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/posix_acl.h>
12 #include <linux/random.h>
13 #include <linux/sort.h>
16 #include "mds_client.h"
18 #include <linux/ceph/decode.h>
21 * Ceph inode operations
23 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
24 * setattr, etc.), xattr helpers, and helpers for assimilating
25 * metadata returned by the MDS into our cache.
27 * Also define helpers for doing asynchronous writeback, invalidation,
28 * and truncation for the benefit of those who can't afford to block
29 * (typically because they are in the message handler path).
32 static const struct inode_operations ceph_symlink_iops;
34 static void ceph_invalidate_work(struct work_struct *work);
35 static void ceph_writeback_work(struct work_struct *work);
36 static void ceph_vmtruncate_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 .setxattr = ceph_setxattr,
97 .getxattr = ceph_getxattr,
98 .listxattr = ceph_listxattr,
99 .removexattr = ceph_removexattr,
100 .get_acl = ceph_get_acl,
101 .set_acl = ceph_set_acl,
106 * We use a 'frag tree' to keep track of the MDS's directory fragments
107 * for a given inode (usually there is just a single fragment). We
108 * need to know when a child frag is delegated to a new MDS, or when
109 * it is flagged as replicated, so we can direct our requests
114 * find/create a frag in the tree
116 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
120 struct rb_node *parent = NULL;
121 struct ceph_inode_frag *frag;
124 p = &ci->i_fragtree.rb_node;
127 frag = rb_entry(parent, struct ceph_inode_frag, node);
128 c = ceph_frag_compare(f, frag->frag);
137 frag = kmalloc(sizeof(*frag), GFP_NOFS);
139 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
140 "frag %x\n", &ci->vfs_inode,
141 ceph_vinop(&ci->vfs_inode), f);
142 return ERR_PTR(-ENOMEM);
149 rb_link_node(&frag->node, parent, p);
150 rb_insert_color(&frag->node, &ci->i_fragtree);
152 dout("get_or_create_frag added %llx.%llx frag %x\n",
153 ceph_vinop(&ci->vfs_inode), f);
158 * find a specific frag @f
160 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
162 struct rb_node *n = ci->i_fragtree.rb_node;
165 struct ceph_inode_frag *frag =
166 rb_entry(n, struct ceph_inode_frag, node);
167 int c = ceph_frag_compare(f, frag->frag);
179 * Choose frag containing the given value @v. If @pfrag is
180 * specified, copy the frag delegation info to the caller if
183 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
184 struct ceph_inode_frag *pfrag, int *found)
186 u32 t = ceph_frag_make(0, 0);
187 struct ceph_inode_frag *frag;
195 WARN_ON(!ceph_frag_contains_value(t, v));
196 frag = __ceph_find_frag(ci, t);
198 break; /* t is a leaf */
199 if (frag->split_by == 0) {
201 memcpy(pfrag, frag, sizeof(*pfrag));
208 nway = 1 << frag->split_by;
209 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
210 frag->split_by, nway);
211 for (i = 0; i < nway; i++) {
212 n = ceph_frag_make_child(t, frag->split_by, i);
213 if (ceph_frag_contains_value(n, v)) {
220 dout("choose_frag(%x) = %x\n", v, t);
225 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
226 struct ceph_inode_frag *pfrag, int *found)
229 mutex_lock(&ci->i_fragtree_mutex);
230 ret = __ceph_choose_frag(ci, v, pfrag, found);
231 mutex_unlock(&ci->i_fragtree_mutex);
236 * Process dirfrag (delegation) info from the mds. Include leaf
237 * fragment in tree ONLY if ndist > 0. Otherwise, only
238 * branches/splits are included in i_fragtree)
240 static int ceph_fill_dirfrag(struct inode *inode,
241 struct ceph_mds_reply_dirfrag *dirinfo)
243 struct ceph_inode_info *ci = ceph_inode(inode);
244 struct ceph_inode_frag *frag;
245 u32 id = le32_to_cpu(dirinfo->frag);
246 int mds = le32_to_cpu(dirinfo->auth);
247 int ndist = le32_to_cpu(dirinfo->ndist);
252 spin_lock(&ci->i_ceph_lock);
254 diri_auth = ci->i_auth_cap->mds;
255 spin_unlock(&ci->i_ceph_lock);
257 if (mds == -1) /* CDIR_AUTH_PARENT */
260 mutex_lock(&ci->i_fragtree_mutex);
261 if (ndist == 0 && mds == diri_auth) {
262 /* no delegation info needed. */
263 frag = __ceph_find_frag(ci, id);
266 if (frag->split_by == 0) {
267 /* tree leaf, remove */
268 dout("fill_dirfrag removed %llx.%llx frag %x"
269 " (no ref)\n", ceph_vinop(inode), id);
270 rb_erase(&frag->node, &ci->i_fragtree);
273 /* tree branch, keep and clear */
274 dout("fill_dirfrag cleared %llx.%llx frag %x"
275 " referral\n", ceph_vinop(inode), id);
283 /* find/add this frag to store mds delegation info */
284 frag = __get_or_create_frag(ci, id);
286 /* this is not the end of the world; we can continue
287 with bad/inaccurate delegation info */
288 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
289 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
295 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
296 for (i = 0; i < frag->ndist; i++)
297 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
298 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
299 ceph_vinop(inode), frag->frag, frag->ndist);
302 mutex_unlock(&ci->i_fragtree_mutex);
306 static int frag_tree_split_cmp(const void *l, const void *r)
308 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
309 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
310 return ceph_frag_compare(ls->frag, rs->frag);
313 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
316 return f == ceph_frag_make(0, 0);
317 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
319 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
322 static int ceph_fill_fragtree(struct inode *inode,
323 struct ceph_frag_tree_head *fragtree,
324 struct ceph_mds_reply_dirfrag *dirinfo)
326 struct ceph_inode_info *ci = ceph_inode(inode);
327 struct ceph_inode_frag *frag, *prev_frag = NULL;
328 struct rb_node *rb_node;
329 unsigned i, split_by, nsplits;
333 mutex_lock(&ci->i_fragtree_mutex);
334 nsplits = le32_to_cpu(fragtree->nsplits);
335 if (nsplits != ci->i_fragtree_nsplits) {
337 } else if (nsplits) {
338 i = prandom_u32() % nsplits;
339 id = le32_to_cpu(fragtree->splits[i].frag);
340 if (!__ceph_find_frag(ci, id))
342 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
343 rb_node = rb_first(&ci->i_fragtree);
344 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
345 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
348 if (!update && dirinfo) {
349 id = le32_to_cpu(dirinfo->frag);
350 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
357 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
358 frag_tree_split_cmp, NULL);
361 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
362 rb_node = rb_first(&ci->i_fragtree);
363 for (i = 0; i < nsplits; i++) {
364 id = le32_to_cpu(fragtree->splits[i].frag);
365 split_by = le32_to_cpu(fragtree->splits[i].by);
366 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
367 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
368 "frag %x split by %d\n", ceph_vinop(inode),
369 i, nsplits, id, split_by);
374 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
375 if (ceph_frag_compare(frag->frag, id) >= 0) {
376 if (frag->frag != id)
379 rb_node = rb_next(rb_node);
382 rb_node = rb_next(rb_node);
383 /* delete stale split/leaf node */
384 if (frag->split_by > 0 ||
385 !is_frag_child(frag->frag, prev_frag)) {
386 rb_erase(&frag->node, &ci->i_fragtree);
387 if (frag->split_by > 0)
388 ci->i_fragtree_nsplits--;
394 frag = __get_or_create_frag(ci, id);
398 if (frag->split_by == 0)
399 ci->i_fragtree_nsplits++;
400 frag->split_by = split_by;
401 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
405 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
406 rb_node = rb_next(rb_node);
407 /* delete stale split/leaf node */
408 if (frag->split_by > 0 ||
409 !is_frag_child(frag->frag, prev_frag)) {
410 rb_erase(&frag->node, &ci->i_fragtree);
411 if (frag->split_by > 0)
412 ci->i_fragtree_nsplits--;
417 mutex_unlock(&ci->i_fragtree_mutex);
422 * initialize a newly allocated inode.
424 struct inode *ceph_alloc_inode(struct super_block *sb)
426 struct ceph_inode_info *ci;
429 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
433 dout("alloc_inode %p\n", &ci->vfs_inode);
435 spin_lock_init(&ci->i_ceph_lock);
438 ci->i_inline_version = 0;
439 ci->i_time_warp_seq = 0;
440 ci->i_ceph_flags = 0;
441 atomic64_set(&ci->i_ordered_count, 1);
442 atomic64_set(&ci->i_release_count, 1);
443 atomic64_set(&ci->i_complete_seq[0], 0);
444 atomic64_set(&ci->i_complete_seq[1], 0);
445 ci->i_symlink = NULL;
447 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
448 ci->i_pool_ns_len = 0;
450 ci->i_fragtree = RB_ROOT;
451 mutex_init(&ci->i_fragtree_mutex);
453 ci->i_xattrs.blob = NULL;
454 ci->i_xattrs.prealloc_blob = NULL;
455 ci->i_xattrs.dirty = false;
456 ci->i_xattrs.index = RB_ROOT;
457 ci->i_xattrs.count = 0;
458 ci->i_xattrs.names_size = 0;
459 ci->i_xattrs.vals_size = 0;
460 ci->i_xattrs.version = 0;
461 ci->i_xattrs.index_version = 0;
463 ci->i_caps = RB_ROOT;
464 ci->i_auth_cap = NULL;
465 ci->i_dirty_caps = 0;
466 ci->i_flushing_caps = 0;
467 INIT_LIST_HEAD(&ci->i_dirty_item);
468 INIT_LIST_HEAD(&ci->i_flushing_item);
469 ci->i_prealloc_cap_flush = NULL;
470 ci->i_cap_flush_tree = RB_ROOT;
471 init_waitqueue_head(&ci->i_cap_wq);
472 ci->i_hold_caps_min = 0;
473 ci->i_hold_caps_max = 0;
474 INIT_LIST_HEAD(&ci->i_cap_delay_list);
475 INIT_LIST_HEAD(&ci->i_cap_snaps);
476 ci->i_head_snapc = NULL;
479 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
480 ci->i_nr_by_mode[i] = 0;
482 mutex_init(&ci->i_truncate_mutex);
483 ci->i_truncate_seq = 0;
484 ci->i_truncate_size = 0;
485 ci->i_truncate_pending = 0;
488 ci->i_reported_size = 0;
489 ci->i_wanted_max_size = 0;
490 ci->i_requested_max_size = 0;
494 ci->i_rdcache_ref = 0;
497 ci->i_wrbuffer_ref = 0;
498 ci->i_wrbuffer_ref_head = 0;
499 ci->i_shared_gen = 0;
500 ci->i_rdcache_gen = 0;
501 ci->i_rdcache_revoking = 0;
503 INIT_LIST_HEAD(&ci->i_unsafe_writes);
504 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
505 INIT_LIST_HEAD(&ci->i_unsafe_iops);
506 spin_lock_init(&ci->i_unsafe_lock);
508 ci->i_snap_realm = NULL;
509 INIT_LIST_HEAD(&ci->i_snap_realm_item);
510 INIT_LIST_HEAD(&ci->i_snap_flush_item);
512 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
513 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
515 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
517 ceph_fscache_inode_init(ci);
519 return &ci->vfs_inode;
522 static void ceph_i_callback(struct rcu_head *head)
524 struct inode *inode = container_of(head, struct inode, i_rcu);
525 struct ceph_inode_info *ci = ceph_inode(inode);
527 kmem_cache_free(ceph_inode_cachep, ci);
530 void ceph_destroy_inode(struct inode *inode)
532 struct ceph_inode_info *ci = ceph_inode(inode);
533 struct ceph_inode_frag *frag;
536 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
538 ceph_fscache_unregister_inode_cookie(ci);
540 ceph_queue_caps_release(inode);
543 * we may still have a snap_realm reference if there are stray
544 * caps in i_snap_caps.
546 if (ci->i_snap_realm) {
547 struct ceph_mds_client *mdsc =
548 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
549 struct ceph_snap_realm *realm = ci->i_snap_realm;
551 dout(" dropping residual ref to snap realm %p\n", realm);
552 spin_lock(&realm->inodes_with_caps_lock);
553 list_del_init(&ci->i_snap_realm_item);
554 spin_unlock(&realm->inodes_with_caps_lock);
555 ceph_put_snap_realm(mdsc, realm);
558 kfree(ci->i_symlink);
559 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
560 frag = rb_entry(n, struct ceph_inode_frag, node);
561 rb_erase(n, &ci->i_fragtree);
564 ci->i_fragtree_nsplits = 0;
566 __ceph_destroy_xattrs(ci);
567 if (ci->i_xattrs.blob)
568 ceph_buffer_put(ci->i_xattrs.blob);
569 if (ci->i_xattrs.prealloc_blob)
570 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
572 call_rcu(&inode->i_rcu, ceph_i_callback);
575 int ceph_drop_inode(struct inode *inode)
578 * Positve dentry and corresponding inode are always accompanied
579 * in MDS reply. So no need to keep inode in the cache after
580 * dropping all its aliases.
585 static inline blkcnt_t calc_inode_blocks(u64 size)
587 return (size + (1<<9) - 1) >> 9;
591 * Helpers to fill in size, ctime, mtime, and atime. We have to be
592 * careful because either the client or MDS may have more up to date
593 * info, depending on which capabilities are held, and whether
594 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
595 * and size are monotonically increasing, except when utimes() or
596 * truncate() increments the corresponding _seq values.)
598 int ceph_fill_file_size(struct inode *inode, int issued,
599 u32 truncate_seq, u64 truncate_size, u64 size)
601 struct ceph_inode_info *ci = ceph_inode(inode);
604 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
605 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
606 dout("size %lld -> %llu\n", inode->i_size, size);
607 if (size > 0 && S_ISDIR(inode->i_mode)) {
608 pr_err("fill_file_size non-zero size for directory\n");
611 i_size_write(inode, size);
612 inode->i_blocks = calc_inode_blocks(size);
613 ci->i_reported_size = size;
614 if (truncate_seq != ci->i_truncate_seq) {
615 dout("truncate_seq %u -> %u\n",
616 ci->i_truncate_seq, truncate_seq);
617 ci->i_truncate_seq = truncate_seq;
619 /* the MDS should have revoked these caps */
620 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
623 CEPH_CAP_FILE_LAZYIO));
625 * If we hold relevant caps, or in the case where we're
626 * not the only client referencing this file and we
627 * don't hold those caps, then we need to check whether
628 * the file is either opened or mmaped
630 if ((issued & (CEPH_CAP_FILE_CACHE|
631 CEPH_CAP_FILE_BUFFER)) ||
632 mapping_mapped(inode->i_mapping) ||
633 __ceph_caps_file_wanted(ci)) {
634 ci->i_truncate_pending++;
639 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
640 ci->i_truncate_size != truncate_size) {
641 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
643 ci->i_truncate_size = truncate_size;
647 ceph_fscache_invalidate(inode);
652 void ceph_fill_file_time(struct inode *inode, int issued,
653 u64 time_warp_seq, struct timespec *ctime,
654 struct timespec *mtime, struct timespec *atime)
656 struct ceph_inode_info *ci = ceph_inode(inode);
659 if (issued & (CEPH_CAP_FILE_EXCL|
661 CEPH_CAP_FILE_BUFFER|
663 CEPH_CAP_XATTR_EXCL)) {
664 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
665 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
666 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
667 ctime->tv_sec, ctime->tv_nsec);
668 inode->i_ctime = *ctime;
670 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
671 /* the MDS did a utimes() */
672 dout("mtime %ld.%09ld -> %ld.%09ld "
674 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
675 mtime->tv_sec, mtime->tv_nsec,
676 ci->i_time_warp_seq, (int)time_warp_seq);
678 inode->i_mtime = *mtime;
679 inode->i_atime = *atime;
680 ci->i_time_warp_seq = time_warp_seq;
681 } else if (time_warp_seq == ci->i_time_warp_seq) {
682 /* nobody did utimes(); take the max */
683 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
684 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
685 inode->i_mtime.tv_sec,
686 inode->i_mtime.tv_nsec,
687 mtime->tv_sec, mtime->tv_nsec);
688 inode->i_mtime = *mtime;
690 if (timespec_compare(atime, &inode->i_atime) > 0) {
691 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
692 inode->i_atime.tv_sec,
693 inode->i_atime.tv_nsec,
694 atime->tv_sec, atime->tv_nsec);
695 inode->i_atime = *atime;
697 } else if (issued & CEPH_CAP_FILE_EXCL) {
698 /* we did a utimes(); ignore mds values */
703 /* we have no write|excl caps; whatever the MDS says is true */
704 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
705 inode->i_ctime = *ctime;
706 inode->i_mtime = *mtime;
707 inode->i_atime = *atime;
708 ci->i_time_warp_seq = time_warp_seq;
713 if (warn) /* time_warp_seq shouldn't go backwards */
714 dout("%p mds time_warp_seq %llu < %u\n",
715 inode, time_warp_seq, ci->i_time_warp_seq);
719 * Populate an inode based on info from mds. May be called on new or
722 static int fill_inode(struct inode *inode, struct page *locked_page,
723 struct ceph_mds_reply_info_in *iinfo,
724 struct ceph_mds_reply_dirfrag *dirinfo,
725 struct ceph_mds_session *session,
726 unsigned long ttl_from, int cap_fmode,
727 struct ceph_cap_reservation *caps_reservation)
729 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
730 struct ceph_mds_reply_inode *info = iinfo->in;
731 struct ceph_inode_info *ci = ceph_inode(inode);
732 int issued = 0, implemented, new_issued;
733 struct timespec mtime, atime, ctime;
734 struct ceph_buffer *xattr_blob = NULL;
735 struct ceph_cap *new_cap = NULL;
738 bool queue_trunc = false;
739 bool new_version = false;
740 bool fill_inline = false;
742 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
743 inode, ceph_vinop(inode), le64_to_cpu(info->version),
746 /* prealloc new cap struct */
747 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
748 new_cap = ceph_get_cap(mdsc, caps_reservation);
751 * prealloc xattr data, if it looks like we'll need it. only
752 * if len > 4 (meaning there are actually xattrs; the first 4
753 * bytes are the xattr count).
755 if (iinfo->xattr_len > 4) {
756 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
758 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
762 spin_lock(&ci->i_ceph_lock);
765 * provided version will be odd if inode value is projected,
766 * even if stable. skip the update if we have newer stable
767 * info (ours>=theirs, e.g. due to racing mds replies), unless
768 * we are getting projected (unstable) info (in which case the
769 * version is odd, and we want ours>theirs).
775 if (ci->i_version == 0 ||
776 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
777 le64_to_cpu(info->version) > (ci->i_version & ~1)))
780 issued = __ceph_caps_issued(ci, &implemented);
781 issued |= implemented | __ceph_caps_dirty(ci);
782 new_issued = ~issued & le32_to_cpu(info->cap.caps);
785 ci->i_version = le64_to_cpu(info->version);
787 inode->i_rdev = le32_to_cpu(info->rdev);
788 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
790 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
791 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
792 inode->i_mode = le32_to_cpu(info->mode);
793 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
794 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
795 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
796 from_kuid(&init_user_ns, inode->i_uid),
797 from_kgid(&init_user_ns, inode->i_gid));
800 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
801 (issued & CEPH_CAP_LINK_EXCL) == 0)
802 set_nlink(inode, le32_to_cpu(info->nlink));
804 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
805 /* be careful with mtime, atime, size */
806 ceph_decode_timespec(&atime, &info->atime);
807 ceph_decode_timespec(&mtime, &info->mtime);
808 ceph_decode_timespec(&ctime, &info->ctime);
809 ceph_fill_file_time(inode, issued,
810 le32_to_cpu(info->time_warp_seq),
811 &ctime, &mtime, &atime);
815 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
816 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
817 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
818 ci->i_layout = info->layout;
819 ci->i_pool_ns_len = iinfo->pool_ns_len;
821 queue_trunc = ceph_fill_file_size(inode, issued,
822 le32_to_cpu(info->truncate_seq),
823 le64_to_cpu(info->truncate_size),
824 le64_to_cpu(info->size));
825 /* only update max_size on auth cap */
826 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
827 ci->i_max_size != le64_to_cpu(info->max_size)) {
828 dout("max_size %lld -> %llu\n", ci->i_max_size,
829 le64_to_cpu(info->max_size));
830 ci->i_max_size = le64_to_cpu(info->max_size);
835 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
836 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
837 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
838 if (ci->i_xattrs.blob)
839 ceph_buffer_put(ci->i_xattrs.blob);
840 ci->i_xattrs.blob = xattr_blob;
842 memcpy(ci->i_xattrs.blob->vec.iov_base,
843 iinfo->xattr_data, iinfo->xattr_len);
844 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
845 ceph_forget_all_cached_acls(inode);
849 inode->i_mapping->a_ops = &ceph_aops;
851 switch (inode->i_mode & S_IFMT) {
856 init_special_inode(inode, inode->i_mode, inode->i_rdev);
857 inode->i_op = &ceph_file_iops;
860 inode->i_op = &ceph_file_iops;
861 inode->i_fop = &ceph_file_fops;
864 inode->i_op = &ceph_symlink_iops;
865 if (!ci->i_symlink) {
866 u32 symlen = iinfo->symlink_len;
869 spin_unlock(&ci->i_ceph_lock);
871 if (symlen != i_size_read(inode)) {
872 pr_err("fill_inode %llx.%llx BAD symlink "
873 "size %lld\n", ceph_vinop(inode),
875 i_size_write(inode, symlen);
876 inode->i_blocks = calc_inode_blocks(symlen);
880 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
884 spin_lock(&ci->i_ceph_lock);
888 kfree(sym); /* lost a race */
890 inode->i_link = ci->i_symlink;
893 inode->i_op = &ceph_dir_iops;
894 inode->i_fop = &ceph_dir_fops;
896 ci->i_dir_layout = iinfo->dir_layout;
898 ci->i_files = le64_to_cpu(info->files);
899 ci->i_subdirs = le64_to_cpu(info->subdirs);
900 ci->i_rbytes = le64_to_cpu(info->rbytes);
901 ci->i_rfiles = le64_to_cpu(info->rfiles);
902 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
903 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
906 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
907 ceph_vinop(inode), inode->i_mode);
910 /* were we issued a capability? */
911 if (info->cap.caps) {
912 if (ceph_snap(inode) == CEPH_NOSNAP) {
913 unsigned caps = le32_to_cpu(info->cap.caps);
914 ceph_add_cap(inode, session,
915 le64_to_cpu(info->cap.cap_id),
917 le32_to_cpu(info->cap.wanted),
918 le32_to_cpu(info->cap.seq),
919 le32_to_cpu(info->cap.mseq),
920 le64_to_cpu(info->cap.realm),
921 info->cap.flags, &new_cap);
923 /* set dir completion flag? */
924 if (S_ISDIR(inode->i_mode) &&
925 ci->i_files == 0 && ci->i_subdirs == 0 &&
926 (caps & CEPH_CAP_FILE_SHARED) &&
927 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
928 !__ceph_dir_is_complete(ci)) {
929 dout(" marking %p complete (empty)\n", inode);
930 i_size_write(inode, 0);
931 __ceph_dir_set_complete(ci,
932 atomic64_read(&ci->i_release_count),
933 atomic64_read(&ci->i_ordered_count));
938 dout(" %p got snap_caps %s\n", inode,
939 ceph_cap_string(le32_to_cpu(info->cap.caps)));
940 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
942 __ceph_get_fmode(ci, cap_fmode);
944 } else if (cap_fmode >= 0) {
945 pr_warn("mds issued no caps on %llx.%llx\n",
947 __ceph_get_fmode(ci, cap_fmode);
950 if (iinfo->inline_version > 0 &&
951 iinfo->inline_version >= ci->i_inline_version) {
952 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
953 ci->i_inline_version = iinfo->inline_version;
954 if (ci->i_inline_version != CEPH_INLINE_NONE &&
956 (le32_to_cpu(info->cap.caps) & cache_caps)))
960 spin_unlock(&ci->i_ceph_lock);
963 ceph_fill_inline_data(inode, locked_page,
964 iinfo->inline_data, iinfo->inline_len);
967 wake_up_all(&ci->i_cap_wq);
969 /* queue truncate if we saw i_size decrease */
971 ceph_queue_vmtruncate(inode);
973 /* populate frag tree */
974 if (S_ISDIR(inode->i_mode))
975 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
977 /* update delegation info? */
979 ceph_fill_dirfrag(inode, dirinfo);
984 ceph_put_cap(mdsc, new_cap);
986 ceph_buffer_put(xattr_blob);
991 * caller should hold session s_mutex.
993 static void update_dentry_lease(struct dentry *dentry,
994 struct ceph_mds_reply_lease *lease,
995 struct ceph_mds_session *session,
996 unsigned long from_time)
998 struct ceph_dentry_info *di = ceph_dentry(dentry);
999 long unsigned duration = le32_to_cpu(lease->duration_ms);
1000 long unsigned ttl = from_time + (duration * HZ) / 1000;
1001 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1004 /* only track leases on regular dentries */
1005 if (dentry->d_op != &ceph_dentry_ops)
1008 spin_lock(&dentry->d_lock);
1009 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1010 dentry, duration, ttl);
1012 /* make lease_rdcache_gen match directory */
1013 dir = d_inode(dentry->d_parent);
1014 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1019 if (di->lease_gen == session->s_cap_gen &&
1020 time_before(ttl, dentry->d_time))
1021 goto out_unlock; /* we already have a newer lease. */
1023 if (di->lease_session && di->lease_session != session)
1026 ceph_dentry_lru_touch(dentry);
1028 if (!di->lease_session)
1029 di->lease_session = ceph_get_mds_session(session);
1030 di->lease_gen = session->s_cap_gen;
1031 di->lease_seq = le32_to_cpu(lease->seq);
1032 di->lease_renew_after = half_ttl;
1033 di->lease_renew_from = 0;
1034 dentry->d_time = ttl;
1036 spin_unlock(&dentry->d_lock);
1041 * splice a dentry to an inode.
1042 * caller must hold directory i_mutex for this to be safe.
1044 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1046 struct dentry *realdn;
1048 BUG_ON(d_inode(dn));
1050 /* dn must be unhashed */
1051 if (!d_unhashed(dn))
1053 realdn = d_splice_alias(in, dn);
1054 if (IS_ERR(realdn)) {
1055 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1056 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1057 dn = realdn; /* note realdn contains the error */
1059 } else if (realdn) {
1060 dout("dn %p (%d) spliced with %p (%d) "
1061 "inode %p ino %llx.%llx\n",
1063 realdn, d_count(realdn),
1064 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1068 BUG_ON(!ceph_dentry(dn));
1069 dout("dn %p attached to %p ino %llx.%llx\n",
1070 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1077 * Incorporate results into the local cache. This is either just
1078 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1081 * A reply may contain
1082 * a directory inode along with a dentry.
1083 * and/or a target inode
1085 * Called with snap_rwsem (read).
1087 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1088 struct ceph_mds_session *session)
1090 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1091 struct inode *in = NULL;
1092 struct ceph_vino vino;
1093 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1096 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1097 rinfo->head->is_dentry, rinfo->head->is_target);
1103 * If we resend completed ops to a recovering mds, we get no
1104 * trace. Since that is very rare, pretend this is the case
1105 * to ensure the 'no trace' handlers in the callers behave.
1107 * Fill in inodes unconditionally to avoid breaking cap
1110 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1111 pr_info("fill_trace faking empty trace on %lld %s\n",
1112 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1113 if (rinfo->head->is_dentry) {
1114 rinfo->head->is_dentry = 0;
1115 err = fill_inode(req->r_locked_dir,
1116 &rinfo->diri, rinfo->dirfrag,
1117 session, req->r_request_started, -1);
1119 if (rinfo->head->is_target) {
1120 rinfo->head->is_target = 0;
1121 ininfo = rinfo->targeti.in;
1122 vino.ino = le64_to_cpu(ininfo->ino);
1123 vino.snap = le64_to_cpu(ininfo->snapid);
1124 in = ceph_get_inode(sb, vino);
1125 err = fill_inode(in, &rinfo->targeti, NULL,
1126 session, req->r_request_started,
1133 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1134 dout("fill_trace reply is empty!\n");
1135 if (rinfo->head->result == 0 && req->r_locked_dir)
1136 ceph_invalidate_dir_request(req);
1140 if (rinfo->head->is_dentry) {
1141 struct inode *dir = req->r_locked_dir;
1144 err = fill_inode(dir, NULL,
1145 &rinfo->diri, rinfo->dirfrag,
1146 session, req->r_request_started, -1,
1147 &req->r_caps_reservation);
1154 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1156 struct dentry *dn, *parent;
1158 BUG_ON(!rinfo->head->is_target);
1159 BUG_ON(req->r_dentry);
1161 parent = d_find_any_alias(dir);
1164 dname.name = rinfo->dname;
1165 dname.len = rinfo->dname_len;
1166 dname.hash = full_name_hash(dname.name, dname.len);
1167 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1168 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1170 dn = d_lookup(parent, &dname);
1171 dout("d_lookup on parent=%p name=%.*s got %p\n",
1172 parent, dname.len, dname.name, dn);
1175 dn = d_alloc(parent, &dname);
1176 dout("d_alloc %p '%.*s' = %p\n", parent,
1177 dname.len, dname.name, dn);
1183 err = ceph_init_dentry(dn);
1189 } else if (d_really_is_positive(dn) &&
1190 (ceph_ino(d_inode(dn)) != vino.ino ||
1191 ceph_snap(d_inode(dn)) != vino.snap)) {
1192 dout(" dn %p points to wrong inode %p\n",
1204 if (rinfo->head->is_target) {
1205 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1206 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1208 in = ceph_get_inode(sb, vino);
1213 req->r_target_inode = in;
1215 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1216 session, req->r_request_started,
1217 (!req->r_aborted && rinfo->head->result == 0) ?
1219 &req->r_caps_reservation);
1221 pr_err("fill_inode badness %p %llx.%llx\n",
1222 in, ceph_vinop(in));
1228 * ignore null lease/binding on snapdir ENOENT, or else we
1229 * will have trouble splicing in the virtual snapdir later
1231 if (rinfo->head->is_dentry && !req->r_aborted &&
1232 req->r_locked_dir &&
1233 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1234 fsc->mount_options->snapdir_name,
1235 req->r_dentry->d_name.len))) {
1237 * lookup link rename : null -> possibly existing inode
1238 * mknod symlink mkdir : null -> new inode
1239 * unlink : linked -> null
1241 struct inode *dir = req->r_locked_dir;
1242 struct dentry *dn = req->r_dentry;
1243 bool have_dir_cap, have_lease;
1247 BUG_ON(d_inode(dn->d_parent) != dir);
1248 BUG_ON(ceph_ino(dir) !=
1249 le64_to_cpu(rinfo->diri.in->ino));
1250 BUG_ON(ceph_snap(dir) !=
1251 le64_to_cpu(rinfo->diri.in->snapid));
1253 /* do we have a lease on the whole dir? */
1255 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1256 CEPH_CAP_FILE_SHARED);
1258 /* do we have a dn lease? */
1259 have_lease = have_dir_cap ||
1260 le32_to_cpu(rinfo->dlease->duration_ms);
1262 dout("fill_trace no dentry lease or dir cap\n");
1265 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1266 struct inode *olddir = req->r_old_dentry_dir;
1269 dout(" src %p '%pd' dst %p '%pd'\n",
1273 dout("fill_trace doing d_move %p -> %p\n",
1274 req->r_old_dentry, dn);
1276 /* d_move screws up sibling dentries' offsets */
1277 ceph_dir_clear_ordered(dir);
1278 ceph_dir_clear_ordered(olddir);
1280 d_move(req->r_old_dentry, dn);
1281 dout(" src %p '%pd' dst %p '%pd'\n",
1286 /* ensure target dentry is invalidated, despite
1287 rehashing bug in vfs_rename_dir */
1288 ceph_invalidate_dentry_lease(dn);
1290 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1291 ceph_dentry(req->r_old_dentry)->offset);
1293 dn = req->r_old_dentry; /* use old_dentry */
1297 if (!rinfo->head->is_target) {
1298 dout("fill_trace null dentry\n");
1299 if (d_really_is_positive(dn)) {
1300 ceph_dir_clear_ordered(dir);
1301 dout("d_delete %p\n", dn);
1304 if (have_lease && d_unhashed(dn))
1306 update_dentry_lease(dn, rinfo->dlease,
1308 req->r_request_started);
1313 /* attach proper inode */
1314 if (d_really_is_negative(dn)) {
1315 ceph_dir_clear_ordered(dir);
1317 dn = splice_dentry(dn, in);
1322 req->r_dentry = dn; /* may have spliced */
1323 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1324 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1325 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1332 update_dentry_lease(dn, rinfo->dlease, session,
1333 req->r_request_started);
1334 dout(" final dn %p\n", dn);
1335 } else if (!req->r_aborted &&
1336 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1337 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1338 struct dentry *dn = req->r_dentry;
1339 struct inode *dir = req->r_locked_dir;
1341 /* fill out a snapdir LOOKUPSNAP dentry */
1344 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1345 dout(" linking snapped dir %p to dn %p\n", in, dn);
1346 ceph_dir_clear_ordered(dir);
1348 dn = splice_dentry(dn, in);
1353 req->r_dentry = dn; /* may have spliced */
1356 dout("fill_trace done err=%d\n", err);
1361 * Prepopulate our cache with readdir results, leases, etc.
1363 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1364 struct ceph_mds_session *session)
1366 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1369 for (i = 0; i < rinfo->dir_nr; i++) {
1370 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1371 struct ceph_vino vino;
1375 vino.ino = le64_to_cpu(rde->inode.in->ino);
1376 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1378 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1381 dout("new_inode badness got %d\n", err);
1384 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1385 req->r_request_started, -1,
1386 &req->r_caps_reservation);
1388 pr_err("fill_inode badness on %p got %d\n", in, rc);
1397 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1401 put_page(ctl->page);
1406 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1407 struct ceph_readdir_cache_control *ctl,
1408 struct ceph_mds_request *req)
1410 struct ceph_inode_info *ci = ceph_inode(dir);
1411 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1412 unsigned idx = ctl->index % nsize;
1413 pgoff_t pgoff = ctl->index / nsize;
1415 if (!ctl->page || pgoff != page_index(ctl->page)) {
1416 ceph_readdir_cache_release(ctl);
1418 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1420 ctl->page = find_lock_page(&dir->i_data, pgoff);
1423 return idx == 0 ? -ENOMEM : 0;
1425 /* reading/filling the cache are serialized by
1426 * i_mutex, no need to use page lock */
1427 unlock_page(ctl->page);
1428 ctl->dentries = kmap(ctl->page);
1430 memset(ctl->dentries, 0, PAGE_SIZE);
1433 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1434 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1435 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1436 ctl->dentries[idx] = dn;
1439 dout("disable readdir cache\n");
1445 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1446 struct ceph_mds_session *session)
1448 struct dentry *parent = req->r_dentry;
1449 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1450 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1454 int err = 0, skipped = 0, ret, i;
1455 struct inode *snapdir = NULL;
1456 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1457 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1460 struct ceph_readdir_cache_control cache_ctl = {};
1463 return readdir_prepopulate_inodes_only(req, session);
1465 if (rinfo->hash_order && req->r_path2) {
1466 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1467 req->r_path2, strlen(req->r_path2));
1468 last_hash = ceph_frag_value(last_hash);
1471 if (rinfo->dir_dir &&
1472 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1473 dout("readdir_prepopulate got new frag %x -> %x\n",
1474 frag, le32_to_cpu(rinfo->dir_dir->frag));
1475 frag = le32_to_cpu(rinfo->dir_dir->frag);
1476 if (!rinfo->hash_order)
1477 req->r_readdir_offset = 2;
1480 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1481 snapdir = ceph_get_snapdir(d_inode(parent));
1482 parent = d_find_alias(snapdir);
1483 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1484 rinfo->dir_nr, parent);
1486 dout("readdir_prepopulate %d items under dn %p\n",
1487 rinfo->dir_nr, parent);
1489 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1492 if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1493 /* note dir version at start of readdir so we can tell
1494 * if any dentries get dropped */
1495 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1496 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1497 req->r_readdir_cache_idx = 0;
1500 cache_ctl.index = req->r_readdir_cache_idx;
1501 fpos_offset = req->r_readdir_offset;
1503 /* FIXME: release caps/leases if error occurs */
1504 for (i = 0; i < rinfo->dir_nr; i++) {
1505 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1506 struct ceph_vino vino;
1508 dname.name = rde->name;
1509 dname.len = rde->name_len;
1510 dname.hash = full_name_hash(dname.name, dname.len);
1512 vino.ino = le64_to_cpu(rde->inode.in->ino);
1513 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1515 if (rinfo->hash_order) {
1516 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1517 rde->name, rde->name_len);
1518 hash = ceph_frag_value(hash);
1519 if (hash != last_hash)
1522 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1524 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1528 dn = d_lookup(parent, &dname);
1529 dout("d_lookup on parent=%p name=%.*s got %p\n",
1530 parent, dname.len, dname.name, dn);
1533 dn = d_alloc(parent, &dname);
1534 dout("d_alloc %p '%.*s' = %p\n", parent,
1535 dname.len, dname.name, dn);
1537 dout("d_alloc badness\n");
1541 ret = ceph_init_dentry(dn);
1547 } else if (d_really_is_positive(dn) &&
1548 (ceph_ino(d_inode(dn)) != vino.ino ||
1549 ceph_snap(d_inode(dn)) != vino.snap)) {
1550 dout(" dn %p points to wrong inode %p\n",
1558 if (d_really_is_positive(dn)) {
1561 in = ceph_get_inode(parent->d_sb, vino);
1563 dout("new_inode badness\n");
1571 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1572 req->r_request_started, -1,
1573 &req->r_caps_reservation);
1575 pr_err("fill_inode badness on %p\n", in);
1576 if (d_really_is_negative(dn))
1583 if (d_really_is_negative(dn)) {
1584 struct dentry *realdn;
1586 if (ceph_security_xattr_deadlock(in)) {
1587 dout(" skip splicing dn %p to inode %p"
1588 " (security xattr deadlock)\n", dn, in);
1594 realdn = splice_dentry(dn, in);
1595 if (IS_ERR(realdn)) {
1596 err = PTR_ERR(realdn);
1604 ceph_dentry(dn)->offset = rde->offset;
1606 update_dentry_lease(dn, rde->lease, req->r_session,
1607 req->r_request_started);
1609 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1610 ret = fill_readdir_cache(d_inode(parent), dn,
1620 if (err == 0 && skipped == 0) {
1621 req->r_did_prepopulate = true;
1622 req->r_readdir_cache_idx = cache_ctl.index;
1624 ceph_readdir_cache_release(&cache_ctl);
1629 dout("readdir_prepopulate done\n");
1633 int ceph_inode_set_size(struct inode *inode, loff_t size)
1635 struct ceph_inode_info *ci = ceph_inode(inode);
1638 spin_lock(&ci->i_ceph_lock);
1639 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1640 i_size_write(inode, size);
1641 inode->i_blocks = calc_inode_blocks(size);
1643 /* tell the MDS if we are approaching max_size */
1644 if ((size << 1) >= ci->i_max_size &&
1645 (ci->i_reported_size << 1) < ci->i_max_size)
1648 spin_unlock(&ci->i_ceph_lock);
1653 * Write back inode data in a worker thread. (This can't be done
1654 * in the message handler context.)
1656 void ceph_queue_writeback(struct inode *inode)
1659 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1660 &ceph_inode(inode)->i_wb_work)) {
1661 dout("ceph_queue_writeback %p\n", inode);
1663 dout("ceph_queue_writeback %p failed\n", inode);
1668 static void ceph_writeback_work(struct work_struct *work)
1670 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1672 struct inode *inode = &ci->vfs_inode;
1674 dout("writeback %p\n", inode);
1675 filemap_fdatawrite(&inode->i_data);
1680 * queue an async invalidation
1682 void ceph_queue_invalidate(struct inode *inode)
1685 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1686 &ceph_inode(inode)->i_pg_inv_work)) {
1687 dout("ceph_queue_invalidate %p\n", inode);
1689 dout("ceph_queue_invalidate %p failed\n", inode);
1695 * Invalidate inode pages in a worker thread. (This can't be done
1696 * in the message handler context.)
1698 static void ceph_invalidate_work(struct work_struct *work)
1700 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1702 struct inode *inode = &ci->vfs_inode;
1703 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1707 mutex_lock(&ci->i_truncate_mutex);
1709 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1710 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1711 inode, ceph_ino(inode));
1712 mapping_set_error(inode->i_mapping, -EIO);
1713 truncate_pagecache(inode, 0);
1714 mutex_unlock(&ci->i_truncate_mutex);
1718 spin_lock(&ci->i_ceph_lock);
1719 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1720 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1721 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1722 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1724 spin_unlock(&ci->i_ceph_lock);
1725 mutex_unlock(&ci->i_truncate_mutex);
1728 orig_gen = ci->i_rdcache_gen;
1729 spin_unlock(&ci->i_ceph_lock);
1731 truncate_pagecache(inode, 0);
1733 spin_lock(&ci->i_ceph_lock);
1734 if (orig_gen == ci->i_rdcache_gen &&
1735 orig_gen == ci->i_rdcache_revoking) {
1736 dout("invalidate_pages %p gen %d successful\n", inode,
1738 ci->i_rdcache_revoking--;
1741 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1742 inode, orig_gen, ci->i_rdcache_gen,
1743 ci->i_rdcache_revoking);
1744 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1747 spin_unlock(&ci->i_ceph_lock);
1748 mutex_unlock(&ci->i_truncate_mutex);
1751 ceph_check_caps(ci, 0, NULL);
1757 * called by trunc_wq;
1759 * We also truncate in a separate thread as well.
1761 static void ceph_vmtruncate_work(struct work_struct *work)
1763 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1765 struct inode *inode = &ci->vfs_inode;
1767 dout("vmtruncate_work %p\n", inode);
1768 __ceph_do_pending_vmtruncate(inode);
1773 * Queue an async vmtruncate. If we fail to queue work, we will handle
1774 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1776 void ceph_queue_vmtruncate(struct inode *inode)
1778 struct ceph_inode_info *ci = ceph_inode(inode);
1782 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1783 &ci->i_vmtruncate_work)) {
1784 dout("ceph_queue_vmtruncate %p\n", inode);
1786 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1787 inode, ci->i_truncate_pending);
1793 * Make sure any pending truncation is applied before doing anything
1794 * that may depend on it.
1796 void __ceph_do_pending_vmtruncate(struct inode *inode)
1798 struct ceph_inode_info *ci = ceph_inode(inode);
1800 int wrbuffer_refs, finish = 0;
1802 mutex_lock(&ci->i_truncate_mutex);
1804 spin_lock(&ci->i_ceph_lock);
1805 if (ci->i_truncate_pending == 0) {
1806 dout("__do_pending_vmtruncate %p none pending\n", inode);
1807 spin_unlock(&ci->i_ceph_lock);
1808 mutex_unlock(&ci->i_truncate_mutex);
1813 * make sure any dirty snapped pages are flushed before we
1814 * possibly truncate them.. so write AND block!
1816 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1817 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1819 spin_unlock(&ci->i_ceph_lock);
1820 filemap_write_and_wait_range(&inode->i_data, 0,
1821 inode->i_sb->s_maxbytes);
1825 /* there should be no reader or writer */
1826 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1828 to = ci->i_truncate_size;
1829 wrbuffer_refs = ci->i_wrbuffer_ref;
1830 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1831 ci->i_truncate_pending, to);
1832 spin_unlock(&ci->i_ceph_lock);
1834 truncate_pagecache(inode, to);
1836 spin_lock(&ci->i_ceph_lock);
1837 if (to == ci->i_truncate_size) {
1838 ci->i_truncate_pending = 0;
1841 spin_unlock(&ci->i_ceph_lock);
1845 mutex_unlock(&ci->i_truncate_mutex);
1847 if (wrbuffer_refs == 0)
1848 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1850 wake_up_all(&ci->i_cap_wq);
1856 static const struct inode_operations ceph_symlink_iops = {
1857 .readlink = generic_readlink,
1858 .get_link = simple_get_link,
1859 .setattr = ceph_setattr,
1860 .getattr = ceph_getattr,
1861 .setxattr = ceph_setxattr,
1862 .getxattr = ceph_getxattr,
1863 .listxattr = ceph_listxattr,
1864 .removexattr = ceph_removexattr,
1870 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1872 struct inode *inode = d_inode(dentry);
1873 struct ceph_inode_info *ci = ceph_inode(inode);
1874 const unsigned int ia_valid = attr->ia_valid;
1875 struct ceph_mds_request *req;
1876 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1877 struct ceph_cap_flush *prealloc_cf;
1879 int release = 0, dirtied = 0;
1882 int inode_dirty_flags = 0;
1883 bool lock_snap_rwsem = false;
1885 if (ceph_snap(inode) != CEPH_NOSNAP)
1888 err = inode_change_ok(inode, attr);
1892 prealloc_cf = ceph_alloc_cap_flush();
1896 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1899 ceph_free_cap_flush(prealloc_cf);
1900 return PTR_ERR(req);
1903 spin_lock(&ci->i_ceph_lock);
1904 issued = __ceph_caps_issued(ci, NULL);
1906 if (!ci->i_head_snapc &&
1907 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1908 lock_snap_rwsem = true;
1909 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1910 spin_unlock(&ci->i_ceph_lock);
1911 down_read(&mdsc->snap_rwsem);
1912 spin_lock(&ci->i_ceph_lock);
1913 issued = __ceph_caps_issued(ci, NULL);
1917 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1919 if (ia_valid & ATTR_UID) {
1920 dout("setattr %p uid %d -> %d\n", inode,
1921 from_kuid(&init_user_ns, inode->i_uid),
1922 from_kuid(&init_user_ns, attr->ia_uid));
1923 if (issued & CEPH_CAP_AUTH_EXCL) {
1924 inode->i_uid = attr->ia_uid;
1925 dirtied |= CEPH_CAP_AUTH_EXCL;
1926 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1927 !uid_eq(attr->ia_uid, inode->i_uid)) {
1928 req->r_args.setattr.uid = cpu_to_le32(
1929 from_kuid(&init_user_ns, attr->ia_uid));
1930 mask |= CEPH_SETATTR_UID;
1931 release |= CEPH_CAP_AUTH_SHARED;
1934 if (ia_valid & ATTR_GID) {
1935 dout("setattr %p gid %d -> %d\n", inode,
1936 from_kgid(&init_user_ns, inode->i_gid),
1937 from_kgid(&init_user_ns, attr->ia_gid));
1938 if (issued & CEPH_CAP_AUTH_EXCL) {
1939 inode->i_gid = attr->ia_gid;
1940 dirtied |= CEPH_CAP_AUTH_EXCL;
1941 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1942 !gid_eq(attr->ia_gid, inode->i_gid)) {
1943 req->r_args.setattr.gid = cpu_to_le32(
1944 from_kgid(&init_user_ns, attr->ia_gid));
1945 mask |= CEPH_SETATTR_GID;
1946 release |= CEPH_CAP_AUTH_SHARED;
1949 if (ia_valid & ATTR_MODE) {
1950 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1952 if (issued & CEPH_CAP_AUTH_EXCL) {
1953 inode->i_mode = attr->ia_mode;
1954 dirtied |= CEPH_CAP_AUTH_EXCL;
1955 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1956 attr->ia_mode != inode->i_mode) {
1957 inode->i_mode = attr->ia_mode;
1958 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1959 mask |= CEPH_SETATTR_MODE;
1960 release |= CEPH_CAP_AUTH_SHARED;
1964 if (ia_valid & ATTR_ATIME) {
1965 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1966 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1967 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1968 if (issued & CEPH_CAP_FILE_EXCL) {
1969 ci->i_time_warp_seq++;
1970 inode->i_atime = attr->ia_atime;
1971 dirtied |= CEPH_CAP_FILE_EXCL;
1972 } else if ((issued & CEPH_CAP_FILE_WR) &&
1973 timespec_compare(&inode->i_atime,
1974 &attr->ia_atime) < 0) {
1975 inode->i_atime = attr->ia_atime;
1976 dirtied |= CEPH_CAP_FILE_WR;
1977 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1978 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1979 ceph_encode_timespec(&req->r_args.setattr.atime,
1981 mask |= CEPH_SETATTR_ATIME;
1982 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1986 if (ia_valid & ATTR_MTIME) {
1987 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1988 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1989 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1990 if (issued & CEPH_CAP_FILE_EXCL) {
1991 ci->i_time_warp_seq++;
1992 inode->i_mtime = attr->ia_mtime;
1993 dirtied |= CEPH_CAP_FILE_EXCL;
1994 } else if ((issued & CEPH_CAP_FILE_WR) &&
1995 timespec_compare(&inode->i_mtime,
1996 &attr->ia_mtime) < 0) {
1997 inode->i_mtime = attr->ia_mtime;
1998 dirtied |= CEPH_CAP_FILE_WR;
1999 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2000 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2001 ceph_encode_timespec(&req->r_args.setattr.mtime,
2003 mask |= CEPH_SETATTR_MTIME;
2004 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2008 if (ia_valid & ATTR_SIZE) {
2009 dout("setattr %p size %lld -> %lld\n", inode,
2010 inode->i_size, attr->ia_size);
2011 if ((issued & CEPH_CAP_FILE_EXCL) &&
2012 attr->ia_size > inode->i_size) {
2013 i_size_write(inode, attr->ia_size);
2014 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2015 inode->i_ctime = attr->ia_ctime;
2016 ci->i_reported_size = attr->ia_size;
2017 dirtied |= CEPH_CAP_FILE_EXCL;
2018 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2019 attr->ia_size != inode->i_size) {
2020 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2021 req->r_args.setattr.old_size =
2022 cpu_to_le64(inode->i_size);
2023 mask |= CEPH_SETATTR_SIZE;
2024 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2029 /* these do nothing */
2030 if (ia_valid & ATTR_CTIME) {
2031 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2032 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2033 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2034 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2035 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2036 only ? "ctime only" : "ignored");
2037 inode->i_ctime = attr->ia_ctime;
2040 * if kernel wants to dirty ctime but nothing else,
2041 * we need to choose a cap to dirty under, or do
2042 * a almost-no-op setattr
2044 if (issued & CEPH_CAP_AUTH_EXCL)
2045 dirtied |= CEPH_CAP_AUTH_EXCL;
2046 else if (issued & CEPH_CAP_FILE_EXCL)
2047 dirtied |= CEPH_CAP_FILE_EXCL;
2048 else if (issued & CEPH_CAP_XATTR_EXCL)
2049 dirtied |= CEPH_CAP_XATTR_EXCL;
2051 mask |= CEPH_SETATTR_CTIME;
2054 if (ia_valid & ATTR_FILE)
2055 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2058 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2060 inode->i_ctime = current_fs_time(inode->i_sb);
2064 spin_unlock(&ci->i_ceph_lock);
2065 if (lock_snap_rwsem)
2066 up_read(&mdsc->snap_rwsem);
2068 if (inode_dirty_flags)
2069 __mark_inode_dirty(inode, inode_dirty_flags);
2071 if (ia_valid & ATTR_MODE) {
2072 err = posix_acl_chmod(inode, attr->ia_mode);
2078 req->r_inode = inode;
2080 req->r_inode_drop = release;
2081 req->r_args.setattr.mask = cpu_to_le32(mask);
2082 req->r_num_caps = 1;
2083 err = ceph_mdsc_do_request(mdsc, NULL, req);
2085 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2086 ceph_cap_string(dirtied), mask);
2088 ceph_mdsc_put_request(req);
2089 if (mask & CEPH_SETATTR_SIZE)
2090 __ceph_do_pending_vmtruncate(inode);
2091 ceph_free_cap_flush(prealloc_cf);
2094 ceph_mdsc_put_request(req);
2095 ceph_free_cap_flush(prealloc_cf);
2100 * Verify that we have a lease on the given mask. If not,
2101 * do a getattr against an mds.
2103 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2104 int mask, bool force)
2106 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2107 struct ceph_mds_client *mdsc = fsc->mdsc;
2108 struct ceph_mds_request *req;
2111 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2112 dout("do_getattr inode %p SNAPDIR\n", inode);
2116 dout("do_getattr inode %p mask %s mode 0%o\n",
2117 inode, ceph_cap_string(mask), inode->i_mode);
2118 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2121 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2123 return PTR_ERR(req);
2124 req->r_inode = inode;
2126 req->r_num_caps = 1;
2127 req->r_args.getattr.mask = cpu_to_le32(mask);
2128 req->r_locked_page = locked_page;
2129 err = ceph_mdsc_do_request(mdsc, NULL, req);
2130 if (locked_page && err == 0) {
2131 u64 inline_version = req->r_reply_info.targeti.inline_version;
2132 if (inline_version == 0) {
2133 /* the reply is supposed to contain inline data */
2135 } else if (inline_version == CEPH_INLINE_NONE) {
2138 err = req->r_reply_info.targeti.inline_len;
2141 ceph_mdsc_put_request(req);
2142 dout("do_getattr result=%d\n", err);
2148 * Check inode permissions. We verify we have a valid value for
2149 * the AUTH cap, then call the generic handler.
2151 int ceph_permission(struct inode *inode, int mask)
2155 if (mask & MAY_NOT_BLOCK)
2158 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2161 err = generic_permission(inode, mask);
2166 * Get all attributes. Hopefully somedata we'll have a statlite()
2167 * and can limit the fields we require to be accurate.
2169 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2172 struct inode *inode = d_inode(dentry);
2173 struct ceph_inode_info *ci = ceph_inode(inode);
2176 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2178 generic_fillattr(inode, stat);
2179 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2180 if (ceph_snap(inode) != CEPH_NOSNAP)
2181 stat->dev = ceph_snap(inode);
2184 if (S_ISDIR(inode->i_mode)) {
2185 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2187 stat->size = ci->i_rbytes;
2189 stat->size = ci->i_files + ci->i_subdirs;
2191 stat->blksize = 65536;