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/namei.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/posix_acl.h>
13 #include <linux/random.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_dir_iops;
86 inode->i_fop = &ceph_dir_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 mutex_lock(&ci->i_fragtree_mutex);
258 if (ndist == 0 && mds == diri_auth) {
259 /* no delegation info needed. */
260 frag = __ceph_find_frag(ci, id);
263 if (frag->split_by == 0) {
264 /* tree leaf, remove */
265 dout("fill_dirfrag removed %llx.%llx frag %x"
266 " (no ref)\n", ceph_vinop(inode), id);
267 rb_erase(&frag->node, &ci->i_fragtree);
270 /* tree branch, keep and clear */
271 dout("fill_dirfrag cleared %llx.%llx frag %x"
272 " referral\n", ceph_vinop(inode), id);
280 /* find/add this frag to store mds delegation info */
281 frag = __get_or_create_frag(ci, id);
283 /* this is not the end of the world; we can continue
284 with bad/inaccurate delegation info */
285 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
286 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
292 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
293 for (i = 0; i < frag->ndist; i++)
294 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
295 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
296 ceph_vinop(inode), frag->frag, frag->ndist);
299 mutex_unlock(&ci->i_fragtree_mutex);
303 static int ceph_fill_fragtree(struct inode *inode,
304 struct ceph_frag_tree_head *fragtree,
305 struct ceph_mds_reply_dirfrag *dirinfo)
307 struct ceph_inode_info *ci = ceph_inode(inode);
308 struct ceph_inode_frag *frag;
309 struct rb_node *rb_node;
314 mutex_lock(&ci->i_fragtree_mutex);
315 nsplits = le32_to_cpu(fragtree->nsplits);
317 i = prandom_u32() % nsplits;
318 id = le32_to_cpu(fragtree->splits[i].frag);
319 if (!__ceph_find_frag(ci, id))
321 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
322 rb_node = rb_first(&ci->i_fragtree);
323 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
324 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
327 if (!update && dirinfo) {
328 id = le32_to_cpu(dirinfo->frag);
329 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
335 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
336 rb_node = rb_first(&ci->i_fragtree);
337 for (i = 0; i < nsplits; i++) {
338 id = le32_to_cpu(fragtree->splits[i].frag);
341 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
342 if (ceph_frag_compare(frag->frag, id) >= 0) {
343 if (frag->frag != id)
346 rb_node = rb_next(rb_node);
349 rb_node = rb_next(rb_node);
350 rb_erase(&frag->node, &ci->i_fragtree);
355 frag = __get_or_create_frag(ci, id);
359 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
360 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
363 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
364 rb_node = rb_next(rb_node);
365 rb_erase(&frag->node, &ci->i_fragtree);
369 mutex_unlock(&ci->i_fragtree_mutex);
374 * initialize a newly allocated inode.
376 struct inode *ceph_alloc_inode(struct super_block *sb)
378 struct ceph_inode_info *ci;
381 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
385 dout("alloc_inode %p\n", &ci->vfs_inode);
387 spin_lock_init(&ci->i_ceph_lock);
390 ci->i_inline_version = 0;
391 ci->i_time_warp_seq = 0;
392 ci->i_ceph_flags = 0;
393 ci->i_ordered_count = 0;
394 atomic_set(&ci->i_release_count, 1);
395 atomic_set(&ci->i_complete_count, 0);
396 ci->i_symlink = NULL;
398 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
400 ci->i_fragtree = RB_ROOT;
401 mutex_init(&ci->i_fragtree_mutex);
403 ci->i_xattrs.blob = NULL;
404 ci->i_xattrs.prealloc_blob = NULL;
405 ci->i_xattrs.dirty = false;
406 ci->i_xattrs.index = RB_ROOT;
407 ci->i_xattrs.count = 0;
408 ci->i_xattrs.names_size = 0;
409 ci->i_xattrs.vals_size = 0;
410 ci->i_xattrs.version = 0;
411 ci->i_xattrs.index_version = 0;
413 ci->i_caps = RB_ROOT;
414 ci->i_auth_cap = NULL;
415 ci->i_dirty_caps = 0;
416 ci->i_flushing_caps = 0;
417 INIT_LIST_HEAD(&ci->i_dirty_item);
418 INIT_LIST_HEAD(&ci->i_flushing_item);
419 ci->i_cap_flush_seq = 0;
420 ci->i_cap_flush_last_tid = 0;
421 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
422 init_waitqueue_head(&ci->i_cap_wq);
423 ci->i_hold_caps_min = 0;
424 ci->i_hold_caps_max = 0;
425 INIT_LIST_HEAD(&ci->i_cap_delay_list);
426 INIT_LIST_HEAD(&ci->i_cap_snaps);
427 ci->i_head_snapc = NULL;
430 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
431 ci->i_nr_by_mode[i] = 0;
433 mutex_init(&ci->i_truncate_mutex);
434 ci->i_truncate_seq = 0;
435 ci->i_truncate_size = 0;
436 ci->i_truncate_pending = 0;
439 ci->i_reported_size = 0;
440 ci->i_wanted_max_size = 0;
441 ci->i_requested_max_size = 0;
445 ci->i_rdcache_ref = 0;
448 ci->i_wrbuffer_ref = 0;
449 ci->i_wrbuffer_ref_head = 0;
450 ci->i_shared_gen = 0;
451 ci->i_rdcache_gen = 0;
452 ci->i_rdcache_revoking = 0;
454 INIT_LIST_HEAD(&ci->i_unsafe_writes);
455 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
456 spin_lock_init(&ci->i_unsafe_lock);
458 ci->i_snap_realm = NULL;
459 INIT_LIST_HEAD(&ci->i_snap_realm_item);
460 INIT_LIST_HEAD(&ci->i_snap_flush_item);
462 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
463 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
465 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
467 ceph_fscache_inode_init(ci);
469 return &ci->vfs_inode;
472 static void ceph_i_callback(struct rcu_head *head)
474 struct inode *inode = container_of(head, struct inode, i_rcu);
475 struct ceph_inode_info *ci = ceph_inode(inode);
477 kmem_cache_free(ceph_inode_cachep, ci);
480 void ceph_destroy_inode(struct inode *inode)
482 struct ceph_inode_info *ci = ceph_inode(inode);
483 struct ceph_inode_frag *frag;
486 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
488 ceph_fscache_unregister_inode_cookie(ci);
490 ceph_queue_caps_release(inode);
493 * we may still have a snap_realm reference if there are stray
494 * caps in i_snap_caps.
496 if (ci->i_snap_realm) {
497 struct ceph_mds_client *mdsc =
498 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
499 struct ceph_snap_realm *realm = ci->i_snap_realm;
501 dout(" dropping residual ref to snap realm %p\n", realm);
502 spin_lock(&realm->inodes_with_caps_lock);
503 list_del_init(&ci->i_snap_realm_item);
504 spin_unlock(&realm->inodes_with_caps_lock);
505 ceph_put_snap_realm(mdsc, realm);
508 kfree(ci->i_symlink);
509 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
510 frag = rb_entry(n, struct ceph_inode_frag, node);
511 rb_erase(n, &ci->i_fragtree);
515 __ceph_destroy_xattrs(ci);
516 if (ci->i_xattrs.blob)
517 ceph_buffer_put(ci->i_xattrs.blob);
518 if (ci->i_xattrs.prealloc_blob)
519 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
521 call_rcu(&inode->i_rcu, ceph_i_callback);
524 int ceph_drop_inode(struct inode *inode)
527 * Positve dentry and corresponding inode are always accompanied
528 * in MDS reply. So no need to keep inode in the cache after
529 * dropping all its aliases.
535 * Helpers to fill in size, ctime, mtime, and atime. We have to be
536 * careful because either the client or MDS may have more up to date
537 * info, depending on which capabilities are held, and whether
538 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
539 * and size are monotonically increasing, except when utimes() or
540 * truncate() increments the corresponding _seq values.)
542 int ceph_fill_file_size(struct inode *inode, int issued,
543 u32 truncate_seq, u64 truncate_size, u64 size)
545 struct ceph_inode_info *ci = ceph_inode(inode);
548 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
549 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
550 dout("size %lld -> %llu\n", inode->i_size, size);
551 inode->i_size = size;
552 inode->i_blocks = (size + (1<<9) - 1) >> 9;
553 ci->i_reported_size = size;
554 if (truncate_seq != ci->i_truncate_seq) {
555 dout("truncate_seq %u -> %u\n",
556 ci->i_truncate_seq, truncate_seq);
557 ci->i_truncate_seq = truncate_seq;
559 /* the MDS should have revoked these caps */
560 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
563 CEPH_CAP_FILE_LAZYIO));
565 * If we hold relevant caps, or in the case where we're
566 * not the only client referencing this file and we
567 * don't hold those caps, then we need to check whether
568 * the file is either opened or mmaped
570 if ((issued & (CEPH_CAP_FILE_CACHE|
571 CEPH_CAP_FILE_BUFFER)) ||
572 mapping_mapped(inode->i_mapping) ||
573 __ceph_caps_file_wanted(ci)) {
574 ci->i_truncate_pending++;
579 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
580 ci->i_truncate_size != truncate_size) {
581 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
583 ci->i_truncate_size = truncate_size;
587 ceph_fscache_invalidate(inode);
592 void ceph_fill_file_time(struct inode *inode, int issued,
593 u64 time_warp_seq, struct timespec *ctime,
594 struct timespec *mtime, struct timespec *atime)
596 struct ceph_inode_info *ci = ceph_inode(inode);
599 if (issued & (CEPH_CAP_FILE_EXCL|
601 CEPH_CAP_FILE_BUFFER|
603 CEPH_CAP_XATTR_EXCL)) {
604 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
605 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
606 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
607 ctime->tv_sec, ctime->tv_nsec);
608 inode->i_ctime = *ctime;
610 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
611 /* the MDS did a utimes() */
612 dout("mtime %ld.%09ld -> %ld.%09ld "
614 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
615 mtime->tv_sec, mtime->tv_nsec,
616 ci->i_time_warp_seq, (int)time_warp_seq);
618 inode->i_mtime = *mtime;
619 inode->i_atime = *atime;
620 ci->i_time_warp_seq = time_warp_seq;
621 } else if (time_warp_seq == ci->i_time_warp_seq) {
622 /* nobody did utimes(); take the max */
623 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
624 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
625 inode->i_mtime.tv_sec,
626 inode->i_mtime.tv_nsec,
627 mtime->tv_sec, mtime->tv_nsec);
628 inode->i_mtime = *mtime;
630 if (timespec_compare(atime, &inode->i_atime) > 0) {
631 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
632 inode->i_atime.tv_sec,
633 inode->i_atime.tv_nsec,
634 atime->tv_sec, atime->tv_nsec);
635 inode->i_atime = *atime;
637 } else if (issued & CEPH_CAP_FILE_EXCL) {
638 /* we did a utimes(); ignore mds values */
643 /* we have no write|excl caps; whatever the MDS says is true */
644 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
645 inode->i_ctime = *ctime;
646 inode->i_mtime = *mtime;
647 inode->i_atime = *atime;
648 ci->i_time_warp_seq = time_warp_seq;
653 if (warn) /* time_warp_seq shouldn't go backwards */
654 dout("%p mds time_warp_seq %llu < %u\n",
655 inode, time_warp_seq, ci->i_time_warp_seq);
659 * Populate an inode based on info from mds. May be called on new or
662 static int fill_inode(struct inode *inode, struct page *locked_page,
663 struct ceph_mds_reply_info_in *iinfo,
664 struct ceph_mds_reply_dirfrag *dirinfo,
665 struct ceph_mds_session *session,
666 unsigned long ttl_from, int cap_fmode,
667 struct ceph_cap_reservation *caps_reservation)
669 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
670 struct ceph_mds_reply_inode *info = iinfo->in;
671 struct ceph_inode_info *ci = ceph_inode(inode);
672 int issued = 0, implemented, new_issued;
673 struct timespec mtime, atime, ctime;
674 struct ceph_buffer *xattr_blob = NULL;
675 struct ceph_cap *new_cap = NULL;
678 bool queue_trunc = false;
679 bool new_version = false;
680 bool fill_inline = false;
682 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
683 inode, ceph_vinop(inode), le64_to_cpu(info->version),
686 /* prealloc new cap struct */
687 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
688 new_cap = ceph_get_cap(mdsc, caps_reservation);
691 * prealloc xattr data, if it looks like we'll need it. only
692 * if len > 4 (meaning there are actually xattrs; the first 4
693 * bytes are the xattr count).
695 if (iinfo->xattr_len > 4) {
696 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
698 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
702 spin_lock(&ci->i_ceph_lock);
705 * provided version will be odd if inode value is projected,
706 * even if stable. skip the update if we have newer stable
707 * info (ours>=theirs, e.g. due to racing mds replies), unless
708 * we are getting projected (unstable) info (in which case the
709 * version is odd, and we want ours>theirs).
715 if (ci->i_version == 0 ||
716 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
717 le64_to_cpu(info->version) > (ci->i_version & ~1)))
720 issued = __ceph_caps_issued(ci, &implemented);
721 issued |= implemented | __ceph_caps_dirty(ci);
722 new_issued = ~issued & le32_to_cpu(info->cap.caps);
725 ci->i_version = le64_to_cpu(info->version);
727 inode->i_rdev = le32_to_cpu(info->rdev);
728 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
730 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
731 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
732 inode->i_mode = le32_to_cpu(info->mode);
733 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
734 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
735 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
736 from_kuid(&init_user_ns, inode->i_uid),
737 from_kgid(&init_user_ns, inode->i_gid));
740 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
741 (issued & CEPH_CAP_LINK_EXCL) == 0)
742 set_nlink(inode, le32_to_cpu(info->nlink));
744 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
745 /* be careful with mtime, atime, size */
746 ceph_decode_timespec(&atime, &info->atime);
747 ceph_decode_timespec(&mtime, &info->mtime);
748 ceph_decode_timespec(&ctime, &info->ctime);
749 ceph_fill_file_time(inode, issued,
750 le32_to_cpu(info->time_warp_seq),
751 &ctime, &mtime, &atime);
755 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
756 ci->i_layout = info->layout;
757 queue_trunc = ceph_fill_file_size(inode, issued,
758 le32_to_cpu(info->truncate_seq),
759 le64_to_cpu(info->truncate_size),
760 le64_to_cpu(info->size));
761 /* only update max_size on auth cap */
762 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
763 ci->i_max_size != le64_to_cpu(info->max_size)) {
764 dout("max_size %lld -> %llu\n", ci->i_max_size,
765 le64_to_cpu(info->max_size));
766 ci->i_max_size = le64_to_cpu(info->max_size);
771 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
772 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
773 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
774 if (ci->i_xattrs.blob)
775 ceph_buffer_put(ci->i_xattrs.blob);
776 ci->i_xattrs.blob = xattr_blob;
778 memcpy(ci->i_xattrs.blob->vec.iov_base,
779 iinfo->xattr_data, iinfo->xattr_len);
780 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
781 ceph_forget_all_cached_acls(inode);
785 inode->i_mapping->a_ops = &ceph_aops;
786 inode->i_mapping->backing_dev_info =
787 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
789 switch (inode->i_mode & S_IFMT) {
794 init_special_inode(inode, inode->i_mode, inode->i_rdev);
795 inode->i_op = &ceph_file_iops;
798 inode->i_op = &ceph_file_iops;
799 inode->i_fop = &ceph_file_fops;
802 inode->i_op = &ceph_symlink_iops;
803 if (!ci->i_symlink) {
804 u32 symlen = iinfo->symlink_len;
807 spin_unlock(&ci->i_ceph_lock);
810 if (WARN_ON(symlen != inode->i_size))
814 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
818 spin_lock(&ci->i_ceph_lock);
822 kfree(sym); /* lost a race */
826 inode->i_op = &ceph_dir_iops;
827 inode->i_fop = &ceph_dir_fops;
829 ci->i_dir_layout = iinfo->dir_layout;
831 ci->i_files = le64_to_cpu(info->files);
832 ci->i_subdirs = le64_to_cpu(info->subdirs);
833 ci->i_rbytes = le64_to_cpu(info->rbytes);
834 ci->i_rfiles = le64_to_cpu(info->rfiles);
835 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
836 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
839 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
840 ceph_vinop(inode), inode->i_mode);
843 /* set dir completion flag? */
844 if (S_ISDIR(inode->i_mode) &&
845 ci->i_files == 0 && ci->i_subdirs == 0 &&
846 ceph_snap(inode) == CEPH_NOSNAP &&
847 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
848 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
849 !__ceph_dir_is_complete(ci)) {
850 dout(" marking %p complete (empty)\n", inode);
851 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count),
852 ci->i_ordered_count);
855 /* were we issued a capability? */
856 if (info->cap.caps) {
857 if (ceph_snap(inode) == CEPH_NOSNAP) {
858 ceph_add_cap(inode, session,
859 le64_to_cpu(info->cap.cap_id),
861 le32_to_cpu(info->cap.caps),
862 le32_to_cpu(info->cap.wanted),
863 le32_to_cpu(info->cap.seq),
864 le32_to_cpu(info->cap.mseq),
865 le64_to_cpu(info->cap.realm),
866 info->cap.flags, &new_cap);
869 dout(" %p got snap_caps %s\n", inode,
870 ceph_cap_string(le32_to_cpu(info->cap.caps)));
871 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
873 __ceph_get_fmode(ci, cap_fmode);
875 } else if (cap_fmode >= 0) {
876 pr_warn("mds issued no caps on %llx.%llx\n",
878 __ceph_get_fmode(ci, cap_fmode);
881 if (iinfo->inline_version > 0 &&
882 iinfo->inline_version >= ci->i_inline_version) {
883 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
884 ci->i_inline_version = iinfo->inline_version;
885 if (ci->i_inline_version != CEPH_INLINE_NONE &&
887 (le32_to_cpu(info->cap.caps) & cache_caps)))
891 spin_unlock(&ci->i_ceph_lock);
894 ceph_fill_inline_data(inode, locked_page,
895 iinfo->inline_data, iinfo->inline_len);
898 wake_up_all(&ci->i_cap_wq);
900 /* queue truncate if we saw i_size decrease */
902 ceph_queue_vmtruncate(inode);
904 /* populate frag tree */
905 if (S_ISDIR(inode->i_mode))
906 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
908 /* update delegation info? */
910 ceph_fill_dirfrag(inode, dirinfo);
915 ceph_put_cap(mdsc, new_cap);
917 ceph_buffer_put(xattr_blob);
922 * caller should hold session s_mutex.
924 static void update_dentry_lease(struct dentry *dentry,
925 struct ceph_mds_reply_lease *lease,
926 struct ceph_mds_session *session,
927 unsigned long from_time)
929 struct ceph_dentry_info *di = ceph_dentry(dentry);
930 long unsigned duration = le32_to_cpu(lease->duration_ms);
931 long unsigned ttl = from_time + (duration * HZ) / 1000;
932 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
935 /* only track leases on regular dentries */
936 if (dentry->d_op != &ceph_dentry_ops)
939 spin_lock(&dentry->d_lock);
940 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
941 dentry, duration, ttl);
943 /* make lease_rdcache_gen match directory */
944 dir = dentry->d_parent->d_inode;
945 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
950 if (di->lease_gen == session->s_cap_gen &&
951 time_before(ttl, dentry->d_time))
952 goto out_unlock; /* we already have a newer lease. */
954 if (di->lease_session && di->lease_session != session)
957 ceph_dentry_lru_touch(dentry);
959 if (!di->lease_session)
960 di->lease_session = ceph_get_mds_session(session);
961 di->lease_gen = session->s_cap_gen;
962 di->lease_seq = le32_to_cpu(lease->seq);
963 di->lease_renew_after = half_ttl;
964 di->lease_renew_from = 0;
965 dentry->d_time = ttl;
967 spin_unlock(&dentry->d_lock);
972 * splice a dentry to an inode.
973 * caller must hold directory i_mutex for this to be safe.
975 * we will only rehash the resulting dentry if @prehash is
976 * true; @prehash will be set to false (for the benefit of
977 * the caller) if we fail.
979 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
982 struct dentry *realdn;
986 /* dn must be unhashed */
989 realdn = d_splice_alias(in, dn);
990 if (IS_ERR(realdn)) {
991 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
992 PTR_ERR(realdn), dn, in, ceph_vinop(in));
994 *prehash = false; /* don't rehash on error */
995 dn = realdn; /* note realdn contains the error */
998 dout("dn %p (%d) spliced with %p (%d) "
999 "inode %p ino %llx.%llx\n",
1001 realdn, d_count(realdn),
1002 realdn->d_inode, ceph_vinop(realdn->d_inode));
1006 BUG_ON(!ceph_dentry(dn));
1007 dout("dn %p attached to %p ino %llx.%llx\n",
1008 dn, dn->d_inode, ceph_vinop(dn->d_inode));
1010 if ((!prehash || *prehash) && d_unhashed(dn))
1017 * Incorporate results into the local cache. This is either just
1018 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1021 * A reply may contain
1022 * a directory inode along with a dentry.
1023 * and/or a target inode
1025 * Called with snap_rwsem (read).
1027 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1028 struct ceph_mds_session *session)
1030 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1031 struct inode *in = NULL;
1032 struct ceph_vino vino;
1033 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1036 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1037 rinfo->head->is_dentry, rinfo->head->is_target);
1043 * If we resend completed ops to a recovering mds, we get no
1044 * trace. Since that is very rare, pretend this is the case
1045 * to ensure the 'no trace' handlers in the callers behave.
1047 * Fill in inodes unconditionally to avoid breaking cap
1050 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1051 pr_info("fill_trace faking empty trace on %lld %s\n",
1052 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1053 if (rinfo->head->is_dentry) {
1054 rinfo->head->is_dentry = 0;
1055 err = fill_inode(req->r_locked_dir,
1056 &rinfo->diri, rinfo->dirfrag,
1057 session, req->r_request_started, -1);
1059 if (rinfo->head->is_target) {
1060 rinfo->head->is_target = 0;
1061 ininfo = rinfo->targeti.in;
1062 vino.ino = le64_to_cpu(ininfo->ino);
1063 vino.snap = le64_to_cpu(ininfo->snapid);
1064 in = ceph_get_inode(sb, vino);
1065 err = fill_inode(in, &rinfo->targeti, NULL,
1066 session, req->r_request_started,
1073 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1074 dout("fill_trace reply is empty!\n");
1075 if (rinfo->head->result == 0 && req->r_locked_dir)
1076 ceph_invalidate_dir_request(req);
1080 if (rinfo->head->is_dentry) {
1081 struct inode *dir = req->r_locked_dir;
1084 err = fill_inode(dir, NULL,
1085 &rinfo->diri, rinfo->dirfrag,
1086 session, req->r_request_started, -1,
1087 &req->r_caps_reservation);
1094 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1096 struct dentry *dn, *parent;
1098 BUG_ON(!rinfo->head->is_target);
1099 BUG_ON(req->r_dentry);
1101 parent = d_find_any_alias(dir);
1104 dname.name = rinfo->dname;
1105 dname.len = rinfo->dname_len;
1106 dname.hash = full_name_hash(dname.name, dname.len);
1107 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1108 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1110 dn = d_lookup(parent, &dname);
1111 dout("d_lookup on parent=%p name=%.*s got %p\n",
1112 parent, dname.len, dname.name, dn);
1115 dn = d_alloc(parent, &dname);
1116 dout("d_alloc %p '%.*s' = %p\n", parent,
1117 dname.len, dname.name, dn);
1123 err = ceph_init_dentry(dn);
1129 } else if (dn->d_inode &&
1130 (ceph_ino(dn->d_inode) != vino.ino ||
1131 ceph_snap(dn->d_inode) != vino.snap)) {
1132 dout(" dn %p points to wrong inode %p\n",
1144 if (rinfo->head->is_target) {
1145 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1146 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1148 in = ceph_get_inode(sb, vino);
1153 req->r_target_inode = in;
1155 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1156 session, req->r_request_started,
1157 (!req->r_aborted && rinfo->head->result == 0) ?
1159 &req->r_caps_reservation);
1161 pr_err("fill_inode badness %p %llx.%llx\n",
1162 in, ceph_vinop(in));
1168 * ignore null lease/binding on snapdir ENOENT, or else we
1169 * will have trouble splicing in the virtual snapdir later
1171 if (rinfo->head->is_dentry && !req->r_aborted &&
1172 req->r_locked_dir &&
1173 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1174 fsc->mount_options->snapdir_name,
1175 req->r_dentry->d_name.len))) {
1177 * lookup link rename : null -> possibly existing inode
1178 * mknod symlink mkdir : null -> new inode
1179 * unlink : linked -> null
1181 struct inode *dir = req->r_locked_dir;
1182 struct dentry *dn = req->r_dentry;
1183 bool have_dir_cap, have_lease;
1187 BUG_ON(dn->d_parent->d_inode != dir);
1188 BUG_ON(ceph_ino(dir) !=
1189 le64_to_cpu(rinfo->diri.in->ino));
1190 BUG_ON(ceph_snap(dir) !=
1191 le64_to_cpu(rinfo->diri.in->snapid));
1193 /* do we have a lease on the whole dir? */
1195 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1196 CEPH_CAP_FILE_SHARED);
1198 /* do we have a dn lease? */
1199 have_lease = have_dir_cap ||
1200 le32_to_cpu(rinfo->dlease->duration_ms);
1202 dout("fill_trace no dentry lease or dir cap\n");
1205 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1206 struct inode *olddir = req->r_old_dentry_dir;
1209 dout(" src %p '%pd' dst %p '%pd'\n",
1213 dout("fill_trace doing d_move %p -> %p\n",
1214 req->r_old_dentry, dn);
1216 d_move(req->r_old_dentry, dn);
1217 dout(" src %p '%pd' dst %p '%pd'\n",
1222 /* ensure target dentry is invalidated, despite
1223 rehashing bug in vfs_rename_dir */
1224 ceph_invalidate_dentry_lease(dn);
1226 /* d_move screws up sibling dentries' offsets */
1227 ceph_dir_clear_ordered(dir);
1228 ceph_dir_clear_ordered(olddir);
1230 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1231 ceph_dentry(req->r_old_dentry)->offset);
1233 dn = req->r_old_dentry; /* use old_dentry */
1237 if (!rinfo->head->is_target) {
1238 dout("fill_trace null dentry\n");
1240 ceph_dir_clear_ordered(dir);
1241 dout("d_delete %p\n", dn);
1244 dout("d_instantiate %p NULL\n", dn);
1245 d_instantiate(dn, NULL);
1246 if (have_lease && d_unhashed(dn))
1248 update_dentry_lease(dn, rinfo->dlease,
1250 req->r_request_started);
1255 /* attach proper inode */
1257 ceph_dir_clear_ordered(dir);
1259 dn = splice_dentry(dn, in, &have_lease);
1264 req->r_dentry = dn; /* may have spliced */
1265 } else if (dn->d_inode && dn->d_inode != in) {
1266 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1267 dn, dn->d_inode, ceph_vinop(dn->d_inode),
1273 update_dentry_lease(dn, rinfo->dlease, session,
1274 req->r_request_started);
1275 dout(" final dn %p\n", dn);
1276 } else if (!req->r_aborted &&
1277 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1278 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1279 struct dentry *dn = req->r_dentry;
1280 struct inode *dir = req->r_locked_dir;
1282 /* fill out a snapdir LOOKUPSNAP dentry */
1285 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1286 dout(" linking snapped dir %p to dn %p\n", in, dn);
1287 ceph_dir_clear_ordered(dir);
1289 dn = splice_dentry(dn, in, NULL);
1294 req->r_dentry = dn; /* may have spliced */
1297 dout("fill_trace done err=%d\n", err);
1302 * Prepopulate our cache with readdir results, leases, etc.
1304 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1305 struct ceph_mds_session *session)
1307 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1310 for (i = 0; i < rinfo->dir_nr; i++) {
1311 struct ceph_vino vino;
1315 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1316 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1318 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1321 dout("new_inode badness got %d\n", err);
1324 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1325 req->r_request_started, -1,
1326 &req->r_caps_reservation);
1328 pr_err("fill_inode badness on %p got %d\n", in, rc);
1337 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1338 struct ceph_mds_session *session)
1340 struct dentry *parent = req->r_dentry;
1341 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1345 int err = 0, ret, i;
1346 struct inode *snapdir = NULL;
1347 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1348 struct ceph_dentry_info *di;
1349 u64 r_readdir_offset = req->r_readdir_offset;
1350 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1352 if (rinfo->dir_dir &&
1353 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1354 dout("readdir_prepopulate got new frag %x -> %x\n",
1355 frag, le32_to_cpu(rinfo->dir_dir->frag));
1356 frag = le32_to_cpu(rinfo->dir_dir->frag);
1357 if (ceph_frag_is_leftmost(frag))
1358 r_readdir_offset = 2;
1360 r_readdir_offset = 0;
1364 return readdir_prepopulate_inodes_only(req, session);
1366 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1367 snapdir = ceph_get_snapdir(parent->d_inode);
1368 parent = d_find_alias(snapdir);
1369 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1370 rinfo->dir_nr, parent);
1372 dout("readdir_prepopulate %d items under dn %p\n",
1373 rinfo->dir_nr, parent);
1375 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1378 /* FIXME: release caps/leases if error occurs */
1379 for (i = 0; i < rinfo->dir_nr; i++) {
1380 struct ceph_vino vino;
1382 dname.name = rinfo->dir_dname[i];
1383 dname.len = rinfo->dir_dname_len[i];
1384 dname.hash = full_name_hash(dname.name, dname.len);
1386 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1387 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1390 dn = d_lookup(parent, &dname);
1391 dout("d_lookup on parent=%p name=%.*s got %p\n",
1392 parent, dname.len, dname.name, dn);
1395 dn = d_alloc(parent, &dname);
1396 dout("d_alloc %p '%.*s' = %p\n", parent,
1397 dname.len, dname.name, dn);
1399 dout("d_alloc badness\n");
1403 ret = ceph_init_dentry(dn);
1409 } else if (dn->d_inode &&
1410 (ceph_ino(dn->d_inode) != vino.ino ||
1411 ceph_snap(dn->d_inode) != vino.snap)) {
1412 dout(" dn %p points to wrong inode %p\n",
1418 /* reorder parent's d_subdirs */
1419 spin_lock(&parent->d_lock);
1420 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1421 list_move(&dn->d_child, &parent->d_subdirs);
1422 spin_unlock(&dn->d_lock);
1423 spin_unlock(&parent->d_lock);
1430 in = ceph_get_inode(parent->d_sb, vino);
1432 dout("new_inode badness\n");
1440 if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1441 req->r_request_started, -1,
1442 &req->r_caps_reservation) < 0) {
1443 pr_err("fill_inode badness on %p\n", in);
1451 dn = splice_dentry(dn, in, NULL);
1460 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1462 update_dentry_lease(dn, rinfo->dir_dlease[i],
1464 req->r_request_started);
1470 req->r_did_prepopulate = true;
1477 dout("readdir_prepopulate done\n");
1481 int ceph_inode_set_size(struct inode *inode, loff_t size)
1483 struct ceph_inode_info *ci = ceph_inode(inode);
1486 spin_lock(&ci->i_ceph_lock);
1487 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1488 inode->i_size = size;
1489 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1491 /* tell the MDS if we are approaching max_size */
1492 if ((size << 1) >= ci->i_max_size &&
1493 (ci->i_reported_size << 1) < ci->i_max_size)
1496 spin_unlock(&ci->i_ceph_lock);
1501 * Write back inode data in a worker thread. (This can't be done
1502 * in the message handler context.)
1504 void ceph_queue_writeback(struct inode *inode)
1507 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1508 &ceph_inode(inode)->i_wb_work)) {
1509 dout("ceph_queue_writeback %p\n", inode);
1511 dout("ceph_queue_writeback %p failed\n", inode);
1516 static void ceph_writeback_work(struct work_struct *work)
1518 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1520 struct inode *inode = &ci->vfs_inode;
1522 dout("writeback %p\n", inode);
1523 filemap_fdatawrite(&inode->i_data);
1528 * queue an async invalidation
1530 void ceph_queue_invalidate(struct inode *inode)
1533 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1534 &ceph_inode(inode)->i_pg_inv_work)) {
1535 dout("ceph_queue_invalidate %p\n", inode);
1537 dout("ceph_queue_invalidate %p failed\n", inode);
1543 * Invalidate inode pages in a worker thread. (This can't be done
1544 * in the message handler context.)
1546 static void ceph_invalidate_work(struct work_struct *work)
1548 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1550 struct inode *inode = &ci->vfs_inode;
1554 mutex_lock(&ci->i_truncate_mutex);
1555 spin_lock(&ci->i_ceph_lock);
1556 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1557 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1558 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1559 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1561 spin_unlock(&ci->i_ceph_lock);
1562 mutex_unlock(&ci->i_truncate_mutex);
1565 orig_gen = ci->i_rdcache_gen;
1566 spin_unlock(&ci->i_ceph_lock);
1568 truncate_pagecache(inode, 0);
1570 spin_lock(&ci->i_ceph_lock);
1571 if (orig_gen == ci->i_rdcache_gen &&
1572 orig_gen == ci->i_rdcache_revoking) {
1573 dout("invalidate_pages %p gen %d successful\n", inode,
1575 ci->i_rdcache_revoking--;
1578 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1579 inode, orig_gen, ci->i_rdcache_gen,
1580 ci->i_rdcache_revoking);
1581 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1584 spin_unlock(&ci->i_ceph_lock);
1585 mutex_unlock(&ci->i_truncate_mutex);
1588 ceph_check_caps(ci, 0, NULL);
1594 * called by trunc_wq;
1596 * We also truncate in a separate thread as well.
1598 static void ceph_vmtruncate_work(struct work_struct *work)
1600 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1602 struct inode *inode = &ci->vfs_inode;
1604 dout("vmtruncate_work %p\n", inode);
1605 __ceph_do_pending_vmtruncate(inode);
1610 * Queue an async vmtruncate. If we fail to queue work, we will handle
1611 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1613 void ceph_queue_vmtruncate(struct inode *inode)
1615 struct ceph_inode_info *ci = ceph_inode(inode);
1619 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1620 &ci->i_vmtruncate_work)) {
1621 dout("ceph_queue_vmtruncate %p\n", inode);
1623 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1624 inode, ci->i_truncate_pending);
1630 * Make sure any pending truncation is applied before doing anything
1631 * that may depend on it.
1633 void __ceph_do_pending_vmtruncate(struct inode *inode)
1635 struct ceph_inode_info *ci = ceph_inode(inode);
1637 int wrbuffer_refs, finish = 0;
1639 mutex_lock(&ci->i_truncate_mutex);
1641 spin_lock(&ci->i_ceph_lock);
1642 if (ci->i_truncate_pending == 0) {
1643 dout("__do_pending_vmtruncate %p none pending\n", inode);
1644 spin_unlock(&ci->i_ceph_lock);
1645 mutex_unlock(&ci->i_truncate_mutex);
1650 * make sure any dirty snapped pages are flushed before we
1651 * possibly truncate them.. so write AND block!
1653 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1654 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1656 spin_unlock(&ci->i_ceph_lock);
1657 filemap_write_and_wait_range(&inode->i_data, 0,
1658 inode->i_sb->s_maxbytes);
1662 /* there should be no reader or writer */
1663 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1665 to = ci->i_truncate_size;
1666 wrbuffer_refs = ci->i_wrbuffer_ref;
1667 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1668 ci->i_truncate_pending, to);
1669 spin_unlock(&ci->i_ceph_lock);
1671 truncate_pagecache(inode, to);
1673 spin_lock(&ci->i_ceph_lock);
1674 if (to == ci->i_truncate_size) {
1675 ci->i_truncate_pending = 0;
1678 spin_unlock(&ci->i_ceph_lock);
1682 mutex_unlock(&ci->i_truncate_mutex);
1684 if (wrbuffer_refs == 0)
1685 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1687 wake_up_all(&ci->i_cap_wq);
1693 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1695 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1696 nd_set_link(nd, ci->i_symlink);
1700 static const struct inode_operations ceph_symlink_iops = {
1701 .readlink = generic_readlink,
1702 .follow_link = ceph_sym_follow_link,
1703 .setattr = ceph_setattr,
1704 .getattr = ceph_getattr,
1705 .setxattr = ceph_setxattr,
1706 .getxattr = ceph_getxattr,
1707 .listxattr = ceph_listxattr,
1708 .removexattr = ceph_removexattr,
1714 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1716 struct inode *inode = dentry->d_inode;
1717 struct ceph_inode_info *ci = ceph_inode(inode);
1718 const unsigned int ia_valid = attr->ia_valid;
1719 struct ceph_mds_request *req;
1720 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1722 int release = 0, dirtied = 0;
1725 int inode_dirty_flags = 0;
1727 if (ceph_snap(inode) != CEPH_NOSNAP)
1730 err = inode_change_ok(inode, attr);
1734 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1737 return PTR_ERR(req);
1739 spin_lock(&ci->i_ceph_lock);
1740 issued = __ceph_caps_issued(ci, NULL);
1741 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1743 if (ia_valid & ATTR_UID) {
1744 dout("setattr %p uid %d -> %d\n", inode,
1745 from_kuid(&init_user_ns, inode->i_uid),
1746 from_kuid(&init_user_ns, attr->ia_uid));
1747 if (issued & CEPH_CAP_AUTH_EXCL) {
1748 inode->i_uid = attr->ia_uid;
1749 dirtied |= CEPH_CAP_AUTH_EXCL;
1750 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1751 !uid_eq(attr->ia_uid, inode->i_uid)) {
1752 req->r_args.setattr.uid = cpu_to_le32(
1753 from_kuid(&init_user_ns, attr->ia_uid));
1754 mask |= CEPH_SETATTR_UID;
1755 release |= CEPH_CAP_AUTH_SHARED;
1758 if (ia_valid & ATTR_GID) {
1759 dout("setattr %p gid %d -> %d\n", inode,
1760 from_kgid(&init_user_ns, inode->i_gid),
1761 from_kgid(&init_user_ns, attr->ia_gid));
1762 if (issued & CEPH_CAP_AUTH_EXCL) {
1763 inode->i_gid = attr->ia_gid;
1764 dirtied |= CEPH_CAP_AUTH_EXCL;
1765 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1766 !gid_eq(attr->ia_gid, inode->i_gid)) {
1767 req->r_args.setattr.gid = cpu_to_le32(
1768 from_kgid(&init_user_ns, attr->ia_gid));
1769 mask |= CEPH_SETATTR_GID;
1770 release |= CEPH_CAP_AUTH_SHARED;
1773 if (ia_valid & ATTR_MODE) {
1774 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1776 if (issued & CEPH_CAP_AUTH_EXCL) {
1777 inode->i_mode = attr->ia_mode;
1778 dirtied |= CEPH_CAP_AUTH_EXCL;
1779 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1780 attr->ia_mode != inode->i_mode) {
1781 inode->i_mode = attr->ia_mode;
1782 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1783 mask |= CEPH_SETATTR_MODE;
1784 release |= CEPH_CAP_AUTH_SHARED;
1788 if (ia_valid & ATTR_ATIME) {
1789 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1790 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1791 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1792 if (issued & CEPH_CAP_FILE_EXCL) {
1793 ci->i_time_warp_seq++;
1794 inode->i_atime = attr->ia_atime;
1795 dirtied |= CEPH_CAP_FILE_EXCL;
1796 } else if ((issued & CEPH_CAP_FILE_WR) &&
1797 timespec_compare(&inode->i_atime,
1798 &attr->ia_atime) < 0) {
1799 inode->i_atime = attr->ia_atime;
1800 dirtied |= CEPH_CAP_FILE_WR;
1801 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1802 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1803 ceph_encode_timespec(&req->r_args.setattr.atime,
1805 mask |= CEPH_SETATTR_ATIME;
1806 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1810 if (ia_valid & ATTR_MTIME) {
1811 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1812 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1813 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1814 if (issued & CEPH_CAP_FILE_EXCL) {
1815 ci->i_time_warp_seq++;
1816 inode->i_mtime = attr->ia_mtime;
1817 dirtied |= CEPH_CAP_FILE_EXCL;
1818 } else if ((issued & CEPH_CAP_FILE_WR) &&
1819 timespec_compare(&inode->i_mtime,
1820 &attr->ia_mtime) < 0) {
1821 inode->i_mtime = attr->ia_mtime;
1822 dirtied |= CEPH_CAP_FILE_WR;
1823 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1824 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1825 ceph_encode_timespec(&req->r_args.setattr.mtime,
1827 mask |= CEPH_SETATTR_MTIME;
1828 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1832 if (ia_valid & ATTR_SIZE) {
1833 dout("setattr %p size %lld -> %lld\n", inode,
1834 inode->i_size, attr->ia_size);
1835 if ((issued & CEPH_CAP_FILE_EXCL) &&
1836 attr->ia_size > inode->i_size) {
1837 inode->i_size = attr->ia_size;
1839 (attr->ia_size + (1 << 9) - 1) >> 9;
1840 inode->i_ctime = attr->ia_ctime;
1841 ci->i_reported_size = attr->ia_size;
1842 dirtied |= CEPH_CAP_FILE_EXCL;
1843 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1844 attr->ia_size != inode->i_size) {
1845 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1846 req->r_args.setattr.old_size =
1847 cpu_to_le64(inode->i_size);
1848 mask |= CEPH_SETATTR_SIZE;
1849 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1854 /* these do nothing */
1855 if (ia_valid & ATTR_CTIME) {
1856 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1857 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1858 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1859 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1860 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1861 only ? "ctime only" : "ignored");
1862 inode->i_ctime = attr->ia_ctime;
1865 * if kernel wants to dirty ctime but nothing else,
1866 * we need to choose a cap to dirty under, or do
1867 * a almost-no-op setattr
1869 if (issued & CEPH_CAP_AUTH_EXCL)
1870 dirtied |= CEPH_CAP_AUTH_EXCL;
1871 else if (issued & CEPH_CAP_FILE_EXCL)
1872 dirtied |= CEPH_CAP_FILE_EXCL;
1873 else if (issued & CEPH_CAP_XATTR_EXCL)
1874 dirtied |= CEPH_CAP_XATTR_EXCL;
1876 mask |= CEPH_SETATTR_CTIME;
1879 if (ia_valid & ATTR_FILE)
1880 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1883 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1884 inode->i_ctime = CURRENT_TIME;
1888 spin_unlock(&ci->i_ceph_lock);
1890 if (inode_dirty_flags)
1891 __mark_inode_dirty(inode, inode_dirty_flags);
1893 if (ia_valid & ATTR_MODE) {
1894 err = posix_acl_chmod(inode, attr->ia_mode);
1900 req->r_inode = inode;
1902 req->r_inode_drop = release;
1903 req->r_args.setattr.mask = cpu_to_le32(mask);
1904 req->r_num_caps = 1;
1905 err = ceph_mdsc_do_request(mdsc, NULL, req);
1907 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1908 ceph_cap_string(dirtied), mask);
1910 ceph_mdsc_put_request(req);
1911 if (mask & CEPH_SETATTR_SIZE)
1912 __ceph_do_pending_vmtruncate(inode);
1915 ceph_mdsc_put_request(req);
1920 * Verify that we have a lease on the given mask. If not,
1921 * do a getattr against an mds.
1923 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1924 int mask, bool force)
1926 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1927 struct ceph_mds_client *mdsc = fsc->mdsc;
1928 struct ceph_mds_request *req;
1931 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1932 dout("do_getattr inode %p SNAPDIR\n", inode);
1936 dout("do_getattr inode %p mask %s mode 0%o\n",
1937 inode, ceph_cap_string(mask), inode->i_mode);
1938 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1941 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1943 return PTR_ERR(req);
1944 req->r_inode = inode;
1946 req->r_num_caps = 1;
1947 req->r_args.getattr.mask = cpu_to_le32(mask);
1948 req->r_locked_page = locked_page;
1949 err = ceph_mdsc_do_request(mdsc, NULL, req);
1950 if (locked_page && err == 0) {
1951 u64 inline_version = req->r_reply_info.targeti.inline_version;
1952 if (inline_version == 0) {
1953 /* the reply is supposed to contain inline data */
1955 } else if (inline_version == CEPH_INLINE_NONE) {
1958 err = req->r_reply_info.targeti.inline_len;
1961 ceph_mdsc_put_request(req);
1962 dout("do_getattr result=%d\n", err);
1968 * Check inode permissions. We verify we have a valid value for
1969 * the AUTH cap, then call the generic handler.
1971 int ceph_permission(struct inode *inode, int mask)
1975 if (mask & MAY_NOT_BLOCK)
1978 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
1981 err = generic_permission(inode, mask);
1986 * Get all attributes. Hopefully somedata we'll have a statlite()
1987 * and can limit the fields we require to be accurate.
1989 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1992 struct inode *inode = dentry->d_inode;
1993 struct ceph_inode_info *ci = ceph_inode(inode);
1996 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
1998 generic_fillattr(inode, stat);
1999 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2000 if (ceph_snap(inode) != CEPH_NOSNAP)
2001 stat->dev = ceph_snap(inode);
2004 if (S_ISDIR(inode->i_mode)) {
2005 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2007 stat->size = ci->i_rbytes;
2009 stat->size = ci->i_files + ci->i_subdirs;
2011 stat->blksize = 65536;