1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <linux/string.h>
7 #include <linux/uaccess.h>
8 #include <linux/kernel.h>
9 #include <linux/writeback.h>
10 #include <linux/vmalloc.h>
11 #include <linux/xattr.h>
12 #include <linux/posix_acl.h>
13 #include <linux/random.h>
14 #include <linux/sort.h>
17 #include "mds_client.h"
19 #include <linux/ceph/decode.h>
22 * Ceph inode operations
24 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
25 * setattr, etc.), xattr helpers, and helpers for assimilating
26 * metadata returned by the MDS into our cache.
28 * Also define helpers for doing asynchronous writeback, invalidation,
29 * and truncation for the benefit of those who can't afford to block
30 * (typically because they are in the message handler path).
33 static const struct inode_operations ceph_symlink_iops;
35 static void ceph_invalidate_work(struct work_struct *work);
36 static void ceph_writeback_work(struct work_struct *work);
37 static void ceph_vmtruncate_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 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
45 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
49 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
52 ino_t t = ceph_vino_to_ino(vino);
54 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
56 return ERR_PTR(-ENOMEM);
57 if (inode->i_state & I_NEW) {
58 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
59 inode, ceph_vinop(inode), (u64)inode->i_ino);
60 unlock_new_inode(inode);
63 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
69 * get/constuct snapdir inode for a given directory
71 struct inode *ceph_get_snapdir(struct inode *parent)
73 struct ceph_vino vino = {
74 .ino = ceph_ino(parent),
77 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
78 struct ceph_inode_info *ci = ceph_inode(inode);
80 BUG_ON(!S_ISDIR(parent->i_mode));
83 inode->i_mode = parent->i_mode;
84 inode->i_uid = parent->i_uid;
85 inode->i_gid = parent->i_gid;
86 inode->i_op = &ceph_snapdir_iops;
87 inode->i_fop = &ceph_snapdir_fops;
88 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
93 const struct inode_operations ceph_file_iops = {
94 .permission = ceph_permission,
95 .setattr = ceph_setattr,
96 .getattr = ceph_getattr,
97 .listxattr = ceph_listxattr,
98 .get_acl = ceph_get_acl,
99 .set_acl = ceph_set_acl,
104 * We use a 'frag tree' to keep track of the MDS's directory fragments
105 * for a given inode (usually there is just a single fragment). We
106 * need to know when a child frag is delegated to a new MDS, or when
107 * it is flagged as replicated, so we can direct our requests
112 * find/create a frag in the tree
114 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
118 struct rb_node *parent = NULL;
119 struct ceph_inode_frag *frag;
122 p = &ci->i_fragtree.rb_node;
125 frag = rb_entry(parent, struct ceph_inode_frag, node);
126 c = ceph_frag_compare(f, frag->frag);
135 frag = kmalloc(sizeof(*frag), GFP_NOFS);
137 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
138 "frag %x\n", &ci->vfs_inode,
139 ceph_vinop(&ci->vfs_inode), f);
140 return ERR_PTR(-ENOMEM);
147 rb_link_node(&frag->node, parent, p);
148 rb_insert_color(&frag->node, &ci->i_fragtree);
150 dout("get_or_create_frag added %llx.%llx frag %x\n",
151 ceph_vinop(&ci->vfs_inode), f);
156 * find a specific frag @f
158 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
160 struct rb_node *n = ci->i_fragtree.rb_node;
163 struct ceph_inode_frag *frag =
164 rb_entry(n, struct ceph_inode_frag, node);
165 int c = ceph_frag_compare(f, frag->frag);
177 * Choose frag containing the given value @v. If @pfrag is
178 * specified, copy the frag delegation info to the caller if
181 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
182 struct ceph_inode_frag *pfrag, int *found)
184 u32 t = ceph_frag_make(0, 0);
185 struct ceph_inode_frag *frag;
193 WARN_ON(!ceph_frag_contains_value(t, v));
194 frag = __ceph_find_frag(ci, t);
196 break; /* t is a leaf */
197 if (frag->split_by == 0) {
199 memcpy(pfrag, frag, sizeof(*pfrag));
206 nway = 1 << frag->split_by;
207 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
208 frag->split_by, nway);
209 for (i = 0; i < nway; i++) {
210 n = ceph_frag_make_child(t, frag->split_by, i);
211 if (ceph_frag_contains_value(n, v)) {
218 dout("choose_frag(%x) = %x\n", v, t);
223 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
224 struct ceph_inode_frag *pfrag, int *found)
227 mutex_lock(&ci->i_fragtree_mutex);
228 ret = __ceph_choose_frag(ci, v, pfrag, found);
229 mutex_unlock(&ci->i_fragtree_mutex);
234 * Process dirfrag (delegation) info from the mds. Include leaf
235 * fragment in tree ONLY if ndist > 0. Otherwise, only
236 * branches/splits are included in i_fragtree)
238 static int ceph_fill_dirfrag(struct inode *inode,
239 struct ceph_mds_reply_dirfrag *dirinfo)
241 struct ceph_inode_info *ci = ceph_inode(inode);
242 struct ceph_inode_frag *frag;
243 u32 id = le32_to_cpu(dirinfo->frag);
244 int mds = le32_to_cpu(dirinfo->auth);
245 int ndist = le32_to_cpu(dirinfo->ndist);
250 spin_lock(&ci->i_ceph_lock);
252 diri_auth = ci->i_auth_cap->mds;
253 spin_unlock(&ci->i_ceph_lock);
255 if (mds == -1) /* CDIR_AUTH_PARENT */
258 mutex_lock(&ci->i_fragtree_mutex);
259 if (ndist == 0 && mds == diri_auth) {
260 /* no delegation info needed. */
261 frag = __ceph_find_frag(ci, id);
264 if (frag->split_by == 0) {
265 /* tree leaf, remove */
266 dout("fill_dirfrag removed %llx.%llx frag %x"
267 " (no ref)\n", ceph_vinop(inode), id);
268 rb_erase(&frag->node, &ci->i_fragtree);
271 /* tree branch, keep and clear */
272 dout("fill_dirfrag cleared %llx.%llx frag %x"
273 " referral\n", ceph_vinop(inode), id);
281 /* find/add this frag to store mds delegation info */
282 frag = __get_or_create_frag(ci, id);
284 /* this is not the end of the world; we can continue
285 with bad/inaccurate delegation info */
286 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
287 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
293 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
294 for (i = 0; i < frag->ndist; i++)
295 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
296 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
297 ceph_vinop(inode), frag->frag, frag->ndist);
300 mutex_unlock(&ci->i_fragtree_mutex);
304 static int frag_tree_split_cmp(const void *l, const void *r)
306 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
307 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
308 return ceph_frag_compare(le32_to_cpu(ls->frag),
309 le32_to_cpu(rs->frag));
312 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
315 return f == ceph_frag_make(0, 0);
316 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
318 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
321 static int ceph_fill_fragtree(struct inode *inode,
322 struct ceph_frag_tree_head *fragtree,
323 struct ceph_mds_reply_dirfrag *dirinfo)
325 struct ceph_inode_info *ci = ceph_inode(inode);
326 struct ceph_inode_frag *frag, *prev_frag = NULL;
327 struct rb_node *rb_node;
328 unsigned i, split_by, nsplits;
332 mutex_lock(&ci->i_fragtree_mutex);
333 nsplits = le32_to_cpu(fragtree->nsplits);
334 if (nsplits != ci->i_fragtree_nsplits) {
336 } else if (nsplits) {
337 i = prandom_u32() % nsplits;
338 id = le32_to_cpu(fragtree->splits[i].frag);
339 if (!__ceph_find_frag(ci, id))
341 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
342 rb_node = rb_first(&ci->i_fragtree);
343 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
344 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
347 if (!update && dirinfo) {
348 id = le32_to_cpu(dirinfo->frag);
349 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
356 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
357 frag_tree_split_cmp, NULL);
360 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
361 rb_node = rb_first(&ci->i_fragtree);
362 for (i = 0; i < nsplits; i++) {
363 id = le32_to_cpu(fragtree->splits[i].frag);
364 split_by = le32_to_cpu(fragtree->splits[i].by);
365 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
366 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
367 "frag %x split by %d\n", ceph_vinop(inode),
368 i, nsplits, id, split_by);
373 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
374 if (ceph_frag_compare(frag->frag, id) >= 0) {
375 if (frag->frag != id)
378 rb_node = rb_next(rb_node);
381 rb_node = rb_next(rb_node);
382 /* delete stale split/leaf node */
383 if (frag->split_by > 0 ||
384 !is_frag_child(frag->frag, prev_frag)) {
385 rb_erase(&frag->node, &ci->i_fragtree);
386 if (frag->split_by > 0)
387 ci->i_fragtree_nsplits--;
393 frag = __get_or_create_frag(ci, id);
397 if (frag->split_by == 0)
398 ci->i_fragtree_nsplits++;
399 frag->split_by = split_by;
400 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
404 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
405 rb_node = rb_next(rb_node);
406 /* delete stale split/leaf node */
407 if (frag->split_by > 0 ||
408 !is_frag_child(frag->frag, prev_frag)) {
409 rb_erase(&frag->node, &ci->i_fragtree);
410 if (frag->split_by > 0)
411 ci->i_fragtree_nsplits--;
416 mutex_unlock(&ci->i_fragtree_mutex);
421 * initialize a newly allocated inode.
423 struct inode *ceph_alloc_inode(struct super_block *sb)
425 struct ceph_inode_info *ci;
428 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
432 dout("alloc_inode %p\n", &ci->vfs_inode);
434 spin_lock_init(&ci->i_ceph_lock);
437 ci->i_inline_version = 0;
438 ci->i_time_warp_seq = 0;
439 ci->i_ceph_flags = 0;
440 atomic64_set(&ci->i_ordered_count, 1);
441 atomic64_set(&ci->i_release_count, 1);
442 atomic64_set(&ci->i_complete_seq[0], 0);
443 atomic64_set(&ci->i_complete_seq[1], 0);
444 ci->i_symlink = NULL;
446 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
447 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
449 ci->i_fragtree = RB_ROOT;
450 mutex_init(&ci->i_fragtree_mutex);
452 ci->i_xattrs.blob = NULL;
453 ci->i_xattrs.prealloc_blob = NULL;
454 ci->i_xattrs.dirty = false;
455 ci->i_xattrs.index = RB_ROOT;
456 ci->i_xattrs.count = 0;
457 ci->i_xattrs.names_size = 0;
458 ci->i_xattrs.vals_size = 0;
459 ci->i_xattrs.version = 0;
460 ci->i_xattrs.index_version = 0;
462 ci->i_caps = RB_ROOT;
463 ci->i_auth_cap = NULL;
464 ci->i_dirty_caps = 0;
465 ci->i_flushing_caps = 0;
466 INIT_LIST_HEAD(&ci->i_dirty_item);
467 INIT_LIST_HEAD(&ci->i_flushing_item);
468 ci->i_prealloc_cap_flush = NULL;
469 INIT_LIST_HEAD(&ci->i_cap_flush_list);
470 init_waitqueue_head(&ci->i_cap_wq);
471 ci->i_hold_caps_min = 0;
472 ci->i_hold_caps_max = 0;
473 INIT_LIST_HEAD(&ci->i_cap_delay_list);
474 INIT_LIST_HEAD(&ci->i_cap_snaps);
475 ci->i_head_snapc = NULL;
478 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
479 ci->i_nr_by_mode[i] = 0;
481 mutex_init(&ci->i_truncate_mutex);
482 ci->i_truncate_seq = 0;
483 ci->i_truncate_size = 0;
484 ci->i_truncate_pending = 0;
487 ci->i_reported_size = 0;
488 ci->i_wanted_max_size = 0;
489 ci->i_requested_max_size = 0;
493 ci->i_rdcache_ref = 0;
496 ci->i_wrbuffer_ref = 0;
497 ci->i_wrbuffer_ref_head = 0;
498 ci->i_shared_gen = 0;
499 ci->i_rdcache_gen = 0;
500 ci->i_rdcache_revoking = 0;
502 INIT_LIST_HEAD(&ci->i_unsafe_writes);
503 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
504 INIT_LIST_HEAD(&ci->i_unsafe_iops);
505 spin_lock_init(&ci->i_unsafe_lock);
507 ci->i_snap_realm = NULL;
508 INIT_LIST_HEAD(&ci->i_snap_realm_item);
509 INIT_LIST_HEAD(&ci->i_snap_flush_item);
511 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
512 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
514 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
516 ceph_fscache_inode_init(ci);
518 return &ci->vfs_inode;
521 static void ceph_i_callback(struct rcu_head *head)
523 struct inode *inode = container_of(head, struct inode, i_rcu);
524 struct ceph_inode_info *ci = ceph_inode(inode);
526 kmem_cache_free(ceph_inode_cachep, ci);
529 void ceph_destroy_inode(struct inode *inode)
531 struct ceph_inode_info *ci = ceph_inode(inode);
532 struct ceph_inode_frag *frag;
535 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
537 ceph_fscache_unregister_inode_cookie(ci);
539 ceph_queue_caps_release(inode);
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_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
548 struct ceph_snap_realm *realm = ci->i_snap_realm;
550 dout(" dropping residual ref to snap realm %p\n", realm);
551 spin_lock(&realm->inodes_with_caps_lock);
552 list_del_init(&ci->i_snap_realm_item);
553 spin_unlock(&realm->inodes_with_caps_lock);
554 ceph_put_snap_realm(mdsc, realm);
557 kfree(ci->i_symlink);
558 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
559 frag = rb_entry(n, struct ceph_inode_frag, node);
560 rb_erase(n, &ci->i_fragtree);
563 ci->i_fragtree_nsplits = 0;
565 __ceph_destroy_xattrs(ci);
566 if (ci->i_xattrs.blob)
567 ceph_buffer_put(ci->i_xattrs.blob);
568 if (ci->i_xattrs.prealloc_blob)
569 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
571 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
573 call_rcu(&inode->i_rcu, ceph_i_callback);
576 int ceph_drop_inode(struct inode *inode)
579 * Positve dentry and corresponding inode are always accompanied
580 * in MDS reply. So no need to keep inode in the cache after
581 * dropping all its aliases.
586 void ceph_evict_inode(struct inode *inode)
588 /* wait unsafe sync writes */
589 ceph_sync_write_wait(inode);
590 truncate_inode_pages_final(&inode->i_data);
594 static inline blkcnt_t calc_inode_blocks(u64 size)
596 return (size + (1<<9) - 1) >> 9;
600 * Helpers to fill in size, ctime, mtime, and atime. We have to be
601 * careful because either the client or MDS may have more up to date
602 * info, depending on which capabilities are held, and whether
603 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
604 * and size are monotonically increasing, except when utimes() or
605 * truncate() increments the corresponding _seq values.)
607 int ceph_fill_file_size(struct inode *inode, int issued,
608 u32 truncate_seq, u64 truncate_size, u64 size)
610 struct ceph_inode_info *ci = ceph_inode(inode);
613 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
614 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
615 dout("size %lld -> %llu\n", inode->i_size, size);
616 if (size > 0 && S_ISDIR(inode->i_mode)) {
617 pr_err("fill_file_size non-zero size for directory\n");
620 i_size_write(inode, size);
621 inode->i_blocks = calc_inode_blocks(size);
622 ci->i_reported_size = size;
623 if (truncate_seq != ci->i_truncate_seq) {
624 dout("truncate_seq %u -> %u\n",
625 ci->i_truncate_seq, truncate_seq);
626 ci->i_truncate_seq = truncate_seq;
628 /* the MDS should have revoked these caps */
629 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
632 CEPH_CAP_FILE_LAZYIO));
634 * If we hold relevant caps, or in the case where we're
635 * not the only client referencing this file and we
636 * don't hold those caps, then we need to check whether
637 * the file is either opened or mmaped
639 if ((issued & (CEPH_CAP_FILE_CACHE|
640 CEPH_CAP_FILE_BUFFER)) ||
641 mapping_mapped(inode->i_mapping) ||
642 __ceph_caps_file_wanted(ci)) {
643 ci->i_truncate_pending++;
648 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
649 ci->i_truncate_size != truncate_size) {
650 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
652 ci->i_truncate_size = truncate_size;
656 ceph_fscache_invalidate(inode);
661 void ceph_fill_file_time(struct inode *inode, int issued,
662 u64 time_warp_seq, struct timespec *ctime,
663 struct timespec *mtime, struct timespec *atime)
665 struct ceph_inode_info *ci = ceph_inode(inode);
668 if (issued & (CEPH_CAP_FILE_EXCL|
670 CEPH_CAP_FILE_BUFFER|
672 CEPH_CAP_XATTR_EXCL)) {
673 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
674 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
675 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
676 ctime->tv_sec, ctime->tv_nsec);
677 inode->i_ctime = *ctime;
679 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
680 /* the MDS did a utimes() */
681 dout("mtime %ld.%09ld -> %ld.%09ld "
683 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
684 mtime->tv_sec, mtime->tv_nsec,
685 ci->i_time_warp_seq, (int)time_warp_seq);
687 inode->i_mtime = *mtime;
688 inode->i_atime = *atime;
689 ci->i_time_warp_seq = time_warp_seq;
690 } else if (time_warp_seq == ci->i_time_warp_seq) {
691 /* nobody did utimes(); take the max */
692 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
693 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
694 inode->i_mtime.tv_sec,
695 inode->i_mtime.tv_nsec,
696 mtime->tv_sec, mtime->tv_nsec);
697 inode->i_mtime = *mtime;
699 if (timespec_compare(atime, &inode->i_atime) > 0) {
700 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
701 inode->i_atime.tv_sec,
702 inode->i_atime.tv_nsec,
703 atime->tv_sec, atime->tv_nsec);
704 inode->i_atime = *atime;
706 } else if (issued & CEPH_CAP_FILE_EXCL) {
707 /* we did a utimes(); ignore mds values */
712 /* we have no write|excl caps; whatever the MDS says is true */
713 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
714 inode->i_ctime = *ctime;
715 inode->i_mtime = *mtime;
716 inode->i_atime = *atime;
717 ci->i_time_warp_seq = time_warp_seq;
722 if (warn) /* time_warp_seq shouldn't go backwards */
723 dout("%p mds time_warp_seq %llu < %u\n",
724 inode, time_warp_seq, ci->i_time_warp_seq);
728 * Populate an inode based on info from mds. May be called on new or
731 static int fill_inode(struct inode *inode, struct page *locked_page,
732 struct ceph_mds_reply_info_in *iinfo,
733 struct ceph_mds_reply_dirfrag *dirinfo,
734 struct ceph_mds_session *session,
735 unsigned long ttl_from, int cap_fmode,
736 struct ceph_cap_reservation *caps_reservation)
738 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
739 struct ceph_mds_reply_inode *info = iinfo->in;
740 struct ceph_inode_info *ci = ceph_inode(inode);
741 int issued = 0, implemented, new_issued;
742 struct timespec mtime, atime, ctime;
743 struct ceph_buffer *xattr_blob = NULL;
744 struct ceph_string *pool_ns = NULL;
745 struct ceph_cap *new_cap = NULL;
748 bool queue_trunc = false;
749 bool new_version = false;
750 bool fill_inline = false;
752 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
753 inode, ceph_vinop(inode), le64_to_cpu(info->version),
756 /* prealloc new cap struct */
757 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
758 new_cap = ceph_get_cap(mdsc, caps_reservation);
761 * prealloc xattr data, if it looks like we'll need it. only
762 * if len > 4 (meaning there are actually xattrs; the first 4
763 * bytes are the xattr count).
765 if (iinfo->xattr_len > 4) {
766 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
768 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
772 if (iinfo->pool_ns_len > 0)
773 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
776 spin_lock(&ci->i_ceph_lock);
779 * provided version will be odd if inode value is projected,
780 * even if stable. skip the update if we have newer stable
781 * info (ours>=theirs, e.g. due to racing mds replies), unless
782 * we are getting projected (unstable) info (in which case the
783 * version is odd, and we want ours>theirs).
789 if (ci->i_version == 0 ||
790 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
791 le64_to_cpu(info->version) > (ci->i_version & ~1)))
794 issued = __ceph_caps_issued(ci, &implemented);
795 issued |= implemented | __ceph_caps_dirty(ci);
796 new_issued = ~issued & le32_to_cpu(info->cap.caps);
799 ci->i_version = le64_to_cpu(info->version);
801 inode->i_rdev = le32_to_cpu(info->rdev);
802 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
804 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
805 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
806 inode->i_mode = le32_to_cpu(info->mode);
807 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
808 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
809 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
810 from_kuid(&init_user_ns, inode->i_uid),
811 from_kgid(&init_user_ns, inode->i_gid));
814 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
815 (issued & CEPH_CAP_LINK_EXCL) == 0)
816 set_nlink(inode, le32_to_cpu(info->nlink));
818 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
819 /* be careful with mtime, atime, size */
820 ceph_decode_timespec(&atime, &info->atime);
821 ceph_decode_timespec(&mtime, &info->mtime);
822 ceph_decode_timespec(&ctime, &info->ctime);
823 ceph_fill_file_time(inode, issued,
824 le32_to_cpu(info->time_warp_seq),
825 &ctime, &mtime, &atime);
829 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
830 s64 old_pool = ci->i_layout.pool_id;
831 struct ceph_string *old_ns;
833 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
834 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
835 lockdep_is_held(&ci->i_ceph_lock));
836 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
838 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
839 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
843 queue_trunc = ceph_fill_file_size(inode, issued,
844 le32_to_cpu(info->truncate_seq),
845 le64_to_cpu(info->truncate_size),
846 le64_to_cpu(info->size));
847 /* only update max_size on auth cap */
848 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
849 ci->i_max_size != le64_to_cpu(info->max_size)) {
850 dout("max_size %lld -> %llu\n", ci->i_max_size,
851 le64_to_cpu(info->max_size));
852 ci->i_max_size = le64_to_cpu(info->max_size);
857 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
858 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
859 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
860 if (ci->i_xattrs.blob)
861 ceph_buffer_put(ci->i_xattrs.blob);
862 ci->i_xattrs.blob = xattr_blob;
864 memcpy(ci->i_xattrs.blob->vec.iov_base,
865 iinfo->xattr_data, iinfo->xattr_len);
866 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
867 ceph_forget_all_cached_acls(inode);
871 inode->i_mapping->a_ops = &ceph_aops;
873 switch (inode->i_mode & S_IFMT) {
878 init_special_inode(inode, inode->i_mode, inode->i_rdev);
879 inode->i_op = &ceph_file_iops;
882 inode->i_op = &ceph_file_iops;
883 inode->i_fop = &ceph_file_fops;
886 inode->i_op = &ceph_symlink_iops;
887 if (!ci->i_symlink) {
888 u32 symlen = iinfo->symlink_len;
891 spin_unlock(&ci->i_ceph_lock);
893 if (symlen != i_size_read(inode)) {
894 pr_err("fill_inode %llx.%llx BAD symlink "
895 "size %lld\n", ceph_vinop(inode),
897 i_size_write(inode, symlen);
898 inode->i_blocks = calc_inode_blocks(symlen);
902 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
906 spin_lock(&ci->i_ceph_lock);
910 kfree(sym); /* lost a race */
912 inode->i_link = ci->i_symlink;
915 inode->i_op = &ceph_dir_iops;
916 inode->i_fop = &ceph_dir_fops;
918 ci->i_dir_layout = iinfo->dir_layout;
920 ci->i_files = le64_to_cpu(info->files);
921 ci->i_subdirs = le64_to_cpu(info->subdirs);
922 ci->i_rbytes = le64_to_cpu(info->rbytes);
923 ci->i_rfiles = le64_to_cpu(info->rfiles);
924 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
925 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
928 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
929 ceph_vinop(inode), inode->i_mode);
932 /* were we issued a capability? */
933 if (info->cap.caps) {
934 if (ceph_snap(inode) == CEPH_NOSNAP) {
935 unsigned caps = le32_to_cpu(info->cap.caps);
936 ceph_add_cap(inode, session,
937 le64_to_cpu(info->cap.cap_id),
939 le32_to_cpu(info->cap.wanted),
940 le32_to_cpu(info->cap.seq),
941 le32_to_cpu(info->cap.mseq),
942 le64_to_cpu(info->cap.realm),
943 info->cap.flags, &new_cap);
945 /* set dir completion flag? */
946 if (S_ISDIR(inode->i_mode) &&
947 ci->i_files == 0 && ci->i_subdirs == 0 &&
948 (caps & CEPH_CAP_FILE_SHARED) &&
949 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
950 !__ceph_dir_is_complete(ci)) {
951 dout(" marking %p complete (empty)\n", inode);
952 i_size_write(inode, 0);
953 __ceph_dir_set_complete(ci,
954 atomic64_read(&ci->i_release_count),
955 atomic64_read(&ci->i_ordered_count));
960 dout(" %p got snap_caps %s\n", inode,
961 ceph_cap_string(le32_to_cpu(info->cap.caps)));
962 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
964 __ceph_get_fmode(ci, cap_fmode);
966 } else if (cap_fmode >= 0) {
967 pr_warn("mds issued no caps on %llx.%llx\n",
969 __ceph_get_fmode(ci, cap_fmode);
972 if (iinfo->inline_version > 0 &&
973 iinfo->inline_version >= ci->i_inline_version) {
974 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
975 ci->i_inline_version = iinfo->inline_version;
976 if (ci->i_inline_version != CEPH_INLINE_NONE &&
978 (le32_to_cpu(info->cap.caps) & cache_caps)))
982 spin_unlock(&ci->i_ceph_lock);
985 ceph_fill_inline_data(inode, locked_page,
986 iinfo->inline_data, iinfo->inline_len);
989 wake_up_all(&ci->i_cap_wq);
991 /* queue truncate if we saw i_size decrease */
993 ceph_queue_vmtruncate(inode);
995 /* populate frag tree */
996 if (S_ISDIR(inode->i_mode))
997 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
999 /* update delegation info? */
1001 ceph_fill_dirfrag(inode, dirinfo);
1006 ceph_put_cap(mdsc, new_cap);
1008 ceph_buffer_put(xattr_blob);
1009 ceph_put_string(pool_ns);
1014 * caller should hold session s_mutex.
1016 static void update_dentry_lease(struct dentry *dentry,
1017 struct ceph_mds_reply_lease *lease,
1018 struct ceph_mds_session *session,
1019 unsigned long from_time)
1021 struct ceph_dentry_info *di = ceph_dentry(dentry);
1022 long unsigned duration = le32_to_cpu(lease->duration_ms);
1023 long unsigned ttl = from_time + (duration * HZ) / 1000;
1024 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1027 spin_lock(&dentry->d_lock);
1028 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1029 dentry, duration, ttl);
1031 /* make lease_rdcache_gen match directory */
1032 dir = d_inode(dentry->d_parent);
1034 /* only track leases on regular dentries */
1035 if (ceph_snap(dir) != CEPH_NOSNAP)
1038 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1043 if (di->lease_gen == session->s_cap_gen &&
1044 time_before(ttl, di->time))
1045 goto out_unlock; /* we already have a newer lease. */
1047 if (di->lease_session && di->lease_session != session)
1050 ceph_dentry_lru_touch(dentry);
1052 if (!di->lease_session)
1053 di->lease_session = ceph_get_mds_session(session);
1054 di->lease_gen = session->s_cap_gen;
1055 di->lease_seq = le32_to_cpu(lease->seq);
1056 di->lease_renew_after = half_ttl;
1057 di->lease_renew_from = 0;
1060 spin_unlock(&dentry->d_lock);
1065 * splice a dentry to an inode.
1066 * caller must hold directory i_mutex for this to be safe.
1068 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1070 struct dentry *realdn;
1072 BUG_ON(d_inode(dn));
1074 /* dn must be unhashed */
1075 if (!d_unhashed(dn))
1077 realdn = d_splice_alias(in, dn);
1078 if (IS_ERR(realdn)) {
1079 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1080 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1081 dn = realdn; /* note realdn contains the error */
1083 } else if (realdn) {
1084 dout("dn %p (%d) spliced with %p (%d) "
1085 "inode %p ino %llx.%llx\n",
1087 realdn, d_count(realdn),
1088 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1092 BUG_ON(!ceph_dentry(dn));
1093 dout("dn %p attached to %p ino %llx.%llx\n",
1094 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1101 * Incorporate results into the local cache. This is either just
1102 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1105 * A reply may contain
1106 * a directory inode along with a dentry.
1107 * and/or a target inode
1109 * Called with snap_rwsem (read).
1111 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1112 struct ceph_mds_session *session)
1114 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1115 struct inode *in = NULL;
1116 struct ceph_vino vino;
1117 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1120 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1121 rinfo->head->is_dentry, rinfo->head->is_target);
1127 * If we resend completed ops to a recovering mds, we get no
1128 * trace. Since that is very rare, pretend this is the case
1129 * to ensure the 'no trace' handlers in the callers behave.
1131 * Fill in inodes unconditionally to avoid breaking cap
1134 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1135 pr_info("fill_trace faking empty trace on %lld %s\n",
1136 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1137 if (rinfo->head->is_dentry) {
1138 rinfo->head->is_dentry = 0;
1139 err = fill_inode(req->r_locked_dir,
1140 &rinfo->diri, rinfo->dirfrag,
1141 session, req->r_request_started, -1);
1143 if (rinfo->head->is_target) {
1144 rinfo->head->is_target = 0;
1145 ininfo = rinfo->targeti.in;
1146 vino.ino = le64_to_cpu(ininfo->ino);
1147 vino.snap = le64_to_cpu(ininfo->snapid);
1148 in = ceph_get_inode(sb, vino);
1149 err = fill_inode(in, &rinfo->targeti, NULL,
1150 session, req->r_request_started,
1157 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1158 dout("fill_trace reply is empty!\n");
1159 if (rinfo->head->result == 0 && req->r_locked_dir)
1160 ceph_invalidate_dir_request(req);
1164 if (rinfo->head->is_dentry) {
1165 struct inode *dir = req->r_locked_dir;
1168 err = fill_inode(dir, NULL,
1169 &rinfo->diri, rinfo->dirfrag,
1170 session, req->r_request_started, -1,
1171 &req->r_caps_reservation);
1178 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1180 struct dentry *dn, *parent;
1182 BUG_ON(!rinfo->head->is_target);
1183 BUG_ON(req->r_dentry);
1185 parent = d_find_any_alias(dir);
1188 dname.name = rinfo->dname;
1189 dname.len = rinfo->dname_len;
1190 dname.hash = full_name_hash(parent, dname.name, dname.len);
1191 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1192 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1194 dn = d_lookup(parent, &dname);
1195 dout("d_lookup on parent=%p name=%.*s got %p\n",
1196 parent, dname.len, dname.name, dn);
1199 dn = d_alloc(parent, &dname);
1200 dout("d_alloc %p '%.*s' = %p\n", parent,
1201 dname.len, dname.name, dn);
1208 } else if (d_really_is_positive(dn) &&
1209 (ceph_ino(d_inode(dn)) != vino.ino ||
1210 ceph_snap(d_inode(dn)) != vino.snap)) {
1211 dout(" dn %p points to wrong inode %p\n",
1223 if (rinfo->head->is_target) {
1224 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1225 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1227 in = ceph_get_inode(sb, vino);
1232 req->r_target_inode = in;
1234 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1235 session, req->r_request_started,
1236 (!req->r_aborted && rinfo->head->result == 0) ?
1238 &req->r_caps_reservation);
1240 pr_err("fill_inode badness %p %llx.%llx\n",
1241 in, ceph_vinop(in));
1247 * ignore null lease/binding on snapdir ENOENT, or else we
1248 * will have trouble splicing in the virtual snapdir later
1250 if (rinfo->head->is_dentry && !req->r_aborted &&
1251 req->r_locked_dir &&
1252 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1253 fsc->mount_options->snapdir_name,
1254 req->r_dentry->d_name.len))) {
1256 * lookup link rename : null -> possibly existing inode
1257 * mknod symlink mkdir : null -> new inode
1258 * unlink : linked -> null
1260 struct inode *dir = req->r_locked_dir;
1261 struct dentry *dn = req->r_dentry;
1262 bool have_dir_cap, have_lease;
1266 BUG_ON(d_inode(dn->d_parent) != dir);
1267 BUG_ON(ceph_ino(dir) !=
1268 le64_to_cpu(rinfo->diri.in->ino));
1269 BUG_ON(ceph_snap(dir) !=
1270 le64_to_cpu(rinfo->diri.in->snapid));
1272 /* do we have a lease on the whole dir? */
1274 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1275 CEPH_CAP_FILE_SHARED);
1277 /* do we have a dn lease? */
1278 have_lease = have_dir_cap ||
1279 le32_to_cpu(rinfo->dlease->duration_ms);
1281 dout("fill_trace no dentry lease or dir cap\n");
1284 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1285 struct inode *olddir = req->r_old_dentry_dir;
1288 dout(" src %p '%pd' dst %p '%pd'\n",
1292 dout("fill_trace doing d_move %p -> %p\n",
1293 req->r_old_dentry, dn);
1295 /* d_move screws up sibling dentries' offsets */
1296 ceph_dir_clear_ordered(dir);
1297 ceph_dir_clear_ordered(olddir);
1299 d_move(req->r_old_dentry, dn);
1300 dout(" src %p '%pd' dst %p '%pd'\n",
1305 /* ensure target dentry is invalidated, despite
1306 rehashing bug in vfs_rename_dir */
1307 ceph_invalidate_dentry_lease(dn);
1309 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1310 ceph_dentry(req->r_old_dentry)->offset);
1312 dn = req->r_old_dentry; /* use old_dentry */
1316 if (!rinfo->head->is_target) {
1317 dout("fill_trace null dentry\n");
1318 if (d_really_is_positive(dn)) {
1319 ceph_dir_clear_ordered(dir);
1320 dout("d_delete %p\n", dn);
1323 if (have_lease && d_unhashed(dn))
1325 update_dentry_lease(dn, rinfo->dlease,
1327 req->r_request_started);
1332 /* attach proper inode */
1333 if (d_really_is_negative(dn)) {
1334 ceph_dir_clear_ordered(dir);
1336 dn = splice_dentry(dn, in);
1341 req->r_dentry = dn; /* may have spliced */
1342 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1343 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1344 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1351 update_dentry_lease(dn, rinfo->dlease, session,
1352 req->r_request_started);
1353 dout(" final dn %p\n", dn);
1354 } else if (!req->r_aborted &&
1355 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1356 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1357 struct dentry *dn = req->r_dentry;
1358 struct inode *dir = req->r_locked_dir;
1360 /* fill out a snapdir LOOKUPSNAP dentry */
1363 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1364 dout(" linking snapped dir %p to dn %p\n", in, dn);
1365 ceph_dir_clear_ordered(dir);
1367 dn = splice_dentry(dn, in);
1372 req->r_dentry = dn; /* may have spliced */
1375 dout("fill_trace done err=%d\n", err);
1380 * Prepopulate our cache with readdir results, leases, etc.
1382 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1383 struct ceph_mds_session *session)
1385 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1388 for (i = 0; i < rinfo->dir_nr; i++) {
1389 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1390 struct ceph_vino vino;
1394 vino.ino = le64_to_cpu(rde->inode.in->ino);
1395 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1397 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1400 dout("new_inode badness got %d\n", err);
1403 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1404 req->r_request_started, -1,
1405 &req->r_caps_reservation);
1407 pr_err("fill_inode badness on %p got %d\n", in, rc);
1416 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1420 put_page(ctl->page);
1425 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1426 struct ceph_readdir_cache_control *ctl,
1427 struct ceph_mds_request *req)
1429 struct ceph_inode_info *ci = ceph_inode(dir);
1430 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1431 unsigned idx = ctl->index % nsize;
1432 pgoff_t pgoff = ctl->index / nsize;
1434 if (!ctl->page || pgoff != page_index(ctl->page)) {
1435 ceph_readdir_cache_release(ctl);
1437 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1439 ctl->page = find_lock_page(&dir->i_data, pgoff);
1442 return idx == 0 ? -ENOMEM : 0;
1444 /* reading/filling the cache are serialized by
1445 * i_mutex, no need to use page lock */
1446 unlock_page(ctl->page);
1447 ctl->dentries = kmap(ctl->page);
1449 memset(ctl->dentries, 0, PAGE_SIZE);
1452 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1453 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1454 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1455 ctl->dentries[idx] = dn;
1458 dout("disable readdir cache\n");
1464 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1465 struct ceph_mds_session *session)
1467 struct dentry *parent = req->r_dentry;
1468 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1469 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1473 int err = 0, skipped = 0, ret, i;
1474 struct inode *snapdir = NULL;
1475 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1476 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1479 struct ceph_readdir_cache_control cache_ctl = {};
1482 return readdir_prepopulate_inodes_only(req, session);
1484 if (rinfo->hash_order && req->r_path2) {
1485 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1486 req->r_path2, strlen(req->r_path2));
1487 last_hash = ceph_frag_value(last_hash);
1490 if (rinfo->dir_dir &&
1491 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1492 dout("readdir_prepopulate got new frag %x -> %x\n",
1493 frag, le32_to_cpu(rinfo->dir_dir->frag));
1494 frag = le32_to_cpu(rinfo->dir_dir->frag);
1495 if (!rinfo->hash_order)
1496 req->r_readdir_offset = 2;
1499 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1500 snapdir = ceph_get_snapdir(d_inode(parent));
1501 parent = d_find_alias(snapdir);
1502 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1503 rinfo->dir_nr, parent);
1505 dout("readdir_prepopulate %d items under dn %p\n",
1506 rinfo->dir_nr, parent);
1508 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1511 if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2 &&
1512 !(rinfo->hash_order && req->r_path2)) {
1513 /* note dir version at start of readdir so we can tell
1514 * if any dentries get dropped */
1515 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1516 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1517 req->r_readdir_cache_idx = 0;
1520 cache_ctl.index = req->r_readdir_cache_idx;
1521 fpos_offset = req->r_readdir_offset;
1523 /* FIXME: release caps/leases if error occurs */
1524 for (i = 0; i < rinfo->dir_nr; i++) {
1525 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1526 struct ceph_vino vino;
1528 dname.name = rde->name;
1529 dname.len = rde->name_len;
1530 dname.hash = full_name_hash(parent, dname.name, dname.len);
1532 vino.ino = le64_to_cpu(rde->inode.in->ino);
1533 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1535 if (rinfo->hash_order) {
1536 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1537 rde->name, rde->name_len);
1538 hash = ceph_frag_value(hash);
1539 if (hash != last_hash)
1542 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1544 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1548 dn = d_lookup(parent, &dname);
1549 dout("d_lookup on parent=%p name=%.*s got %p\n",
1550 parent, dname.len, dname.name, dn);
1553 dn = d_alloc(parent, &dname);
1554 dout("d_alloc %p '%.*s' = %p\n", parent,
1555 dname.len, dname.name, dn);
1557 dout("d_alloc badness\n");
1561 } else if (d_really_is_positive(dn) &&
1562 (ceph_ino(d_inode(dn)) != vino.ino ||
1563 ceph_snap(d_inode(dn)) != vino.snap)) {
1564 dout(" dn %p points to wrong inode %p\n",
1572 if (d_really_is_positive(dn)) {
1575 in = ceph_get_inode(parent->d_sb, vino);
1577 dout("new_inode badness\n");
1585 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1586 req->r_request_started, -1,
1587 &req->r_caps_reservation);
1589 pr_err("fill_inode badness on %p\n", in);
1590 if (d_really_is_negative(dn))
1597 if (d_really_is_negative(dn)) {
1598 struct dentry *realdn;
1600 if (ceph_security_xattr_deadlock(in)) {
1601 dout(" skip splicing dn %p to inode %p"
1602 " (security xattr deadlock)\n", dn, in);
1608 realdn = splice_dentry(dn, in);
1609 if (IS_ERR(realdn)) {
1610 err = PTR_ERR(realdn);
1618 ceph_dentry(dn)->offset = rde->offset;
1620 update_dentry_lease(dn, rde->lease, req->r_session,
1621 req->r_request_started);
1623 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1624 ret = fill_readdir_cache(d_inode(parent), dn,
1634 if (err == 0 && skipped == 0) {
1635 req->r_did_prepopulate = true;
1636 req->r_readdir_cache_idx = cache_ctl.index;
1638 ceph_readdir_cache_release(&cache_ctl);
1643 dout("readdir_prepopulate done\n");
1647 int ceph_inode_set_size(struct inode *inode, loff_t size)
1649 struct ceph_inode_info *ci = ceph_inode(inode);
1652 spin_lock(&ci->i_ceph_lock);
1653 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1654 i_size_write(inode, size);
1655 inode->i_blocks = calc_inode_blocks(size);
1657 /* tell the MDS if we are approaching max_size */
1658 if ((size << 1) >= ci->i_max_size &&
1659 (ci->i_reported_size << 1) < ci->i_max_size)
1662 spin_unlock(&ci->i_ceph_lock);
1667 * Write back inode data in a worker thread. (This can't be done
1668 * in the message handler context.)
1670 void ceph_queue_writeback(struct inode *inode)
1673 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1674 &ceph_inode(inode)->i_wb_work)) {
1675 dout("ceph_queue_writeback %p\n", inode);
1677 dout("ceph_queue_writeback %p failed\n", inode);
1682 static void ceph_writeback_work(struct work_struct *work)
1684 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1686 struct inode *inode = &ci->vfs_inode;
1688 dout("writeback %p\n", inode);
1689 filemap_fdatawrite(&inode->i_data);
1694 * queue an async invalidation
1696 void ceph_queue_invalidate(struct inode *inode)
1699 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1700 &ceph_inode(inode)->i_pg_inv_work)) {
1701 dout("ceph_queue_invalidate %p\n", inode);
1703 dout("ceph_queue_invalidate %p failed\n", inode);
1709 * Invalidate inode pages in a worker thread. (This can't be done
1710 * in the message handler context.)
1712 static void ceph_invalidate_work(struct work_struct *work)
1714 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1716 struct inode *inode = &ci->vfs_inode;
1717 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1721 mutex_lock(&ci->i_truncate_mutex);
1723 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1724 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1725 inode, ceph_ino(inode));
1726 mapping_set_error(inode->i_mapping, -EIO);
1727 truncate_pagecache(inode, 0);
1728 mutex_unlock(&ci->i_truncate_mutex);
1732 spin_lock(&ci->i_ceph_lock);
1733 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1734 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1735 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1736 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1738 spin_unlock(&ci->i_ceph_lock);
1739 mutex_unlock(&ci->i_truncate_mutex);
1742 orig_gen = ci->i_rdcache_gen;
1743 spin_unlock(&ci->i_ceph_lock);
1745 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1746 pr_err("invalidate_pages %p fails\n", inode);
1749 spin_lock(&ci->i_ceph_lock);
1750 if (orig_gen == ci->i_rdcache_gen &&
1751 orig_gen == ci->i_rdcache_revoking) {
1752 dout("invalidate_pages %p gen %d successful\n", inode,
1754 ci->i_rdcache_revoking--;
1757 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1758 inode, orig_gen, ci->i_rdcache_gen,
1759 ci->i_rdcache_revoking);
1760 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1763 spin_unlock(&ci->i_ceph_lock);
1764 mutex_unlock(&ci->i_truncate_mutex);
1767 ceph_check_caps(ci, 0, NULL);
1773 * called by trunc_wq;
1775 * We also truncate in a separate thread as well.
1777 static void ceph_vmtruncate_work(struct work_struct *work)
1779 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1781 struct inode *inode = &ci->vfs_inode;
1783 dout("vmtruncate_work %p\n", inode);
1784 __ceph_do_pending_vmtruncate(inode);
1789 * Queue an async vmtruncate. If we fail to queue work, we will handle
1790 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1792 void ceph_queue_vmtruncate(struct inode *inode)
1794 struct ceph_inode_info *ci = ceph_inode(inode);
1798 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1799 &ci->i_vmtruncate_work)) {
1800 dout("ceph_queue_vmtruncate %p\n", inode);
1802 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1803 inode, ci->i_truncate_pending);
1809 * Make sure any pending truncation is applied before doing anything
1810 * that may depend on it.
1812 void __ceph_do_pending_vmtruncate(struct inode *inode)
1814 struct ceph_inode_info *ci = ceph_inode(inode);
1816 int wrbuffer_refs, finish = 0;
1818 mutex_lock(&ci->i_truncate_mutex);
1820 spin_lock(&ci->i_ceph_lock);
1821 if (ci->i_truncate_pending == 0) {
1822 dout("__do_pending_vmtruncate %p none pending\n", inode);
1823 spin_unlock(&ci->i_ceph_lock);
1824 mutex_unlock(&ci->i_truncate_mutex);
1829 * make sure any dirty snapped pages are flushed before we
1830 * possibly truncate them.. so write AND block!
1832 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1833 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1835 spin_unlock(&ci->i_ceph_lock);
1836 filemap_write_and_wait_range(&inode->i_data, 0,
1837 inode->i_sb->s_maxbytes);
1841 /* there should be no reader or writer */
1842 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1844 to = ci->i_truncate_size;
1845 wrbuffer_refs = ci->i_wrbuffer_ref;
1846 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1847 ci->i_truncate_pending, to);
1848 spin_unlock(&ci->i_ceph_lock);
1850 truncate_pagecache(inode, to);
1852 spin_lock(&ci->i_ceph_lock);
1853 if (to == ci->i_truncate_size) {
1854 ci->i_truncate_pending = 0;
1857 spin_unlock(&ci->i_ceph_lock);
1861 mutex_unlock(&ci->i_truncate_mutex);
1863 if (wrbuffer_refs == 0)
1864 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1866 wake_up_all(&ci->i_cap_wq);
1872 static const struct inode_operations ceph_symlink_iops = {
1873 .get_link = simple_get_link,
1874 .setattr = ceph_setattr,
1875 .getattr = ceph_getattr,
1876 .listxattr = ceph_listxattr,
1879 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1881 struct ceph_inode_info *ci = ceph_inode(inode);
1882 const unsigned int ia_valid = attr->ia_valid;
1883 struct ceph_mds_request *req;
1884 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1885 struct ceph_cap_flush *prealloc_cf;
1887 int release = 0, dirtied = 0;
1890 int inode_dirty_flags = 0;
1891 bool lock_snap_rwsem = false;
1893 prealloc_cf = ceph_alloc_cap_flush();
1897 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1900 ceph_free_cap_flush(prealloc_cf);
1901 return PTR_ERR(req);
1904 spin_lock(&ci->i_ceph_lock);
1905 issued = __ceph_caps_issued(ci, NULL);
1907 if (!ci->i_head_snapc &&
1908 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1909 lock_snap_rwsem = true;
1910 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1911 spin_unlock(&ci->i_ceph_lock);
1912 down_read(&mdsc->snap_rwsem);
1913 spin_lock(&ci->i_ceph_lock);
1914 issued = __ceph_caps_issued(ci, NULL);
1918 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1920 if (ia_valid & ATTR_UID) {
1921 dout("setattr %p uid %d -> %d\n", inode,
1922 from_kuid(&init_user_ns, inode->i_uid),
1923 from_kuid(&init_user_ns, attr->ia_uid));
1924 if (issued & CEPH_CAP_AUTH_EXCL) {
1925 inode->i_uid = attr->ia_uid;
1926 dirtied |= CEPH_CAP_AUTH_EXCL;
1927 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1928 !uid_eq(attr->ia_uid, inode->i_uid)) {
1929 req->r_args.setattr.uid = cpu_to_le32(
1930 from_kuid(&init_user_ns, attr->ia_uid));
1931 mask |= CEPH_SETATTR_UID;
1932 release |= CEPH_CAP_AUTH_SHARED;
1935 if (ia_valid & ATTR_GID) {
1936 dout("setattr %p gid %d -> %d\n", inode,
1937 from_kgid(&init_user_ns, inode->i_gid),
1938 from_kgid(&init_user_ns, attr->ia_gid));
1939 if (issued & CEPH_CAP_AUTH_EXCL) {
1940 inode->i_gid = attr->ia_gid;
1941 dirtied |= CEPH_CAP_AUTH_EXCL;
1942 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1943 !gid_eq(attr->ia_gid, inode->i_gid)) {
1944 req->r_args.setattr.gid = cpu_to_le32(
1945 from_kgid(&init_user_ns, attr->ia_gid));
1946 mask |= CEPH_SETATTR_GID;
1947 release |= CEPH_CAP_AUTH_SHARED;
1950 if (ia_valid & ATTR_MODE) {
1951 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1953 if (issued & CEPH_CAP_AUTH_EXCL) {
1954 inode->i_mode = attr->ia_mode;
1955 dirtied |= CEPH_CAP_AUTH_EXCL;
1956 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1957 attr->ia_mode != inode->i_mode) {
1958 inode->i_mode = attr->ia_mode;
1959 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1960 mask |= CEPH_SETATTR_MODE;
1961 release |= CEPH_CAP_AUTH_SHARED;
1965 if (ia_valid & ATTR_ATIME) {
1966 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1967 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1968 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1969 if (issued & CEPH_CAP_FILE_EXCL) {
1970 ci->i_time_warp_seq++;
1971 inode->i_atime = attr->ia_atime;
1972 dirtied |= CEPH_CAP_FILE_EXCL;
1973 } else if ((issued & CEPH_CAP_FILE_WR) &&
1974 timespec_compare(&inode->i_atime,
1975 &attr->ia_atime) < 0) {
1976 inode->i_atime = attr->ia_atime;
1977 dirtied |= CEPH_CAP_FILE_WR;
1978 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1979 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1980 ceph_encode_timespec(&req->r_args.setattr.atime,
1982 mask |= CEPH_SETATTR_ATIME;
1983 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1987 if (ia_valid & ATTR_MTIME) {
1988 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1989 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1990 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1991 if (issued & CEPH_CAP_FILE_EXCL) {
1992 ci->i_time_warp_seq++;
1993 inode->i_mtime = attr->ia_mtime;
1994 dirtied |= CEPH_CAP_FILE_EXCL;
1995 } else if ((issued & CEPH_CAP_FILE_WR) &&
1996 timespec_compare(&inode->i_mtime,
1997 &attr->ia_mtime) < 0) {
1998 inode->i_mtime = attr->ia_mtime;
1999 dirtied |= CEPH_CAP_FILE_WR;
2000 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2001 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2002 ceph_encode_timespec(&req->r_args.setattr.mtime,
2004 mask |= CEPH_SETATTR_MTIME;
2005 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2009 if (ia_valid & ATTR_SIZE) {
2010 dout("setattr %p size %lld -> %lld\n", inode,
2011 inode->i_size, attr->ia_size);
2012 if ((issued & CEPH_CAP_FILE_EXCL) &&
2013 attr->ia_size > inode->i_size) {
2014 i_size_write(inode, attr->ia_size);
2015 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2016 inode->i_ctime = attr->ia_ctime;
2017 ci->i_reported_size = attr->ia_size;
2018 dirtied |= CEPH_CAP_FILE_EXCL;
2019 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2020 attr->ia_size != inode->i_size) {
2021 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2022 req->r_args.setattr.old_size =
2023 cpu_to_le64(inode->i_size);
2024 mask |= CEPH_SETATTR_SIZE;
2025 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2030 /* these do nothing */
2031 if (ia_valid & ATTR_CTIME) {
2032 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2033 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2034 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2035 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2036 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2037 only ? "ctime only" : "ignored");
2038 inode->i_ctime = attr->ia_ctime;
2041 * if kernel wants to dirty ctime but nothing else,
2042 * we need to choose a cap to dirty under, or do
2043 * a almost-no-op setattr
2045 if (issued & CEPH_CAP_AUTH_EXCL)
2046 dirtied |= CEPH_CAP_AUTH_EXCL;
2047 else if (issued & CEPH_CAP_FILE_EXCL)
2048 dirtied |= CEPH_CAP_FILE_EXCL;
2049 else if (issued & CEPH_CAP_XATTR_EXCL)
2050 dirtied |= CEPH_CAP_XATTR_EXCL;
2052 mask |= CEPH_SETATTR_CTIME;
2055 if (ia_valid & ATTR_FILE)
2056 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2059 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2061 inode->i_ctime = current_time(inode);
2065 spin_unlock(&ci->i_ceph_lock);
2066 if (lock_snap_rwsem)
2067 up_read(&mdsc->snap_rwsem);
2069 if (inode_dirty_flags)
2070 __mark_inode_dirty(inode, inode_dirty_flags);
2072 if (ia_valid & ATTR_MODE) {
2073 err = posix_acl_chmod(inode, attr->ia_mode);
2079 req->r_inode = inode;
2081 req->r_inode_drop = release;
2082 req->r_args.setattr.mask = cpu_to_le32(mask);
2083 req->r_num_caps = 1;
2084 err = ceph_mdsc_do_request(mdsc, NULL, req);
2086 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2087 ceph_cap_string(dirtied), mask);
2089 ceph_mdsc_put_request(req);
2090 if (mask & CEPH_SETATTR_SIZE)
2091 __ceph_do_pending_vmtruncate(inode);
2092 ceph_free_cap_flush(prealloc_cf);
2095 ceph_mdsc_put_request(req);
2096 ceph_free_cap_flush(prealloc_cf);
2103 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2105 struct inode *inode = d_inode(dentry);
2108 if (ceph_snap(inode) != CEPH_NOSNAP)
2111 err = setattr_prepare(dentry, attr);
2115 return __ceph_setattr(inode, attr);
2119 * Verify that we have a lease on the given mask. If not,
2120 * do a getattr against an mds.
2122 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2123 int mask, bool force)
2125 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2126 struct ceph_mds_client *mdsc = fsc->mdsc;
2127 struct ceph_mds_request *req;
2130 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2131 dout("do_getattr inode %p SNAPDIR\n", inode);
2135 dout("do_getattr inode %p mask %s mode 0%o\n",
2136 inode, ceph_cap_string(mask), inode->i_mode);
2137 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2140 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2142 return PTR_ERR(req);
2143 req->r_inode = inode;
2145 req->r_num_caps = 1;
2146 req->r_args.getattr.mask = cpu_to_le32(mask);
2147 req->r_locked_page = locked_page;
2148 err = ceph_mdsc_do_request(mdsc, NULL, req);
2149 if (locked_page && err == 0) {
2150 u64 inline_version = req->r_reply_info.targeti.inline_version;
2151 if (inline_version == 0) {
2152 /* the reply is supposed to contain inline data */
2154 } else if (inline_version == CEPH_INLINE_NONE) {
2157 err = req->r_reply_info.targeti.inline_len;
2160 ceph_mdsc_put_request(req);
2161 dout("do_getattr result=%d\n", err);
2167 * Check inode permissions. We verify we have a valid value for
2168 * the AUTH cap, then call the generic handler.
2170 int ceph_permission(struct inode *inode, int mask)
2174 if (mask & MAY_NOT_BLOCK)
2177 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2180 err = generic_permission(inode, mask);
2185 * Get all attributes. Hopefully somedata we'll have a statlite()
2186 * and can limit the fields we require to be accurate.
2188 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2191 struct inode *inode = d_inode(dentry);
2192 struct ceph_inode_info *ci = ceph_inode(inode);
2195 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2197 generic_fillattr(inode, stat);
2198 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2199 if (ceph_snap(inode) != CEPH_NOSNAP)
2200 stat->dev = ceph_snap(inode);
2203 if (S_ISDIR(inode->i_mode)) {
2204 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2206 stat->size = ci->i_rbytes;
2208 stat->size = ci->i_files + ci->i_subdirs;
2210 stat->blksize = 65536;