 * in the message handler context.)
1831 void ceph_queue_writeback(struct inode *inode)
1833 struct ceph_inode_info *ci = ceph_inode(inode);
1834 set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1837 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1839 dout("ceph_queue_writeback %p\n", inode);
1841 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1842 inode, ci->i_work_mask);
1848 * queue an async invalidation
1850 void ceph_queue_invalidate(struct inode *inode)
1852 struct ceph_inode_info *ci = ceph_inode(inode);
1853 set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1856 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1857 &ceph_inode(inode)->i_work)) {
1858 dout("ceph_queue_invalidate %p\n", inode);
1860 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1861 inode, ci->i_work_mask);
1867 * Queue an async vmtruncate. If we fail to queue work, we will handle
1868 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1870 void ceph_queue_vmtruncate(struct inode *inode)
1872 struct ceph_inode_info *ci = ceph_inode(inode);
1873 set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1876 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1878 dout("ceph_queue_vmtruncate %p\n", inode);
1880 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1881 inode, ci->i_work_mask);
1886 static void ceph_do_invalidate_pages(struct inode *inode)
1888 struct ceph_inode_info *ci = ceph_inode(inode);
1889 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1893 mutex_lock(&ci->i_truncate_mutex);
1895 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1896 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1897 inode, ceph_ino(inode));
1898 mapping_set_error(inode->i_mapping, -EIO);
1899 truncate_pagecache(inode, 0);
1900 mutex_unlock(&ci->i_truncate_mutex);
1904 spin_lock(&ci->i_ceph_lock);
1905 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1906 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1907 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1908 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1910 spin_unlock(&ci->i_ceph_lock);
1911 mutex_unlock(&ci->i_truncate_mutex);
1914 orig_gen = ci->i_rdcache_gen;
1915 spin_unlock(&ci->i_ceph_lock);
1917 ceph_fscache_invalidate(inode);
1918 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1919 pr_err("invalidate_pages %p fails\n", inode);
1922 spin_lock(&ci->i_ceph_lock);
1923 if (orig_gen == ci->i_rdcache_gen &&
1924 orig_gen == ci->i_rdcache_revoking) {
1925 dout("invalidate_pages %p gen %d successful\n", inode,
1927 ci->i_rdcache_revoking--;
1930 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1931 inode, orig_gen, ci->i_rdcache_gen,
1932 ci->i_rdcache_revoking);
1933 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1936 spin_unlock(&ci->i_ceph_lock);
1937 mutex_unlock(&ci->i_truncate_mutex);
1940 ceph_check_caps(ci, 0, NULL);
1944 * Make sure any pending truncation is applied before doing anything
1945 * that may depend on it.
1947 void __ceph_do_pending_vmtruncate(struct inode *inode)
1949 struct ceph_inode_info *ci = ceph_inode(inode);
1951 int wrbuffer_refs, finish = 0;
1953 mutex_lock(&ci->i_truncate_mutex);
1955 spin_lock(&ci->i_ceph_lock);
1956 if (ci->i_truncate_pending == 0) {
1957 dout("__do_pending_vmtruncate %p none pending\n", inode);
1958 spin_unlock(&ci->i_ceph_lock);
1959 mutex_unlock(&ci->i_truncate_mutex);
1964 * make sure any dirty snapped pages are flushed before we
1965 * possibly truncate them.. so write AND block!
1967 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1968 spin_unlock(&ci->i_ceph_lock);
1969 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1971 filemap_write_and_wait_range(&inode->i_data, 0,
1972 inode->i_sb->s_maxbytes);
1976 /* there should be no reader or writer */
1977 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1979 to = ci->i_truncate_size;
1980 wrbuffer_refs = ci->i_wrbuffer_ref;
1981 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1982 ci->i_truncate_pending, to);
1983 spin_unlock(&ci->i_ceph_lock);
1985 truncate_pagecache(inode, to);
1987 spin_lock(&ci->i_ceph_lock);
1988 if (to == ci->i_truncate_size) {
1989 ci->i_truncate_pending = 0;
1992 spin_unlock(&ci->i_ceph_lock);
1996 mutex_unlock(&ci->i_truncate_mutex);
1998 if (wrbuffer_refs == 0)
1999 ceph_check_caps(ci, 0, NULL);
2001 wake_up_all(&ci->i_cap_wq);
2004 static void ceph_inode_work(struct work_struct *work)
2006 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
2008 struct inode *inode = &ci->vfs_inode;
2010 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2011 dout("writeback %p\n", inode);
2012 filemap_fdatawrite(&inode->i_data);
2014 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2015 ceph_do_invalidate_pages(inode);
2017 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2018 __ceph_do_pending_vmtruncate(inode);
2026 static const struct inode_operations ceph_symlink_iops = {
2027 .get_link = simple_get_link,
2028 .setattr = ceph_setattr,
2029 .getattr = ceph_getattr,
2030 .listxattr = ceph_listxattr,
2033 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2035 struct ceph_inode_info *ci = ceph_inode(inode);
2036 unsigned int ia_valid = attr->ia_valid;
2037 struct ceph_mds_request *req;
2038 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2039 struct ceph_cap_flush *prealloc_cf;
2041 int release = 0, dirtied = 0;
2044 int inode_dirty_flags = 0;
2045 bool lock_snap_rwsem = false;
2047 prealloc_cf = ceph_alloc_cap_flush();
2051 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2054 ceph_free_cap_flush(prealloc_cf);
2055 return PTR_ERR(req);
2058 spin_lock(&ci->i_ceph_lock);
2059 issued = __ceph_caps_issued(ci, NULL);
2061 if (!ci->i_head_snapc &&
2062 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2063 lock_snap_rwsem = true;
2064 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2065 spin_unlock(&ci->i_ceph_lock);
2066 down_read(&mdsc->snap_rwsem);
2067 spin_lock(&ci->i_ceph_lock);
2068 issued = __ceph_caps_issued(ci, NULL);
2072 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2074 if (ia_valid & ATTR_UID) {
2075 dout("setattr %p uid %d -> %d\n", inode,
2076 from_kuid(&init_user_ns, inode->i_uid),
2077 from_kuid(&init_user_ns, attr->ia_uid));
2078 if (issued & CEPH_CAP_AUTH_EXCL) {
2079 inode->i_uid = attr->ia_uid;
2080 dirtied |= CEPH_CAP_AUTH_EXCL;
2081 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2082 !uid_eq(attr->ia_uid, inode->i_uid)) {
2083 req->r_args.setattr.uid = cpu_to_le32(
2084 from_kuid(&init_user_ns, attr->ia_uid));
2085 mask |= CEPH_SETATTR_UID;
2086 release |= CEPH_CAP_AUTH_SHARED;
2089 if (ia_valid & ATTR_GID) {
2090 dout("setattr %p gid %d -> %d\n", inode,
2091 from_kgid(&init_user_ns, inode->i_gid),
2092 from_kgid(&init_user_ns, attr->ia_gid));
2093 if (issued & CEPH_CAP_AUTH_EXCL) {
2094 inode->i_gid = attr->ia_gid;
2095 dirtied |= CEPH_CAP_AUTH_EXCL;
2096 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2097 !gid_eq(attr->ia_gid, inode->i_gid)) {
2098 req->r_args.setattr.gid = cpu_to_le32(
2099 from_kgid(&init_user_ns, attr->ia_gid));
2100 mask |= CEPH_SETATTR_GID;
2101 release |= CEPH_CAP_AUTH_SHARED;
2104 if (ia_valid & ATTR_MODE) {
2105 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2107 if (issued & CEPH_CAP_AUTH_EXCL) {
2108 inode->i_mode = attr->ia_mode;
2109 dirtied |= CEPH_CAP_AUTH_EXCL;
2110 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2111 attr->ia_mode != inode->i_mode) {
2112 inode->i_mode = attr->ia_mode;
2113 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2114 mask |= CEPH_SETATTR_MODE;
2115 release |= CEPH_CAP_AUTH_SHARED;
2119 if (ia_valid & ATTR_ATIME) {
2120 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2121 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2122 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2123 if (issued & CEPH_CAP_FILE_EXCL) {
2124 ci->i_time_warp_seq++;
2125 inode->i_atime = attr->ia_atime;
2126 dirtied |= CEPH_CAP_FILE_EXCL;
2127 } else if ((issued & CEPH_CAP_FILE_WR) &&
2128 timespec64_compare(&inode->i_atime,
2129 &attr->ia_atime) < 0) {
2130 inode->i_atime = attr->ia_atime;
2131 dirtied |= CEPH_CAP_FILE_WR;
2132 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2133 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2134 ceph_encode_timespec64(&req->r_args.setattr.atime,
2136 mask |= CEPH_SETATTR_ATIME;
2137 release |= CEPH_CAP_FILE_SHARED |
2138 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2141 if (ia_valid & ATTR_SIZE) {
2142 dout("setattr %p size %lld -> %lld\n", inode,
2143 inode->i_size, attr->ia_size);
2144 if ((issued & CEPH_CAP_FILE_EXCL) &&
2145 attr->ia_size > inode->i_size) {
2146 i_size_write(inode, attr->ia_size);
2147 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2148 ci->i_reported_size = attr->ia_size;
2149 dirtied |= CEPH_CAP_FILE_EXCL;
2150 ia_valid |= ATTR_MTIME;
2151 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2152 attr->ia_size != inode->i_size) {
2153 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2154 req->r_args.setattr.old_size =
2155 cpu_to_le64(inode->i_size);
2156 mask |= CEPH_SETATTR_SIZE;
2157 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2158 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2161 if (ia_valid & ATTR_MTIME) {
2162 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2163 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2164 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2165 if (issued & CEPH_CAP_FILE_EXCL) {
2166 ci->i_time_warp_seq++;
2167 inode->i_mtime = attr->ia_mtime;
2168 dirtied |= CEPH_CAP_FILE_EXCL;
2169 } else if ((issued & CEPH_CAP_FILE_WR) &&
2170 timespec64_compare(&inode->i_mtime,
2171 &attr->ia_mtime) < 0) {
2172 inode->i_mtime = attr->ia_mtime;
2173 dirtied |= CEPH_CAP_FILE_WR;
2174 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2175 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2176 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2178 mask |= CEPH_SETATTR_MTIME;
2179 release |= CEPH_CAP_FILE_SHARED |
2180 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2184 /* these do nothing */
2185 if (ia_valid & ATTR_CTIME) {
2186 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2187 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2188 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2189 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2190 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2191 only ? "ctime only" : "ignored");
2194 * if kernel wants to dirty ctime but nothing else,
2195 * we need to choose a cap to dirty under, or do
2196 * a almost-no-op setattr
2198 if (issued & CEPH_CAP_AUTH_EXCL)
2199 dirtied |= CEPH_CAP_AUTH_EXCL;
2200 else if (issued & CEPH_CAP_FILE_EXCL)
2201 dirtied |= CEPH_CAP_FILE_EXCL;
2202 else if (issued & CEPH_CAP_XATTR_EXCL)
2203 dirtied |= CEPH_CAP_XATTR_EXCL;
2205 mask |= CEPH_SETATTR_CTIME;
2208 if (ia_valid & ATTR_FILE)
2209 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2212 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2214 inode->i_ctime = attr->ia_ctime;
2218 spin_unlock(&ci->i_ceph_lock);
2219 if (lock_snap_rwsem)
2220 up_read(&mdsc->snap_rwsem);
2222 if (inode_dirty_flags)
2223 __mark_inode_dirty(inode, inode_dirty_flags);
2227 req->r_inode = inode;
2229 req->r_inode_drop = release;
2230 req->r_args.setattr.mask = cpu_to_le32(mask);
2231 req->r_num_caps = 1;
2232 req->r_stamp = attr->ia_ctime;
2233 err = ceph_mdsc_do_request(mdsc, NULL, req);
2235 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2236 ceph_cap_string(dirtied), mask);
2238 ceph_mdsc_put_request(req);
2239 ceph_free_cap_flush(prealloc_cf);
2241 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2242 __ceph_do_pending_vmtruncate(inode);
2250 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2252 struct inode *inode = d_inode(dentry);
2253 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2256 if (ceph_snap(inode) != CEPH_NOSNAP)
2259 err = setattr_prepare(dentry, attr);
2263 if ((attr->ia_valid & ATTR_SIZE) &&
2264 attr->ia_size > max(inode->i_size, fsc->max_file_size))
2267 if ((attr->ia_valid & ATTR_SIZE) &&
2268 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2271 err = __ceph_setattr(inode, attr);
2273 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2274 err = posix_acl_chmod(inode, attr->ia_mode);
2280 * Verify that we have a lease on the given mask. If not,
2281 * do a getattr against an mds.
2283 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2284 int mask, bool force)
2286 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2287 struct ceph_mds_client *mdsc = fsc->mdsc;
2288 struct ceph_mds_request *req;
2292 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2293 dout("do_getattr inode %p SNAPDIR\n", inode);
2297 dout("do_getattr inode %p mask %s mode 0%o\n",
2298 inode, ceph_cap_string(mask), inode->i_mode);
2299 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2302 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2303 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2305 return PTR_ERR(req);
2306 req->r_inode = inode;
2308 req->r_num_caps = 1;
2309 req->r_args.getattr.mask = cpu_to_le32(mask);
2310 req->r_locked_page = locked_page;
2311 err = ceph_mdsc_do_request(mdsc, NULL, req);
2312 if (locked_page && err == 0) {
2313 u64 inline_version = req->r_reply_info.targeti.inline_version;
2314 if (inline_version == 0) {
2315 /* the reply is supposed to contain inline data */
2317 } else if (inline_version == CEPH_INLINE_NONE) {
2320 err = req->r_reply_info.targeti.inline_len;
2323 ceph_mdsc_put_request(req);
2324 dout("do_getattr result=%d\n", err);
2330 * Check inode permissions. We verify we have a valid value for
2331 * the AUTH cap, then call the generic handler.
2333 int ceph_permission(struct inode *inode, int mask)
2337 if (mask & MAY_NOT_BLOCK)
2340 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2343 err = generic_permission(inode, mask);
2347 /* Craft a mask of needed caps given a set of requested statx attrs. */
2348 static int statx_to_caps(u32 want)
2352 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2353 mask |= CEPH_CAP_AUTH_SHARED;
2355 if (want & (STATX_NLINK|STATX_CTIME))
2356 mask |= CEPH_CAP_LINK_SHARED;
2358 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2360 mask |= CEPH_CAP_FILE_SHARED;
2362 if (want & (STATX_CTIME))
2363 mask |= CEPH_CAP_XATTR_SHARED;
2369 * Get all the attributes. If we have sufficient caps for the requested attrs,
2370 * then we can avoid talking to the MDS at all.
2372 int ceph_getattr(const struct path *path, struct kstat *stat,
2373 u32 request_mask, unsigned int flags)
2375 struct inode *inode = d_inode(path->dentry);
2376 struct ceph_inode_info *ci = ceph_inode(inode);
2377 u32 valid_mask = STATX_BASIC_STATS;
2380 /* Skip the getattr altogether if we're asked not to sync */
2381 if (!(flags & AT_STATX_DONT_SYNC)) {
2382 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2383 flags & AT_STATX_FORCE_SYNC);
2388 generic_fillattr(inode, stat);
2389 stat->ino = ceph_present_inode(inode);
2392 * btime on newly-allocated inodes is 0, so if this is still set to
2393 * that, then assume that it's not valid.
2395 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2396 stat->btime = ci->i_btime;
2397 valid_mask |= STATX_BTIME;
2400 if (ceph_snap(inode) == CEPH_NOSNAP)
2401 stat->dev = inode->i_sb->s_dev;
2403 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2405 if (S_ISDIR(inode->i_mode)) {
2406 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2408 stat->size = ci->i_rbytes;
2410 stat->size = ci->i_files + ci->i_subdirs;
2412 stat->blksize = 65536;
2414 * Some applications rely on the number of st_nlink
2415 * value on directories to be either 0 (if unlinked)
2416 * or 2 + number of subdirectories.
2418 if (stat->nlink == 1)
2419 /* '.' + '..' + subdirs */
2420 stat->nlink = 1 + 1 + ci->i_subdirs;
2423 stat->result_mask = request_mask & valid_mask;