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(inode);
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));
818 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
819 (issued & CEPH_CAP_LINK_EXCL) == 0)
820 set_nlink(inode, le32_to_cpu(info->nlink));
822 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
823 /* be careful with mtime, atime, size */
824 ceph_decode_timespec64(&atime, &info->atime);
825 ceph_decode_timespec64(&mtime, &info->mtime);
826 ceph_decode_timespec64(&ctime, &info->ctime);
827 ceph_fill_file_time(inode, issued,
828 le32_to_cpu(info->time_warp_seq),
829 &ctime, &mtime, &atime);
832 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
833 ci->i_files = le64_to_cpu(info->files);
834 ci->i_subdirs = le64_to_cpu(info->subdirs);
838 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
839 s64 old_pool = ci->i_layout.pool_id;
840 struct ceph_string *old_ns;
842 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
843 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
844 lockdep_is_held(&ci->i_ceph_lock));
845 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
847 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
848 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
852 queue_trunc = ceph_fill_file_size(inode, issued,
853 le32_to_cpu(info->truncate_seq),
854 le64_to_cpu(info->truncate_size),
855 le64_to_cpu(info->size));
856 /* only update max_size on auth cap */
857 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
858 ci->i_max_size != le64_to_cpu(info->max_size)) {
859 dout("max_size %lld -> %llu\n", ci->i_max_size,
860 le64_to_cpu(info->max_size));
861 ci->i_max_size = le64_to_cpu(info->max_size);
865 /* layout and rstat are not tracked by capability, update them if
866 * the inode info is from auth mds */
867 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
868 if (S_ISDIR(inode->i_mode)) {
869 ci->i_dir_layout = iinfo->dir_layout;
870 ci->i_rbytes = le64_to_cpu(info->rbytes);
871 ci->i_rfiles = le64_to_cpu(info->rfiles);
872 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
873 ci->i_dir_pin = iinfo->dir_pin;
874 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
879 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
880 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
881 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
882 if (ci->i_xattrs.blob)
883 ceph_buffer_put(ci->i_xattrs.blob);
884 ci->i_xattrs.blob = xattr_blob;
886 memcpy(ci->i_xattrs.blob->vec.iov_base,
887 iinfo->xattr_data, iinfo->xattr_len);
888 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
889 ceph_forget_all_cached_acls(inode);
893 /* finally update i_version */
894 if (le64_to_cpu(info->version) > ci->i_version)
895 ci->i_version = le64_to_cpu(info->version);
897 inode->i_mapping->a_ops = &ceph_aops;
899 switch (inode->i_mode & S_IFMT) {
904 inode->i_blkbits = PAGE_SHIFT;
905 init_special_inode(inode, inode->i_mode, inode->i_rdev);
906 inode->i_op = &ceph_file_iops;
909 inode->i_op = &ceph_file_iops;
910 inode->i_fop = &ceph_file_fops;
913 inode->i_op = &ceph_symlink_iops;
914 if (!ci->i_symlink) {
915 u32 symlen = iinfo->symlink_len;
918 spin_unlock(&ci->i_ceph_lock);
920 if (symlen != i_size_read(inode)) {
921 pr_err("fill_inode %llx.%llx BAD symlink "
922 "size %lld\n", ceph_vinop(inode),
924 i_size_write(inode, symlen);
925 inode->i_blocks = calc_inode_blocks(symlen);
929 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
933 spin_lock(&ci->i_ceph_lock);
937 kfree(sym); /* lost a race */
939 inode->i_link = ci->i_symlink;
942 inode->i_op = &ceph_dir_iops;
943 inode->i_fop = &ceph_dir_fops;
946 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
947 ceph_vinop(inode), inode->i_mode);
950 /* were we issued a capability? */
952 if (ceph_snap(inode) == CEPH_NOSNAP) {
953 ceph_add_cap(inode, session,
954 le64_to_cpu(info->cap.cap_id),
955 cap_fmode, info_caps,
956 le32_to_cpu(info->cap.wanted),
957 le32_to_cpu(info->cap.seq),
958 le32_to_cpu(info->cap.mseq),
959 le64_to_cpu(info->cap.realm),
960 info->cap.flags, &new_cap);
962 /* set dir completion flag? */
963 if (S_ISDIR(inode->i_mode) &&
964 ci->i_files == 0 && ci->i_subdirs == 0 &&
965 (info_caps & CEPH_CAP_FILE_SHARED) &&
966 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
967 !__ceph_dir_is_complete(ci)) {
968 dout(" marking %p complete (empty)\n", inode);
969 i_size_write(inode, 0);
970 __ceph_dir_set_complete(ci,
971 atomic64_read(&ci->i_release_count),
972 atomic64_read(&ci->i_ordered_count));
977 dout(" %p got snap_caps %s\n", inode,
978 ceph_cap_string(info_caps));
979 ci->i_snap_caps |= info_caps;
981 __ceph_get_fmode(ci, cap_fmode);
983 } else if (cap_fmode >= 0) {
984 pr_warn("mds issued no caps on %llx.%llx\n",
986 __ceph_get_fmode(ci, cap_fmode);
989 if (iinfo->inline_version > 0 &&
990 iinfo->inline_version >= ci->i_inline_version) {
991 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
992 ci->i_inline_version = iinfo->inline_version;
993 if (ci->i_inline_version != CEPH_INLINE_NONE &&
994 (locked_page || (info_caps & cache_caps)))
998 spin_unlock(&ci->i_ceph_lock);
1001 ceph_fill_inline_data(inode, locked_page,
1002 iinfo->inline_data, iinfo->inline_len);
1005 wake_up_all(&ci->i_cap_wq);
1007 /* queue truncate if we saw i_size decrease */
1009 ceph_queue_vmtruncate(inode);
1011 /* populate frag tree */
1012 if (S_ISDIR(inode->i_mode))
1013 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1015 /* update delegation info? */
1017 ceph_fill_dirfrag(inode, dirinfo);
1022 ceph_put_cap(mdsc, new_cap);
1024 ceph_buffer_put(xattr_blob);
1025 ceph_put_string(pool_ns);
1030 * caller should hold session s_mutex.
1032 static void update_dentry_lease(struct dentry *dentry,
1033 struct ceph_mds_reply_lease *lease,
1034 struct ceph_mds_session *session,
1035 unsigned long from_time,
1036 struct ceph_vino *tgt_vino,
1037 struct ceph_vino *dir_vino)
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 struct ceph_mds_session *old_lease_session = NULL;
1047 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1048 * we expect a negative dentry.
1050 if (!tgt_vino && d_really_is_positive(dentry))
1053 if (tgt_vino && (d_really_is_negative(dentry) ||
1054 !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1057 spin_lock(&dentry->d_lock);
1058 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1059 dentry, duration, ttl);
1061 dir = d_inode(dentry->d_parent);
1063 /* make sure parent matches dir_vino */
1064 if (!ceph_ino_compare(dir, dir_vino))
1067 /* only track leases on regular dentries */
1068 if (ceph_snap(dir) != CEPH_NOSNAP)
1071 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1072 if (duration == 0) {
1073 __ceph_dentry_dir_lease_touch(di);
1077 if (di->lease_gen == session->s_cap_gen &&
1078 time_before(ttl, di->time))
1079 goto out_unlock; /* we already have a newer lease. */
1081 if (di->lease_session && di->lease_session != session) {
1082 old_lease_session = di->lease_session;
1083 di->lease_session = NULL;
1086 if (!di->lease_session)
1087 di->lease_session = ceph_get_mds_session(session);
1088 di->lease_gen = session->s_cap_gen;
1089 di->lease_seq = le32_to_cpu(lease->seq);
1090 di->lease_renew_after = half_ttl;
1091 di->lease_renew_from = 0;
1094 __ceph_dentry_lease_touch(di);
1096 spin_unlock(&dentry->d_lock);
1097 if (old_lease_session)
1098 ceph_put_mds_session(old_lease_session);
1102 * splice a dentry to an inode.
1103 * caller must hold directory i_mutex for this to be safe.
1105 static int splice_dentry(struct dentry **pdn, struct inode *in)
1107 struct dentry *dn = *pdn;
1108 struct dentry *realdn;
1110 BUG_ON(d_inode(dn));
1112 if (S_ISDIR(in->i_mode)) {
1113 /* If inode is directory, d_splice_alias() below will remove
1114 * 'realdn' from its origin parent. We need to ensure that
1115 * origin parent's readdir cache will not reference 'realdn'
1117 realdn = d_find_any_alias(in);
1119 struct ceph_dentry_info *di = ceph_dentry(realdn);
1120 spin_lock(&realdn->d_lock);
1122 realdn->d_op->d_prune(realdn);
1125 di->lease_shared_gen = 0;
1128 spin_unlock(&realdn->d_lock);
1133 /* dn must be unhashed */
1134 if (!d_unhashed(dn))
1136 realdn = d_splice_alias(in, dn);
1137 if (IS_ERR(realdn)) {
1138 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1139 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1140 return PTR_ERR(realdn);
1144 dout("dn %p (%d) spliced with %p (%d) "
1145 "inode %p ino %llx.%llx\n",
1147 realdn, d_count(realdn),
1148 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1152 BUG_ON(!ceph_dentry(dn));
1153 dout("dn %p attached to %p ino %llx.%llx\n",
1154 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1159 static int d_name_cmp(struct dentry *dentry, const char *name, size_t len)
1163 /* take d_lock to ensure dentry->d_name stability */
1164 spin_lock(&dentry->d_lock);
1165 ret = dentry->d_name.len - len;
1167 ret = memcmp(dentry->d_name.name, name, len);
1168 spin_unlock(&dentry->d_lock);
1173 * Incorporate results into the local cache. This is either just
1174 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1177 * A reply may contain
1178 * a directory inode along with a dentry.
1179 * and/or a target inode
1181 * Called with snap_rwsem (read).
1183 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1185 struct ceph_mds_session *session = req->r_session;
1186 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1187 struct inode *in = NULL;
1188 struct ceph_vino tvino, dvino;
1189 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1192 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1193 rinfo->head->is_dentry, rinfo->head->is_target);
1195 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1196 dout("fill_trace reply is empty!\n");
1197 if (rinfo->head->result == 0 && req->r_parent)
1198 ceph_invalidate_dir_request(req);
1202 if (rinfo->head->is_dentry) {
1203 struct inode *dir = req->r_parent;
1206 err = fill_inode(dir, NULL,
1207 &rinfo->diri, rinfo->dirfrag,
1208 session, req->r_request_started, -1,
1209 &req->r_caps_reservation);
1216 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1217 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1218 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1220 struct dentry *dn, *parent;
1222 BUG_ON(!rinfo->head->is_target);
1223 BUG_ON(req->r_dentry);
1225 parent = d_find_any_alias(dir);
1228 dname.name = rinfo->dname;
1229 dname.len = rinfo->dname_len;
1230 dname.hash = full_name_hash(parent, dname.name, dname.len);
1231 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1232 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1234 dn = d_lookup(parent, &dname);
1235 dout("d_lookup on parent=%p name=%.*s got %p\n",
1236 parent, dname.len, dname.name, dn);
1239 dn = d_alloc(parent, &dname);
1240 dout("d_alloc %p '%.*s' = %p\n", parent,
1241 dname.len, dname.name, dn);
1248 } else if (d_really_is_positive(dn) &&
1249 (ceph_ino(d_inode(dn)) != tvino.ino ||
1250 ceph_snap(d_inode(dn)) != tvino.snap)) {
1251 dout(" dn %p points to wrong inode %p\n",
1253 ceph_dir_clear_ordered(dir);
1264 if (rinfo->head->is_target) {
1265 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1266 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1268 in = ceph_get_inode(sb, tvino);
1273 req->r_target_inode = in;
1275 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1276 session, req->r_request_started,
1277 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1278 rinfo->head->result == 0) ? req->r_fmode : -1,
1279 &req->r_caps_reservation);
1281 pr_err("fill_inode badness %p %llx.%llx\n",
1282 in, ceph_vinop(in));
1288 * ignore null lease/binding on snapdir ENOENT, or else we
1289 * will have trouble splicing in the virtual snapdir later
1291 if (rinfo->head->is_dentry &&
1292 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1293 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1294 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1295 fsc->mount_options->snapdir_name,
1296 req->r_dentry->d_name.len))) {
1298 * lookup link rename : null -> possibly existing inode
1299 * mknod symlink mkdir : null -> new inode
1300 * unlink : linked -> null
1302 struct inode *dir = req->r_parent;
1303 struct dentry *dn = req->r_dentry;
1304 bool have_dir_cap, have_lease;
1308 BUG_ON(d_inode(dn->d_parent) != dir);
1310 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1311 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1313 BUG_ON(ceph_ino(dir) != dvino.ino);
1314 BUG_ON(ceph_snap(dir) != dvino.snap);
1316 /* do we have a lease on the whole dir? */
1318 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1319 CEPH_CAP_FILE_SHARED);
1321 /* do we have a dn lease? */
1322 have_lease = have_dir_cap ||
1323 le32_to_cpu(rinfo->dlease->duration_ms);
1325 dout("fill_trace no dentry lease or dir cap\n");
1328 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1329 struct inode *olddir = req->r_old_dentry_dir;
1332 dout(" src %p '%pd' dst %p '%pd'\n",
1336 dout("fill_trace doing d_move %p -> %p\n",
1337 req->r_old_dentry, dn);
1339 /* d_move screws up sibling dentries' offsets */
1340 ceph_dir_clear_ordered(dir);
1341 ceph_dir_clear_ordered(olddir);
1343 d_move(req->r_old_dentry, dn);
1344 dout(" src %p '%pd' dst %p '%pd'\n",
1349 /* ensure target dentry is invalidated, despite
1350 rehashing bug in vfs_rename_dir */
1351 ceph_invalidate_dentry_lease(dn);
1353 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1354 ceph_dentry(req->r_old_dentry)->offset);
1356 /* swap r_dentry and r_old_dentry in case that
1357 * splice_dentry() gets called later. This is safe
1358 * because no other place will use them */
1359 req->r_dentry = req->r_old_dentry;
1360 req->r_old_dentry = dn;
1365 if (!rinfo->head->is_target) {
1366 dout("fill_trace null dentry\n");
1367 if (d_really_is_positive(dn)) {
1368 dout("d_delete %p\n", dn);
1369 ceph_dir_clear_ordered(dir);
1371 } else if (have_lease) {
1374 update_dentry_lease(dn, rinfo->dlease,
1376 req->r_request_started,
1382 /* attach proper inode */
1383 if (d_really_is_negative(dn)) {
1384 ceph_dir_clear_ordered(dir);
1386 err = splice_dentry(&req->r_dentry, in);
1389 dn = req->r_dentry; /* may have spliced */
1390 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1391 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1392 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1399 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1400 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1401 update_dentry_lease(dn, rinfo->dlease, session,
1402 req->r_request_started,
1405 dout(" final dn %p\n", dn);
1406 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1407 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1408 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1409 struct inode *dir = req->r_parent;
1411 /* fill out a snapdir LOOKUPSNAP dentry */
1413 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1414 BUG_ON(!req->r_dentry);
1415 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1416 ceph_dir_clear_ordered(dir);
1418 err = splice_dentry(&req->r_dentry, in);
1421 } else if (rinfo->head->is_dentry &&
1422 !d_name_cmp(req->r_dentry, rinfo->dname, rinfo->dname_len)) {
1423 struct ceph_vino *ptvino = NULL;
1425 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1426 le32_to_cpu(rinfo->dlease->duration_ms)) {
1427 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1428 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1430 if (rinfo->head->is_target) {
1431 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1432 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1436 update_dentry_lease(req->r_dentry, rinfo->dlease,
1437 session, req->r_request_started, ptvino,
1440 dout("%s: no dentry lease or dir cap\n", __func__);
1444 dout("fill_trace done err=%d\n", err);
1449 * Prepopulate our cache with readdir results, leases, etc.
1451 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1452 struct ceph_mds_session *session)
1454 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1457 for (i = 0; i < rinfo->dir_nr; i++) {
1458 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1459 struct ceph_vino vino;
1463 vino.ino = le64_to_cpu(rde->inode.in->ino);
1464 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1466 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1469 dout("new_inode badness got %d\n", err);
1472 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1473 req->r_request_started, -1,
1474 &req->r_caps_reservation);
1476 pr_err("fill_inode badness on %p got %d\n", in, rc);
1479 /* avoid calling iput_final() in mds dispatch threads */
1480 ceph_async_iput(in);
1486 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1490 put_page(ctl->page);
1495 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1496 struct ceph_readdir_cache_control *ctl,
1497 struct ceph_mds_request *req)
1499 struct ceph_inode_info *ci = ceph_inode(dir);
1500 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1501 unsigned idx = ctl->index % nsize;
1502 pgoff_t pgoff = ctl->index / nsize;
1504 if (!ctl->page || pgoff != page_index(ctl->page)) {
1505 ceph_readdir_cache_release(ctl);
1507 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1509 ctl->page = find_lock_page(&dir->i_data, pgoff);
1512 return idx == 0 ? -ENOMEM : 0;
1514 /* reading/filling the cache are serialized by
1515 * i_mutex, no need to use page lock */
1516 unlock_page(ctl->page);
1517 ctl->dentries = kmap(ctl->page);
1519 memset(ctl->dentries, 0, PAGE_SIZE);
1522 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1523 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1524 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1525 ctl->dentries[idx] = dn;
1528 dout("disable readdir cache\n");
1534 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1535 struct ceph_mds_session *session)
1537 struct dentry *parent = req->r_dentry;
1538 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1539 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1543 int err = 0, skipped = 0, ret, i;
1544 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1545 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1548 struct ceph_readdir_cache_control cache_ctl = {};
1550 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1551 return readdir_prepopulate_inodes_only(req, session);
1553 if (rinfo->hash_order) {
1555 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1557 strlen(req->r_path2));
1558 last_hash = ceph_frag_value(last_hash);
1559 } else if (rinfo->offset_hash) {
1560 /* mds understands offset_hash */
1561 WARN_ON_ONCE(req->r_readdir_offset != 2);
1562 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1566 if (rinfo->dir_dir &&
1567 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1568 dout("readdir_prepopulate got new frag %x -> %x\n",
1569 frag, le32_to_cpu(rinfo->dir_dir->frag));
1570 frag = le32_to_cpu(rinfo->dir_dir->frag);
1571 if (!rinfo->hash_order)
1572 req->r_readdir_offset = 2;
1575 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1576 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1577 rinfo->dir_nr, parent);
1579 dout("readdir_prepopulate %d items under dn %p\n",
1580 rinfo->dir_nr, parent);
1582 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1584 if (ceph_frag_is_leftmost(frag) &&
1585 req->r_readdir_offset == 2 &&
1586 !(rinfo->hash_order && last_hash)) {
1587 /* note dir version at start of readdir so we can
1588 * tell if any dentries get dropped */
1589 req->r_dir_release_cnt =
1590 atomic64_read(&ci->i_release_count);
1591 req->r_dir_ordered_cnt =
1592 atomic64_read(&ci->i_ordered_count);
1593 req->r_readdir_cache_idx = 0;
1597 cache_ctl.index = req->r_readdir_cache_idx;
1598 fpos_offset = req->r_readdir_offset;
1600 /* FIXME: release caps/leases if error occurs */
1601 for (i = 0; i < rinfo->dir_nr; i++) {
1602 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1603 struct ceph_vino tvino, dvino;
1605 dname.name = rde->name;
1606 dname.len = rde->name_len;
1607 dname.hash = full_name_hash(parent, dname.name, dname.len);
1609 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1610 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1612 if (rinfo->hash_order) {
1613 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1614 rde->name, rde->name_len);
1615 hash = ceph_frag_value(hash);
1616 if (hash != last_hash)
1619 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1621 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1625 dn = d_lookup(parent, &dname);
1626 dout("d_lookup on parent=%p name=%.*s got %p\n",
1627 parent, dname.len, dname.name, dn);
1630 dn = d_alloc(parent, &dname);
1631 dout("d_alloc %p '%.*s' = %p\n", parent,
1632 dname.len, dname.name, dn);
1634 dout("d_alloc badness\n");
1638 } else if (d_really_is_positive(dn) &&
1639 (ceph_ino(d_inode(dn)) != tvino.ino ||
1640 ceph_snap(d_inode(dn)) != tvino.snap)) {
1641 struct ceph_dentry_info *di = ceph_dentry(dn);
1642 dout(" dn %p points to wrong inode %p\n",
1645 spin_lock(&dn->d_lock);
1646 if (di->offset > 0 &&
1647 di->lease_shared_gen ==
1648 atomic_read(&ci->i_shared_gen)) {
1649 __ceph_dir_clear_ordered(ci);
1652 spin_unlock(&dn->d_lock);
1660 if (d_really_is_positive(dn)) {
1663 in = ceph_get_inode(parent->d_sb, tvino);
1665 dout("new_inode badness\n");
1673 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1674 req->r_request_started, -1,
1675 &req->r_caps_reservation);
1677 pr_err("fill_inode badness on %p\n", in);
1678 if (d_really_is_negative(dn)) {
1679 /* avoid calling iput_final() in mds
1680 * dispatch threads */
1681 ceph_async_iput(in);
1688 if (d_really_is_negative(dn)) {
1689 if (ceph_security_xattr_deadlock(in)) {
1690 dout(" skip splicing dn %p to inode %p"
1691 " (security xattr deadlock)\n", dn, in);
1692 ceph_async_iput(in);
1697 err = splice_dentry(&dn, in);
1702 ceph_dentry(dn)->offset = rde->offset;
1704 dvino = ceph_vino(d_inode(parent));
1705 update_dentry_lease(dn, rde->lease, req->r_session,
1706 req->r_request_started, &tvino, &dvino);
1708 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1709 ret = fill_readdir_cache(d_inode(parent), dn,
1718 if (err == 0 && skipped == 0) {
1719 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1720 req->r_readdir_cache_idx = cache_ctl.index;
1722 ceph_readdir_cache_release(&cache_ctl);
1723 dout("readdir_prepopulate done\n");
1727 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1729 struct ceph_inode_info *ci = ceph_inode(inode);
1732 spin_lock(&ci->i_ceph_lock);
1733 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1734 i_size_write(inode, size);
1735 inode->i_blocks = calc_inode_blocks(size);
1737 ret = __ceph_should_report_size(ci);
1739 spin_unlock(&ci->i_ceph_lock);
1744 * Put reference to inode, but avoid calling iput_final() in current thread.
1745 * iput_final() may wait for reahahead pages. The wait can cause deadlock in
1748 void ceph_async_iput(struct inode *inode)
1753 if (atomic_add_unless(&inode->i_count, -1, 1))
1755 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1756 &ceph_inode(inode)->i_work))
1758 /* queue work failed, i_count must be at least 2 */
1763 * Write back inode data in a worker thread. (This can't be done
1764 * in the message handler context.)
1766 void ceph_queue_writeback(struct inode *inode)
1768 struct ceph_inode_info *ci = ceph_inode(inode);
1769 set_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask);
1772 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1774 dout("ceph_queue_writeback %p\n", inode);
1776 dout("ceph_queue_writeback %p already queued, mask=%lx\n",
1777 inode, ci->i_work_mask);
1783 * queue an async invalidation
1785 void ceph_queue_invalidate(struct inode *inode)
1787 struct ceph_inode_info *ci = ceph_inode(inode);
1788 set_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask);
1791 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1792 &ceph_inode(inode)->i_work)) {
1793 dout("ceph_queue_invalidate %p\n", inode);
1795 dout("ceph_queue_invalidate %p already queued, mask=%lx\n",
1796 inode, ci->i_work_mask);
1802 * Queue an async vmtruncate. If we fail to queue work, we will handle
1803 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1805 void ceph_queue_vmtruncate(struct inode *inode)
1807 struct ceph_inode_info *ci = ceph_inode(inode);
1808 set_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask);
1811 if (queue_work(ceph_inode_to_client(inode)->inode_wq,
1813 dout("ceph_queue_vmtruncate %p\n", inode);
1815 dout("ceph_queue_vmtruncate %p already queued, mask=%lx\n",
1816 inode, ci->i_work_mask);
1821 static void ceph_do_invalidate_pages(struct inode *inode)
1823 struct ceph_inode_info *ci = ceph_inode(inode);
1824 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1828 mutex_lock(&ci->i_truncate_mutex);
1830 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1831 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1832 inode, ceph_ino(inode));
1833 mapping_set_error(inode->i_mapping, -EIO);
1834 truncate_pagecache(inode, 0);
1835 mutex_unlock(&ci->i_truncate_mutex);
1839 spin_lock(&ci->i_ceph_lock);
1840 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1841 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1842 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1843 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1845 spin_unlock(&ci->i_ceph_lock);
1846 mutex_unlock(&ci->i_truncate_mutex);
1849 orig_gen = ci->i_rdcache_gen;
1850 spin_unlock(&ci->i_ceph_lock);
1852 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1853 pr_err("invalidate_pages %p fails\n", inode);
1856 spin_lock(&ci->i_ceph_lock);
1857 if (orig_gen == ci->i_rdcache_gen &&
1858 orig_gen == ci->i_rdcache_revoking) {
1859 dout("invalidate_pages %p gen %d successful\n", inode,
1861 ci->i_rdcache_revoking--;
1864 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1865 inode, orig_gen, ci->i_rdcache_gen,
1866 ci->i_rdcache_revoking);
1867 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1870 spin_unlock(&ci->i_ceph_lock);
1871 mutex_unlock(&ci->i_truncate_mutex);
1874 ceph_check_caps(ci, 0, NULL);
1878 * Make sure any pending truncation is applied before doing anything
1879 * that may depend on it.
1881 void __ceph_do_pending_vmtruncate(struct inode *inode)
1883 struct ceph_inode_info *ci = ceph_inode(inode);
1885 int wrbuffer_refs, finish = 0;
1887 mutex_lock(&ci->i_truncate_mutex);
1889 spin_lock(&ci->i_ceph_lock);
1890 if (ci->i_truncate_pending == 0) {
1891 dout("__do_pending_vmtruncate %p none pending\n", inode);
1892 spin_unlock(&ci->i_ceph_lock);
1893 mutex_unlock(&ci->i_truncate_mutex);
1898 * make sure any dirty snapped pages are flushed before we
1899 * possibly truncate them.. so write AND block!
1901 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1902 spin_unlock(&ci->i_ceph_lock);
1903 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1905 filemap_write_and_wait_range(&inode->i_data, 0,
1906 inode->i_sb->s_maxbytes);
1910 /* there should be no reader or writer */
1911 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1913 to = ci->i_truncate_size;
1914 wrbuffer_refs = ci->i_wrbuffer_ref;
1915 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1916 ci->i_truncate_pending, to);
1917 spin_unlock(&ci->i_ceph_lock);
1919 truncate_pagecache(inode, to);
1921 spin_lock(&ci->i_ceph_lock);
1922 if (to == ci->i_truncate_size) {
1923 ci->i_truncate_pending = 0;
1926 spin_unlock(&ci->i_ceph_lock);
1930 mutex_unlock(&ci->i_truncate_mutex);
1932 if (wrbuffer_refs == 0)
1933 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1935 wake_up_all(&ci->i_cap_wq);
1938 static void ceph_inode_work(struct work_struct *work)
1940 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1942 struct inode *inode = &ci->vfs_inode;
1944 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1945 dout("writeback %p\n", inode);
1946 filemap_fdatawrite(&inode->i_data);
1948 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1949 ceph_do_invalidate_pages(inode);
1951 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1952 __ceph_do_pending_vmtruncate(inode);
1960 static const struct inode_operations ceph_symlink_iops = {
1961 .get_link = simple_get_link,
1962 .setattr = ceph_setattr,
1963 .getattr = ceph_getattr,
1964 .listxattr = ceph_listxattr,
1967 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1969 struct ceph_inode_info *ci = ceph_inode(inode);
1970 const unsigned int ia_valid = attr->ia_valid;
1971 struct ceph_mds_request *req;
1972 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1973 struct ceph_cap_flush *prealloc_cf;
1975 int release = 0, dirtied = 0;
1978 int inode_dirty_flags = 0;
1979 bool lock_snap_rwsem = false;
1981 prealloc_cf = ceph_alloc_cap_flush();
1985 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1988 ceph_free_cap_flush(prealloc_cf);
1989 return PTR_ERR(req);
1992 spin_lock(&ci->i_ceph_lock);
1993 issued = __ceph_caps_issued(ci, NULL);
1995 if (!ci->i_head_snapc &&
1996 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1997 lock_snap_rwsem = true;
1998 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1999 spin_unlock(&ci->i_ceph_lock);
2000 down_read(&mdsc->snap_rwsem);
2001 spin_lock(&ci->i_ceph_lock);
2002 issued = __ceph_caps_issued(ci, NULL);
2006 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2008 if (ia_valid & ATTR_UID) {
2009 dout("setattr %p uid %d -> %d\n", inode,
2010 from_kuid(&init_user_ns, inode->i_uid),
2011 from_kuid(&init_user_ns, attr->ia_uid));
2012 if (issued & CEPH_CAP_AUTH_EXCL) {
2013 inode->i_uid = attr->ia_uid;
2014 dirtied |= CEPH_CAP_AUTH_EXCL;
2015 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2016 !uid_eq(attr->ia_uid, inode->i_uid)) {
2017 req->r_args.setattr.uid = cpu_to_le32(
2018 from_kuid(&init_user_ns, attr->ia_uid));
2019 mask |= CEPH_SETATTR_UID;
2020 release |= CEPH_CAP_AUTH_SHARED;
2023 if (ia_valid & ATTR_GID) {
2024 dout("setattr %p gid %d -> %d\n", inode,
2025 from_kgid(&init_user_ns, inode->i_gid),
2026 from_kgid(&init_user_ns, attr->ia_gid));
2027 if (issued & CEPH_CAP_AUTH_EXCL) {
2028 inode->i_gid = attr->ia_gid;
2029 dirtied |= CEPH_CAP_AUTH_EXCL;
2030 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2031 !gid_eq(attr->ia_gid, inode->i_gid)) {
2032 req->r_args.setattr.gid = cpu_to_le32(
2033 from_kgid(&init_user_ns, attr->ia_gid));
2034 mask |= CEPH_SETATTR_GID;
2035 release |= CEPH_CAP_AUTH_SHARED;
2038 if (ia_valid & ATTR_MODE) {
2039 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2041 if (issued & CEPH_CAP_AUTH_EXCL) {
2042 inode->i_mode = attr->ia_mode;
2043 dirtied |= CEPH_CAP_AUTH_EXCL;
2044 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2045 attr->ia_mode != inode->i_mode) {
2046 inode->i_mode = attr->ia_mode;
2047 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2048 mask |= CEPH_SETATTR_MODE;
2049 release |= CEPH_CAP_AUTH_SHARED;
2053 if (ia_valid & ATTR_ATIME) {
2054 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2055 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2056 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2057 if (issued & CEPH_CAP_FILE_EXCL) {
2058 ci->i_time_warp_seq++;
2059 inode->i_atime = attr->ia_atime;
2060 dirtied |= CEPH_CAP_FILE_EXCL;
2061 } else if ((issued & CEPH_CAP_FILE_WR) &&
2062 timespec64_compare(&inode->i_atime,
2063 &attr->ia_atime) < 0) {
2064 inode->i_atime = attr->ia_atime;
2065 dirtied |= CEPH_CAP_FILE_WR;
2066 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2067 !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2068 ceph_encode_timespec64(&req->r_args.setattr.atime,
2070 mask |= CEPH_SETATTR_ATIME;
2071 release |= CEPH_CAP_FILE_SHARED |
2072 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2075 if (ia_valid & ATTR_MTIME) {
2076 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2077 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2078 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2079 if (issued & CEPH_CAP_FILE_EXCL) {
2080 ci->i_time_warp_seq++;
2081 inode->i_mtime = attr->ia_mtime;
2082 dirtied |= CEPH_CAP_FILE_EXCL;
2083 } else if ((issued & CEPH_CAP_FILE_WR) &&
2084 timespec64_compare(&inode->i_mtime,
2085 &attr->ia_mtime) < 0) {
2086 inode->i_mtime = attr->ia_mtime;
2087 dirtied |= CEPH_CAP_FILE_WR;
2088 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2089 !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2090 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2092 mask |= CEPH_SETATTR_MTIME;
2093 release |= CEPH_CAP_FILE_SHARED |
2094 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2097 if (ia_valid & ATTR_SIZE) {
2098 dout("setattr %p size %lld -> %lld\n", inode,
2099 inode->i_size, attr->ia_size);
2100 if ((issued & CEPH_CAP_FILE_EXCL) &&
2101 attr->ia_size > inode->i_size) {
2102 i_size_write(inode, attr->ia_size);
2103 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2104 ci->i_reported_size = attr->ia_size;
2105 dirtied |= CEPH_CAP_FILE_EXCL;
2106 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2107 attr->ia_size != inode->i_size) {
2108 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2109 req->r_args.setattr.old_size =
2110 cpu_to_le64(inode->i_size);
2111 mask |= CEPH_SETATTR_SIZE;
2112 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2113 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2117 /* these do nothing */
2118 if (ia_valid & ATTR_CTIME) {
2119 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2120 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2121 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2122 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2123 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2124 only ? "ctime only" : "ignored");
2127 * if kernel wants to dirty ctime but nothing else,
2128 * we need to choose a cap to dirty under, or do
2129 * a almost-no-op setattr
2131 if (issued & CEPH_CAP_AUTH_EXCL)
2132 dirtied |= CEPH_CAP_AUTH_EXCL;
2133 else if (issued & CEPH_CAP_FILE_EXCL)
2134 dirtied |= CEPH_CAP_FILE_EXCL;
2135 else if (issued & CEPH_CAP_XATTR_EXCL)
2136 dirtied |= CEPH_CAP_XATTR_EXCL;
2138 mask |= CEPH_SETATTR_CTIME;
2141 if (ia_valid & ATTR_FILE)
2142 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2145 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2147 inode->i_ctime = attr->ia_ctime;
2151 spin_unlock(&ci->i_ceph_lock);
2152 if (lock_snap_rwsem)
2153 up_read(&mdsc->snap_rwsem);
2155 if (inode_dirty_flags)
2156 __mark_inode_dirty(inode, inode_dirty_flags);
2160 req->r_inode = inode;
2162 req->r_inode_drop = release;
2163 req->r_args.setattr.mask = cpu_to_le32(mask);
2164 req->r_num_caps = 1;
2165 req->r_stamp = attr->ia_ctime;
2166 err = ceph_mdsc_do_request(mdsc, NULL, req);
2168 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2169 ceph_cap_string(dirtied), mask);
2171 ceph_mdsc_put_request(req);
2172 ceph_free_cap_flush(prealloc_cf);
2174 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2175 __ceph_do_pending_vmtruncate(inode);
2183 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2185 struct inode *inode = d_inode(dentry);
2186 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2189 if (ceph_snap(inode) != CEPH_NOSNAP)
2192 err = setattr_prepare(dentry, attr);
2196 if ((attr->ia_valid & ATTR_SIZE) &&
2197 attr->ia_size > max(inode->i_size, fsc->max_file_size))
2200 if ((attr->ia_valid & ATTR_SIZE) &&
2201 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2204 err = __ceph_setattr(inode, attr);
2206 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2207 err = posix_acl_chmod(inode, attr->ia_mode);
2213 * Verify that we have a lease on the given mask. If not,
2214 * do a getattr against an mds.
2216 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2217 int mask, bool force)
2219 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2220 struct ceph_mds_client *mdsc = fsc->mdsc;
2221 struct ceph_mds_request *req;
2225 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2226 dout("do_getattr inode %p SNAPDIR\n", inode);
2230 dout("do_getattr inode %p mask %s mode 0%o\n",
2231 inode, ceph_cap_string(mask), inode->i_mode);
2232 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2235 mode = (mask & CEPH_STAT_RSTAT) ? USE_AUTH_MDS : USE_ANY_MDS;
2236 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2238 return PTR_ERR(req);
2239 req->r_inode = inode;
2241 req->r_num_caps = 1;
2242 req->r_args.getattr.mask = cpu_to_le32(mask);
2243 req->r_locked_page = locked_page;
2244 err = ceph_mdsc_do_request(mdsc, NULL, req);
2245 if (locked_page && err == 0) {
2246 u64 inline_version = req->r_reply_info.targeti.inline_version;
2247 if (inline_version == 0) {
2248 /* the reply is supposed to contain inline data */
2250 } else if (inline_version == CEPH_INLINE_NONE) {
2253 err = req->r_reply_info.targeti.inline_len;
2256 ceph_mdsc_put_request(req);
2257 dout("do_getattr result=%d\n", err);
2263 * Check inode permissions. We verify we have a valid value for
2264 * the AUTH cap, then call the generic handler.
2266 int ceph_permission(struct inode *inode, int mask)
2270 if (mask & MAY_NOT_BLOCK)
2273 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2276 err = generic_permission(inode, mask);
2280 /* Craft a mask of needed caps given a set of requested statx attrs. */
2281 static int statx_to_caps(u32 want)
2285 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME))
2286 mask |= CEPH_CAP_AUTH_SHARED;
2288 if (want & (STATX_NLINK|STATX_CTIME))
2289 mask |= CEPH_CAP_LINK_SHARED;
2291 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2293 mask |= CEPH_CAP_FILE_SHARED;
2295 if (want & (STATX_CTIME))
2296 mask |= CEPH_CAP_XATTR_SHARED;
2302 * Get all the attributes. If we have sufficient caps for the requested attrs,
2303 * then we can avoid talking to the MDS at all.
2305 int ceph_getattr(const struct path *path, struct kstat *stat,
2306 u32 request_mask, unsigned int flags)
2308 struct inode *inode = d_inode(path->dentry);
2309 struct ceph_inode_info *ci = ceph_inode(inode);
2312 /* Skip the getattr altogether if we're asked not to sync */
2313 if (!(flags & AT_STATX_DONT_SYNC)) {
2314 err = ceph_do_getattr(inode, statx_to_caps(request_mask),
2315 flags & AT_STATX_FORCE_SYNC);
2320 generic_fillattr(inode, stat);
2321 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2322 if (ceph_snap(inode) == CEPH_NOSNAP)
2323 stat->dev = inode->i_sb->s_dev;
2325 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2327 if (S_ISDIR(inode->i_mode)) {
2328 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2330 stat->size = ci->i_rbytes;
2332 stat->size = ci->i_files + ci->i_subdirs;
2334 stat->blksize = 65536;
2336 * Some applications rely on the number of st_nlink
2337 * value on directories to be either 0 (if unlinked)
2338 * or 2 + number of subdirectories.
2340 if (stat->nlink == 1)
2341 /* '.' + '..' + subdirs */
2342 stat->nlink = 1 + 1 + ci->i_subdirs;
2345 /* Mask off any higher bits (e.g. btime) until we have support */
2346 stat->result_mask = request_mask & STATX_BASIC_STATS;