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>
15 #include "mds_client.h"
17 #include <linux/ceph/decode.h>
20 * Ceph inode operations
22 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
23 * setattr, etc.), xattr helpers, and helpers for assimilating
24 * metadata returned by the MDS into our cache.
26 * Also define helpers for doing asynchronous writeback, invalidation,
27 * and truncation for the benefit of those who can't afford to block
28 * (typically because they are in the message handler path).
31 static const struct inode_operations ceph_symlink_iops;
33 static void ceph_invalidate_work(struct work_struct *work);
34 static void ceph_writeback_work(struct work_struct *work);
35 static void ceph_vmtruncate_work(struct work_struct *work);
38 * find or create an inode, given the ceph ino number
40 static int ceph_set_ino_cb(struct inode *inode, void *data)
42 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
43 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
47 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
50 ino_t t = ceph_vino_to_ino(vino);
52 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
54 return ERR_PTR(-ENOMEM);
55 if (inode->i_state & I_NEW) {
56 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
57 inode, ceph_vinop(inode), (u64)inode->i_ino);
58 unlock_new_inode(inode);
61 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
67 * get/constuct snapdir inode for a given directory
69 struct inode *ceph_get_snapdir(struct inode *parent)
71 struct ceph_vino vino = {
72 .ino = ceph_ino(parent),
75 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
76 struct ceph_inode_info *ci = ceph_inode(inode);
78 BUG_ON(!S_ISDIR(parent->i_mode));
81 inode->i_mode = parent->i_mode;
82 inode->i_uid = parent->i_uid;
83 inode->i_gid = parent->i_gid;
84 inode->i_op = &ceph_snapdir_iops;
85 inode->i_fop = &ceph_snapdir_fops;
86 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
91 const struct inode_operations ceph_file_iops = {
92 .permission = ceph_permission,
93 .setattr = ceph_setattr,
94 .getattr = ceph_getattr,
95 .setxattr = ceph_setxattr,
96 .getxattr = ceph_getxattr,
97 .listxattr = ceph_listxattr,
98 .removexattr = ceph_removexattr,
99 .get_acl = ceph_get_acl,
100 .set_acl = ceph_set_acl,
105 * We use a 'frag tree' to keep track of the MDS's directory fragments
106 * for a given inode (usually there is just a single fragment). We
107 * need to know when a child frag is delegated to a new MDS, or when
108 * it is flagged as replicated, so we can direct our requests
113 * find/create a frag in the tree
115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
119 struct rb_node *parent = NULL;
120 struct ceph_inode_frag *frag;
123 p = &ci->i_fragtree.rb_node;
126 frag = rb_entry(parent, struct ceph_inode_frag, node);
127 c = ceph_frag_compare(f, frag->frag);
136 frag = kmalloc(sizeof(*frag), GFP_NOFS);
138 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
139 "frag %x\n", &ci->vfs_inode,
140 ceph_vinop(&ci->vfs_inode), f);
141 return ERR_PTR(-ENOMEM);
148 rb_link_node(&frag->node, parent, p);
149 rb_insert_color(&frag->node, &ci->i_fragtree);
151 dout("get_or_create_frag added %llx.%llx frag %x\n",
152 ceph_vinop(&ci->vfs_inode), f);
157 * find a specific frag @f
159 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
161 struct rb_node *n = ci->i_fragtree.rb_node;
164 struct ceph_inode_frag *frag =
165 rb_entry(n, struct ceph_inode_frag, node);
166 int c = ceph_frag_compare(f, frag->frag);
178 * Choose frag containing the given value @v. If @pfrag is
179 * specified, copy the frag delegation info to the caller if
182 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
183 struct ceph_inode_frag *pfrag, int *found)
185 u32 t = ceph_frag_make(0, 0);
186 struct ceph_inode_frag *frag;
194 WARN_ON(!ceph_frag_contains_value(t, v));
195 frag = __ceph_find_frag(ci, t);
197 break; /* t is a leaf */
198 if (frag->split_by == 0) {
200 memcpy(pfrag, frag, sizeof(*pfrag));
207 nway = 1 << frag->split_by;
208 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
209 frag->split_by, nway);
210 for (i = 0; i < nway; i++) {
211 n = ceph_frag_make_child(t, frag->split_by, i);
212 if (ceph_frag_contains_value(n, v)) {
219 dout("choose_frag(%x) = %x\n", v, t);
224 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
225 struct ceph_inode_frag *pfrag, int *found)
228 mutex_lock(&ci->i_fragtree_mutex);
229 ret = __ceph_choose_frag(ci, v, pfrag, found);
230 mutex_unlock(&ci->i_fragtree_mutex);
235 * Process dirfrag (delegation) info from the mds. Include leaf
236 * fragment in tree ONLY if ndist > 0. Otherwise, only
237 * branches/splits are included in i_fragtree)
239 static int ceph_fill_dirfrag(struct inode *inode,
240 struct ceph_mds_reply_dirfrag *dirinfo)
242 struct ceph_inode_info *ci = ceph_inode(inode);
243 struct ceph_inode_frag *frag;
244 u32 id = le32_to_cpu(dirinfo->frag);
245 int mds = le32_to_cpu(dirinfo->auth);
246 int ndist = le32_to_cpu(dirinfo->ndist);
251 spin_lock(&ci->i_ceph_lock);
253 diri_auth = ci->i_auth_cap->mds;
254 spin_unlock(&ci->i_ceph_lock);
256 mutex_lock(&ci->i_fragtree_mutex);
257 if (ndist == 0 && mds == diri_auth) {
258 /* no delegation info needed. */
259 frag = __ceph_find_frag(ci, id);
262 if (frag->split_by == 0) {
263 /* tree leaf, remove */
264 dout("fill_dirfrag removed %llx.%llx frag %x"
265 " (no ref)\n", ceph_vinop(inode), id);
266 rb_erase(&frag->node, &ci->i_fragtree);
269 /* tree branch, keep and clear */
270 dout("fill_dirfrag cleared %llx.%llx frag %x"
271 " referral\n", ceph_vinop(inode), id);
279 /* find/add this frag to store mds delegation info */
280 frag = __get_or_create_frag(ci, id);
282 /* this is not the end of the world; we can continue
283 with bad/inaccurate delegation info */
284 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
285 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
291 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
292 for (i = 0; i < frag->ndist; i++)
293 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
294 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
295 ceph_vinop(inode), frag->frag, frag->ndist);
298 mutex_unlock(&ci->i_fragtree_mutex);
302 static int ceph_fill_fragtree(struct inode *inode,
303 struct ceph_frag_tree_head *fragtree,
304 struct ceph_mds_reply_dirfrag *dirinfo)
306 struct ceph_inode_info *ci = ceph_inode(inode);
307 struct ceph_inode_frag *frag;
308 struct rb_node *rb_node;
313 mutex_lock(&ci->i_fragtree_mutex);
314 nsplits = le32_to_cpu(fragtree->nsplits);
316 i = prandom_u32() % nsplits;
317 id = le32_to_cpu(fragtree->splits[i].frag);
318 if (!__ceph_find_frag(ci, id))
320 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
321 rb_node = rb_first(&ci->i_fragtree);
322 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
323 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
326 if (!update && dirinfo) {
327 id = le32_to_cpu(dirinfo->frag);
328 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
334 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
335 rb_node = rb_first(&ci->i_fragtree);
336 for (i = 0; i < nsplits; i++) {
337 id = le32_to_cpu(fragtree->splits[i].frag);
340 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
341 if (ceph_frag_compare(frag->frag, id) >= 0) {
342 if (frag->frag != id)
345 rb_node = rb_next(rb_node);
348 rb_node = rb_next(rb_node);
349 rb_erase(&frag->node, &ci->i_fragtree);
354 frag = __get_or_create_frag(ci, id);
358 frag->split_by = le32_to_cpu(fragtree->splits[i].by);
359 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
362 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
363 rb_node = rb_next(rb_node);
364 rb_erase(&frag->node, &ci->i_fragtree);
368 mutex_unlock(&ci->i_fragtree_mutex);
373 * initialize a newly allocated inode.
375 struct inode *ceph_alloc_inode(struct super_block *sb)
377 struct ceph_inode_info *ci;
380 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
384 dout("alloc_inode %p\n", &ci->vfs_inode);
386 spin_lock_init(&ci->i_ceph_lock);
389 ci->i_inline_version = 0;
390 ci->i_time_warp_seq = 0;
391 ci->i_ceph_flags = 0;
392 atomic64_set(&ci->i_ordered_count, 1);
393 atomic64_set(&ci->i_release_count, 1);
394 atomic64_set(&ci->i_complete_seq[0], 0);
395 atomic64_set(&ci->i_complete_seq[1], 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_prealloc_cap_flush = NULL;
420 ci->i_cap_flush_tree = RB_ROOT;
421 init_waitqueue_head(&ci->i_cap_wq);
422 ci->i_hold_caps_min = 0;
423 ci->i_hold_caps_max = 0;
424 INIT_LIST_HEAD(&ci->i_cap_delay_list);
425 INIT_LIST_HEAD(&ci->i_cap_snaps);
426 ci->i_head_snapc = NULL;
429 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
430 ci->i_nr_by_mode[i] = 0;
432 mutex_init(&ci->i_truncate_mutex);
433 ci->i_truncate_seq = 0;
434 ci->i_truncate_size = 0;
435 ci->i_truncate_pending = 0;
438 ci->i_reported_size = 0;
439 ci->i_wanted_max_size = 0;
440 ci->i_requested_max_size = 0;
444 ci->i_rdcache_ref = 0;
447 ci->i_wrbuffer_ref = 0;
448 ci->i_wrbuffer_ref_head = 0;
449 ci->i_shared_gen = 0;
450 ci->i_rdcache_gen = 0;
451 ci->i_rdcache_revoking = 0;
453 INIT_LIST_HEAD(&ci->i_unsafe_writes);
454 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
455 INIT_LIST_HEAD(&ci->i_unsafe_iops);
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 i_size_write(inode, 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 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
757 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
758 ci->i_layout = info->layout;
760 queue_trunc = ceph_fill_file_size(inode, issued,
761 le32_to_cpu(info->truncate_seq),
762 le64_to_cpu(info->truncate_size),
763 le64_to_cpu(info->size));
764 /* only update max_size on auth cap */
765 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
766 ci->i_max_size != le64_to_cpu(info->max_size)) {
767 dout("max_size %lld -> %llu\n", ci->i_max_size,
768 le64_to_cpu(info->max_size));
769 ci->i_max_size = le64_to_cpu(info->max_size);
774 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
775 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
776 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
777 if (ci->i_xattrs.blob)
778 ceph_buffer_put(ci->i_xattrs.blob);
779 ci->i_xattrs.blob = xattr_blob;
781 memcpy(ci->i_xattrs.blob->vec.iov_base,
782 iinfo->xattr_data, iinfo->xattr_len);
783 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
784 ceph_forget_all_cached_acls(inode);
788 inode->i_mapping->a_ops = &ceph_aops;
790 switch (inode->i_mode & S_IFMT) {
795 init_special_inode(inode, inode->i_mode, inode->i_rdev);
796 inode->i_op = &ceph_file_iops;
799 inode->i_op = &ceph_file_iops;
800 inode->i_fop = &ceph_file_fops;
803 inode->i_op = &ceph_symlink_iops;
804 if (!ci->i_symlink) {
805 u32 symlen = iinfo->symlink_len;
808 spin_unlock(&ci->i_ceph_lock);
811 if (WARN_ON(symlen != i_size_read(inode)))
815 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
819 spin_lock(&ci->i_ceph_lock);
823 kfree(sym); /* lost a race */
825 inode->i_link = ci->i_symlink;
828 inode->i_op = &ceph_dir_iops;
829 inode->i_fop = &ceph_dir_fops;
831 ci->i_dir_layout = iinfo->dir_layout;
833 ci->i_files = le64_to_cpu(info->files);
834 ci->i_subdirs = le64_to_cpu(info->subdirs);
835 ci->i_rbytes = le64_to_cpu(info->rbytes);
836 ci->i_rfiles = le64_to_cpu(info->rfiles);
837 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
838 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
841 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
842 ceph_vinop(inode), inode->i_mode);
845 /* were we issued a capability? */
846 if (info->cap.caps) {
847 if (ceph_snap(inode) == CEPH_NOSNAP) {
848 unsigned caps = le32_to_cpu(info->cap.caps);
849 ceph_add_cap(inode, session,
850 le64_to_cpu(info->cap.cap_id),
852 le32_to_cpu(info->cap.wanted),
853 le32_to_cpu(info->cap.seq),
854 le32_to_cpu(info->cap.mseq),
855 le64_to_cpu(info->cap.realm),
856 info->cap.flags, &new_cap);
858 /* set dir completion flag? */
859 if (S_ISDIR(inode->i_mode) &&
860 ci->i_files == 0 && ci->i_subdirs == 0 &&
861 (caps & CEPH_CAP_FILE_SHARED) &&
862 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
863 !__ceph_dir_is_complete(ci)) {
864 dout(" marking %p complete (empty)\n", inode);
865 i_size_write(inode, 0);
866 __ceph_dir_set_complete(ci,
867 atomic64_read(&ci->i_release_count),
868 atomic64_read(&ci->i_ordered_count));
873 dout(" %p got snap_caps %s\n", inode,
874 ceph_cap_string(le32_to_cpu(info->cap.caps)));
875 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
877 __ceph_get_fmode(ci, cap_fmode);
879 } else if (cap_fmode >= 0) {
880 pr_warn("mds issued no caps on %llx.%llx\n",
882 __ceph_get_fmode(ci, cap_fmode);
885 if (iinfo->inline_version > 0 &&
886 iinfo->inline_version >= ci->i_inline_version) {
887 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
888 ci->i_inline_version = iinfo->inline_version;
889 if (ci->i_inline_version != CEPH_INLINE_NONE &&
891 (le32_to_cpu(info->cap.caps) & cache_caps)))
895 spin_unlock(&ci->i_ceph_lock);
898 ceph_fill_inline_data(inode, locked_page,
899 iinfo->inline_data, iinfo->inline_len);
902 wake_up_all(&ci->i_cap_wq);
904 /* queue truncate if we saw i_size decrease */
906 ceph_queue_vmtruncate(inode);
908 /* populate frag tree */
909 if (S_ISDIR(inode->i_mode))
910 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
912 /* update delegation info? */
914 ceph_fill_dirfrag(inode, dirinfo);
919 ceph_put_cap(mdsc, new_cap);
921 ceph_buffer_put(xattr_blob);
926 * caller should hold session s_mutex.
928 static void update_dentry_lease(struct dentry *dentry,
929 struct ceph_mds_reply_lease *lease,
930 struct ceph_mds_session *session,
931 unsigned long from_time)
933 struct ceph_dentry_info *di = ceph_dentry(dentry);
934 long unsigned duration = le32_to_cpu(lease->duration_ms);
935 long unsigned ttl = from_time + (duration * HZ) / 1000;
936 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
939 /* only track leases on regular dentries */
940 if (dentry->d_op != &ceph_dentry_ops)
943 spin_lock(&dentry->d_lock);
944 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
945 dentry, duration, ttl);
947 /* make lease_rdcache_gen match directory */
948 dir = d_inode(dentry->d_parent);
949 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
954 if (di->lease_gen == session->s_cap_gen &&
955 time_before(ttl, dentry->d_time))
956 goto out_unlock; /* we already have a newer lease. */
958 if (di->lease_session && di->lease_session != session)
961 ceph_dentry_lru_touch(dentry);
963 if (!di->lease_session)
964 di->lease_session = ceph_get_mds_session(session);
965 di->lease_gen = session->s_cap_gen;
966 di->lease_seq = le32_to_cpu(lease->seq);
967 di->lease_renew_after = half_ttl;
968 di->lease_renew_from = 0;
969 dentry->d_time = ttl;
971 spin_unlock(&dentry->d_lock);
976 * splice a dentry to an inode.
977 * caller must hold directory i_mutex for this to be safe.
979 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
981 struct dentry *realdn;
985 /* dn must be unhashed */
988 realdn = d_splice_alias(in, dn);
989 if (IS_ERR(realdn)) {
990 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
991 PTR_ERR(realdn), dn, in, ceph_vinop(in));
992 dn = realdn; /* note realdn contains the error */
995 dout("dn %p (%d) spliced with %p (%d) "
996 "inode %p ino %llx.%llx\n",
998 realdn, d_count(realdn),
999 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1003 BUG_ON(!ceph_dentry(dn));
1004 dout("dn %p attached to %p ino %llx.%llx\n",
1005 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1012 * Incorporate results into the local cache. This is either just
1013 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1016 * A reply may contain
1017 * a directory inode along with a dentry.
1018 * and/or a target inode
1020 * Called with snap_rwsem (read).
1022 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1023 struct ceph_mds_session *session)
1025 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1026 struct inode *in = NULL;
1027 struct ceph_vino vino;
1028 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1031 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1032 rinfo->head->is_dentry, rinfo->head->is_target);
1038 * If we resend completed ops to a recovering mds, we get no
1039 * trace. Since that is very rare, pretend this is the case
1040 * to ensure the 'no trace' handlers in the callers behave.
1042 * Fill in inodes unconditionally to avoid breaking cap
1045 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1046 pr_info("fill_trace faking empty trace on %lld %s\n",
1047 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1048 if (rinfo->head->is_dentry) {
1049 rinfo->head->is_dentry = 0;
1050 err = fill_inode(req->r_locked_dir,
1051 &rinfo->diri, rinfo->dirfrag,
1052 session, req->r_request_started, -1);
1054 if (rinfo->head->is_target) {
1055 rinfo->head->is_target = 0;
1056 ininfo = rinfo->targeti.in;
1057 vino.ino = le64_to_cpu(ininfo->ino);
1058 vino.snap = le64_to_cpu(ininfo->snapid);
1059 in = ceph_get_inode(sb, vino);
1060 err = fill_inode(in, &rinfo->targeti, NULL,
1061 session, req->r_request_started,
1068 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1069 dout("fill_trace reply is empty!\n");
1070 if (rinfo->head->result == 0 && req->r_locked_dir)
1071 ceph_invalidate_dir_request(req);
1075 if (rinfo->head->is_dentry) {
1076 struct inode *dir = req->r_locked_dir;
1079 err = fill_inode(dir, NULL,
1080 &rinfo->diri, rinfo->dirfrag,
1081 session, req->r_request_started, -1,
1082 &req->r_caps_reservation);
1089 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1091 struct dentry *dn, *parent;
1093 BUG_ON(!rinfo->head->is_target);
1094 BUG_ON(req->r_dentry);
1096 parent = d_find_any_alias(dir);
1099 dname.name = rinfo->dname;
1100 dname.len = rinfo->dname_len;
1101 dname.hash = full_name_hash(dname.name, dname.len);
1102 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1103 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1105 dn = d_lookup(parent, &dname);
1106 dout("d_lookup on parent=%p name=%.*s got %p\n",
1107 parent, dname.len, dname.name, dn);
1110 dn = d_alloc(parent, &dname);
1111 dout("d_alloc %p '%.*s' = %p\n", parent,
1112 dname.len, dname.name, dn);
1118 err = ceph_init_dentry(dn);
1124 } else if (d_really_is_positive(dn) &&
1125 (ceph_ino(d_inode(dn)) != vino.ino ||
1126 ceph_snap(d_inode(dn)) != vino.snap)) {
1127 dout(" dn %p points to wrong inode %p\n",
1139 if (rinfo->head->is_target) {
1140 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1141 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1143 in = ceph_get_inode(sb, vino);
1148 req->r_target_inode = in;
1150 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1151 session, req->r_request_started,
1152 (!req->r_aborted && rinfo->head->result == 0) ?
1154 &req->r_caps_reservation);
1156 pr_err("fill_inode badness %p %llx.%llx\n",
1157 in, ceph_vinop(in));
1163 * ignore null lease/binding on snapdir ENOENT, or else we
1164 * will have trouble splicing in the virtual snapdir later
1166 if (rinfo->head->is_dentry && !req->r_aborted &&
1167 req->r_locked_dir &&
1168 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1169 fsc->mount_options->snapdir_name,
1170 req->r_dentry->d_name.len))) {
1172 * lookup link rename : null -> possibly existing inode
1173 * mknod symlink mkdir : null -> new inode
1174 * unlink : linked -> null
1176 struct inode *dir = req->r_locked_dir;
1177 struct dentry *dn = req->r_dentry;
1178 bool have_dir_cap, have_lease;
1182 BUG_ON(d_inode(dn->d_parent) != dir);
1183 BUG_ON(ceph_ino(dir) !=
1184 le64_to_cpu(rinfo->diri.in->ino));
1185 BUG_ON(ceph_snap(dir) !=
1186 le64_to_cpu(rinfo->diri.in->snapid));
1188 /* do we have a lease on the whole dir? */
1190 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1191 CEPH_CAP_FILE_SHARED);
1193 /* do we have a dn lease? */
1194 have_lease = have_dir_cap ||
1195 le32_to_cpu(rinfo->dlease->duration_ms);
1197 dout("fill_trace no dentry lease or dir cap\n");
1200 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1201 struct inode *olddir = req->r_old_dentry_dir;
1204 dout(" src %p '%pd' dst %p '%pd'\n",
1208 dout("fill_trace doing d_move %p -> %p\n",
1209 req->r_old_dentry, dn);
1211 /* d_move screws up sibling dentries' offsets */
1212 ceph_dir_clear_ordered(dir);
1213 ceph_dir_clear_ordered(olddir);
1215 d_move(req->r_old_dentry, dn);
1216 dout(" src %p '%pd' dst %p '%pd'\n",
1221 /* ensure target dentry is invalidated, despite
1222 rehashing bug in vfs_rename_dir */
1223 ceph_invalidate_dentry_lease(dn);
1225 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1226 ceph_dentry(req->r_old_dentry)->offset);
1228 dn = req->r_old_dentry; /* use old_dentry */
1232 if (!rinfo->head->is_target) {
1233 dout("fill_trace null dentry\n");
1234 if (d_really_is_positive(dn)) {
1235 ceph_dir_clear_ordered(dir);
1236 dout("d_delete %p\n", dn);
1239 if (have_lease && d_unhashed(dn))
1241 update_dentry_lease(dn, rinfo->dlease,
1243 req->r_request_started);
1248 /* attach proper inode */
1249 if (d_really_is_negative(dn)) {
1250 ceph_dir_clear_ordered(dir);
1252 dn = splice_dentry(dn, in);
1257 req->r_dentry = dn; /* may have spliced */
1258 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1259 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1260 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1266 update_dentry_lease(dn, rinfo->dlease, session,
1267 req->r_request_started);
1268 dout(" final dn %p\n", dn);
1269 } else if (!req->r_aborted &&
1270 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1271 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1272 struct dentry *dn = req->r_dentry;
1273 struct inode *dir = req->r_locked_dir;
1275 /* fill out a snapdir LOOKUPSNAP dentry */
1278 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1279 dout(" linking snapped dir %p to dn %p\n", in, dn);
1280 ceph_dir_clear_ordered(dir);
1282 dn = splice_dentry(dn, in);
1287 req->r_dentry = dn; /* may have spliced */
1290 dout("fill_trace done err=%d\n", err);
1295 * Prepopulate our cache with readdir results, leases, etc.
1297 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1298 struct ceph_mds_session *session)
1300 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1303 for (i = 0; i < rinfo->dir_nr; i++) {
1304 struct ceph_vino vino;
1308 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1309 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1311 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1314 dout("new_inode badness got %d\n", err);
1317 rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1318 req->r_request_started, -1,
1319 &req->r_caps_reservation);
1321 pr_err("fill_inode badness on %p got %d\n", in, rc);
1330 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1334 page_cache_release(ctl->page);
1339 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1340 struct ceph_readdir_cache_control *ctl,
1341 struct ceph_mds_request *req)
1343 struct ceph_inode_info *ci = ceph_inode(dir);
1344 unsigned nsize = PAGE_CACHE_SIZE / sizeof(struct dentry*);
1345 unsigned idx = ctl->index % nsize;
1346 pgoff_t pgoff = ctl->index / nsize;
1348 if (!ctl->page || pgoff != page_index(ctl->page)) {
1349 ceph_readdir_cache_release(ctl);
1350 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1355 /* reading/filling the cache are serialized by
1356 * i_mutex, no need to use page lock */
1357 unlock_page(ctl->page);
1358 ctl->dentries = kmap(ctl->page);
1361 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1362 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1363 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1364 ctl->dentries[idx] = dn;
1367 dout("disable readdir cache\n");
1373 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1374 struct ceph_mds_session *session)
1376 struct dentry *parent = req->r_dentry;
1377 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1381 int err = 0, ret, i;
1382 struct inode *snapdir = NULL;
1383 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1384 struct ceph_dentry_info *di;
1385 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1386 struct ceph_readdir_cache_control cache_ctl = {};
1389 return readdir_prepopulate_inodes_only(req, session);
1391 if (rinfo->dir_dir &&
1392 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1393 dout("readdir_prepopulate got new frag %x -> %x\n",
1394 frag, le32_to_cpu(rinfo->dir_dir->frag));
1395 frag = le32_to_cpu(rinfo->dir_dir->frag);
1396 if (ceph_frag_is_leftmost(frag))
1397 req->r_readdir_offset = 2;
1399 req->r_readdir_offset = 0;
1402 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1403 snapdir = ceph_get_snapdir(d_inode(parent));
1404 parent = d_find_alias(snapdir);
1405 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1406 rinfo->dir_nr, parent);
1408 dout("readdir_prepopulate %d items under dn %p\n",
1409 rinfo->dir_nr, parent);
1411 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1414 if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1415 /* note dir version at start of readdir so we can tell
1416 * if any dentries get dropped */
1417 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1418 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1419 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1420 req->r_readdir_cache_idx = 0;
1423 cache_ctl.index = req->r_readdir_cache_idx;
1425 /* FIXME: release caps/leases if error occurs */
1426 for (i = 0; i < rinfo->dir_nr; i++) {
1427 struct ceph_vino vino;
1429 dname.name = rinfo->dir_dname[i];
1430 dname.len = rinfo->dir_dname_len[i];
1431 dname.hash = full_name_hash(dname.name, dname.len);
1433 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1434 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1437 dn = d_lookup(parent, &dname);
1438 dout("d_lookup on parent=%p name=%.*s got %p\n",
1439 parent, dname.len, dname.name, dn);
1442 dn = d_alloc(parent, &dname);
1443 dout("d_alloc %p '%.*s' = %p\n", parent,
1444 dname.len, dname.name, dn);
1446 dout("d_alloc badness\n");
1450 ret = ceph_init_dentry(dn);
1456 } else if (d_really_is_positive(dn) &&
1457 (ceph_ino(d_inode(dn)) != vino.ino ||
1458 ceph_snap(d_inode(dn)) != vino.snap)) {
1459 dout(" dn %p points to wrong inode %p\n",
1467 if (d_really_is_positive(dn)) {
1470 in = ceph_get_inode(parent->d_sb, vino);
1472 dout("new_inode badness\n");
1480 ret = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1481 req->r_request_started, -1,
1482 &req->r_caps_reservation);
1484 pr_err("fill_inode badness on %p\n", in);
1485 if (d_really_is_negative(dn))
1492 if (d_really_is_negative(dn)) {
1493 struct dentry *realdn = splice_dentry(dn, in);
1494 if (IS_ERR(realdn)) {
1495 err = PTR_ERR(realdn);
1504 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1506 update_dentry_lease(dn, rinfo->dir_dlease[i],
1508 req->r_request_started);
1510 if (err == 0 && cache_ctl.index >= 0) {
1511 ret = fill_readdir_cache(d_inode(parent), dn,
1522 req->r_did_prepopulate = true;
1523 req->r_readdir_cache_idx = cache_ctl.index;
1525 ceph_readdir_cache_release(&cache_ctl);
1530 dout("readdir_prepopulate done\n");
1534 int ceph_inode_set_size(struct inode *inode, loff_t size)
1536 struct ceph_inode_info *ci = ceph_inode(inode);
1539 spin_lock(&ci->i_ceph_lock);
1540 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1541 i_size_write(inode, size);
1542 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1544 /* tell the MDS if we are approaching max_size */
1545 if ((size << 1) >= ci->i_max_size &&
1546 (ci->i_reported_size << 1) < ci->i_max_size)
1549 spin_unlock(&ci->i_ceph_lock);
1554 * Write back inode data in a worker thread. (This can't be done
1555 * in the message handler context.)
1557 void ceph_queue_writeback(struct inode *inode)
1560 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1561 &ceph_inode(inode)->i_wb_work)) {
1562 dout("ceph_queue_writeback %p\n", inode);
1564 dout("ceph_queue_writeback %p failed\n", inode);
1569 static void ceph_writeback_work(struct work_struct *work)
1571 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1573 struct inode *inode = &ci->vfs_inode;
1575 dout("writeback %p\n", inode);
1576 filemap_fdatawrite(&inode->i_data);
1581 * queue an async invalidation
1583 void ceph_queue_invalidate(struct inode *inode)
1586 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1587 &ceph_inode(inode)->i_pg_inv_work)) {
1588 dout("ceph_queue_invalidate %p\n", inode);
1590 dout("ceph_queue_invalidate %p failed\n", inode);
1596 * Invalidate inode pages in a worker thread. (This can't be done
1597 * in the message handler context.)
1599 static void ceph_invalidate_work(struct work_struct *work)
1601 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1603 struct inode *inode = &ci->vfs_inode;
1607 mutex_lock(&ci->i_truncate_mutex);
1608 spin_lock(&ci->i_ceph_lock);
1609 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1610 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1611 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1612 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1614 spin_unlock(&ci->i_ceph_lock);
1615 mutex_unlock(&ci->i_truncate_mutex);
1618 orig_gen = ci->i_rdcache_gen;
1619 spin_unlock(&ci->i_ceph_lock);
1621 truncate_pagecache(inode, 0);
1623 spin_lock(&ci->i_ceph_lock);
1624 if (orig_gen == ci->i_rdcache_gen &&
1625 orig_gen == ci->i_rdcache_revoking) {
1626 dout("invalidate_pages %p gen %d successful\n", inode,
1628 ci->i_rdcache_revoking--;
1631 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1632 inode, orig_gen, ci->i_rdcache_gen,
1633 ci->i_rdcache_revoking);
1634 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1637 spin_unlock(&ci->i_ceph_lock);
1638 mutex_unlock(&ci->i_truncate_mutex);
1641 ceph_check_caps(ci, 0, NULL);
1647 * called by trunc_wq;
1649 * We also truncate in a separate thread as well.
1651 static void ceph_vmtruncate_work(struct work_struct *work)
1653 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1655 struct inode *inode = &ci->vfs_inode;
1657 dout("vmtruncate_work %p\n", inode);
1658 __ceph_do_pending_vmtruncate(inode);
1663 * Queue an async vmtruncate. If we fail to queue work, we will handle
1664 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1666 void ceph_queue_vmtruncate(struct inode *inode)
1668 struct ceph_inode_info *ci = ceph_inode(inode);
1672 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1673 &ci->i_vmtruncate_work)) {
1674 dout("ceph_queue_vmtruncate %p\n", inode);
1676 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1677 inode, ci->i_truncate_pending);
1683 * Make sure any pending truncation is applied before doing anything
1684 * that may depend on it.
1686 void __ceph_do_pending_vmtruncate(struct inode *inode)
1688 struct ceph_inode_info *ci = ceph_inode(inode);
1690 int wrbuffer_refs, finish = 0;
1692 mutex_lock(&ci->i_truncate_mutex);
1694 spin_lock(&ci->i_ceph_lock);
1695 if (ci->i_truncate_pending == 0) {
1696 dout("__do_pending_vmtruncate %p none pending\n", inode);
1697 spin_unlock(&ci->i_ceph_lock);
1698 mutex_unlock(&ci->i_truncate_mutex);
1703 * make sure any dirty snapped pages are flushed before we
1704 * possibly truncate them.. so write AND block!
1706 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1707 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1709 spin_unlock(&ci->i_ceph_lock);
1710 filemap_write_and_wait_range(&inode->i_data, 0,
1711 inode->i_sb->s_maxbytes);
1715 /* there should be no reader or writer */
1716 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1718 to = ci->i_truncate_size;
1719 wrbuffer_refs = ci->i_wrbuffer_ref;
1720 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1721 ci->i_truncate_pending, to);
1722 spin_unlock(&ci->i_ceph_lock);
1724 truncate_pagecache(inode, to);
1726 spin_lock(&ci->i_ceph_lock);
1727 if (to == ci->i_truncate_size) {
1728 ci->i_truncate_pending = 0;
1731 spin_unlock(&ci->i_ceph_lock);
1735 mutex_unlock(&ci->i_truncate_mutex);
1737 if (wrbuffer_refs == 0)
1738 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1740 wake_up_all(&ci->i_cap_wq);
1746 static const struct inode_operations ceph_symlink_iops = {
1747 .readlink = generic_readlink,
1748 .get_link = simple_get_link,
1749 .setattr = ceph_setattr,
1750 .getattr = ceph_getattr,
1751 .setxattr = ceph_setxattr,
1752 .getxattr = ceph_getxattr,
1753 .listxattr = ceph_listxattr,
1754 .removexattr = ceph_removexattr,
1760 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1762 struct inode *inode = d_inode(dentry);
1763 struct ceph_inode_info *ci = ceph_inode(inode);
1764 const unsigned int ia_valid = attr->ia_valid;
1765 struct ceph_mds_request *req;
1766 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1767 struct ceph_cap_flush *prealloc_cf;
1769 int release = 0, dirtied = 0;
1772 int inode_dirty_flags = 0;
1773 bool lock_snap_rwsem = false;
1775 if (ceph_snap(inode) != CEPH_NOSNAP)
1778 err = inode_change_ok(inode, attr);
1782 prealloc_cf = ceph_alloc_cap_flush();
1786 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1789 ceph_free_cap_flush(prealloc_cf);
1790 return PTR_ERR(req);
1793 spin_lock(&ci->i_ceph_lock);
1794 issued = __ceph_caps_issued(ci, NULL);
1796 if (!ci->i_head_snapc &&
1797 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1798 lock_snap_rwsem = true;
1799 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1800 spin_unlock(&ci->i_ceph_lock);
1801 down_read(&mdsc->snap_rwsem);
1802 spin_lock(&ci->i_ceph_lock);
1803 issued = __ceph_caps_issued(ci, NULL);
1807 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1809 if (ia_valid & ATTR_UID) {
1810 dout("setattr %p uid %d -> %d\n", inode,
1811 from_kuid(&init_user_ns, inode->i_uid),
1812 from_kuid(&init_user_ns, attr->ia_uid));
1813 if (issued & CEPH_CAP_AUTH_EXCL) {
1814 inode->i_uid = attr->ia_uid;
1815 dirtied |= CEPH_CAP_AUTH_EXCL;
1816 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1817 !uid_eq(attr->ia_uid, inode->i_uid)) {
1818 req->r_args.setattr.uid = cpu_to_le32(
1819 from_kuid(&init_user_ns, attr->ia_uid));
1820 mask |= CEPH_SETATTR_UID;
1821 release |= CEPH_CAP_AUTH_SHARED;
1824 if (ia_valid & ATTR_GID) {
1825 dout("setattr %p gid %d -> %d\n", inode,
1826 from_kgid(&init_user_ns, inode->i_gid),
1827 from_kgid(&init_user_ns, attr->ia_gid));
1828 if (issued & CEPH_CAP_AUTH_EXCL) {
1829 inode->i_gid = attr->ia_gid;
1830 dirtied |= CEPH_CAP_AUTH_EXCL;
1831 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1832 !gid_eq(attr->ia_gid, inode->i_gid)) {
1833 req->r_args.setattr.gid = cpu_to_le32(
1834 from_kgid(&init_user_ns, attr->ia_gid));
1835 mask |= CEPH_SETATTR_GID;
1836 release |= CEPH_CAP_AUTH_SHARED;
1839 if (ia_valid & ATTR_MODE) {
1840 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1842 if (issued & CEPH_CAP_AUTH_EXCL) {
1843 inode->i_mode = attr->ia_mode;
1844 dirtied |= CEPH_CAP_AUTH_EXCL;
1845 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1846 attr->ia_mode != inode->i_mode) {
1847 inode->i_mode = attr->ia_mode;
1848 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1849 mask |= CEPH_SETATTR_MODE;
1850 release |= CEPH_CAP_AUTH_SHARED;
1854 if (ia_valid & ATTR_ATIME) {
1855 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1856 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1857 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1858 if (issued & CEPH_CAP_FILE_EXCL) {
1859 ci->i_time_warp_seq++;
1860 inode->i_atime = attr->ia_atime;
1861 dirtied |= CEPH_CAP_FILE_EXCL;
1862 } else if ((issued & CEPH_CAP_FILE_WR) &&
1863 timespec_compare(&inode->i_atime,
1864 &attr->ia_atime) < 0) {
1865 inode->i_atime = attr->ia_atime;
1866 dirtied |= CEPH_CAP_FILE_WR;
1867 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1868 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1869 ceph_encode_timespec(&req->r_args.setattr.atime,
1871 mask |= CEPH_SETATTR_ATIME;
1872 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1876 if (ia_valid & ATTR_MTIME) {
1877 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1878 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1879 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1880 if (issued & CEPH_CAP_FILE_EXCL) {
1881 ci->i_time_warp_seq++;
1882 inode->i_mtime = attr->ia_mtime;
1883 dirtied |= CEPH_CAP_FILE_EXCL;
1884 } else if ((issued & CEPH_CAP_FILE_WR) &&
1885 timespec_compare(&inode->i_mtime,
1886 &attr->ia_mtime) < 0) {
1887 inode->i_mtime = attr->ia_mtime;
1888 dirtied |= CEPH_CAP_FILE_WR;
1889 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1890 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1891 ceph_encode_timespec(&req->r_args.setattr.mtime,
1893 mask |= CEPH_SETATTR_MTIME;
1894 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1898 if (ia_valid & ATTR_SIZE) {
1899 dout("setattr %p size %lld -> %lld\n", inode,
1900 inode->i_size, attr->ia_size);
1901 if ((issued & CEPH_CAP_FILE_EXCL) &&
1902 attr->ia_size > inode->i_size) {
1903 i_size_write(inode, attr->ia_size);
1905 (attr->ia_size + (1 << 9) - 1) >> 9;
1906 inode->i_ctime = attr->ia_ctime;
1907 ci->i_reported_size = attr->ia_size;
1908 dirtied |= CEPH_CAP_FILE_EXCL;
1909 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1910 attr->ia_size != inode->i_size) {
1911 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1912 req->r_args.setattr.old_size =
1913 cpu_to_le64(inode->i_size);
1914 mask |= CEPH_SETATTR_SIZE;
1915 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1920 /* these do nothing */
1921 if (ia_valid & ATTR_CTIME) {
1922 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1923 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1924 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1925 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1926 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1927 only ? "ctime only" : "ignored");
1928 inode->i_ctime = attr->ia_ctime;
1931 * if kernel wants to dirty ctime but nothing else,
1932 * we need to choose a cap to dirty under, or do
1933 * a almost-no-op setattr
1935 if (issued & CEPH_CAP_AUTH_EXCL)
1936 dirtied |= CEPH_CAP_AUTH_EXCL;
1937 else if (issued & CEPH_CAP_FILE_EXCL)
1938 dirtied |= CEPH_CAP_FILE_EXCL;
1939 else if (issued & CEPH_CAP_XATTR_EXCL)
1940 dirtied |= CEPH_CAP_XATTR_EXCL;
1942 mask |= CEPH_SETATTR_CTIME;
1945 if (ia_valid & ATTR_FILE)
1946 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1949 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
1951 inode->i_ctime = CURRENT_TIME;
1955 spin_unlock(&ci->i_ceph_lock);
1956 if (lock_snap_rwsem)
1957 up_read(&mdsc->snap_rwsem);
1959 if (inode_dirty_flags)
1960 __mark_inode_dirty(inode, inode_dirty_flags);
1962 if (ia_valid & ATTR_MODE) {
1963 err = posix_acl_chmod(inode, attr->ia_mode);
1969 req->r_inode = inode;
1971 req->r_inode_drop = release;
1972 req->r_args.setattr.mask = cpu_to_le32(mask);
1973 req->r_num_caps = 1;
1974 err = ceph_mdsc_do_request(mdsc, NULL, req);
1976 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1977 ceph_cap_string(dirtied), mask);
1979 ceph_mdsc_put_request(req);
1980 if (mask & CEPH_SETATTR_SIZE)
1981 __ceph_do_pending_vmtruncate(inode);
1982 ceph_free_cap_flush(prealloc_cf);
1985 ceph_mdsc_put_request(req);
1986 ceph_free_cap_flush(prealloc_cf);
1991 * Verify that we have a lease on the given mask. If not,
1992 * do a getattr against an mds.
1994 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1995 int mask, bool force)
1997 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1998 struct ceph_mds_client *mdsc = fsc->mdsc;
1999 struct ceph_mds_request *req;
2002 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2003 dout("do_getattr inode %p SNAPDIR\n", inode);
2007 dout("do_getattr inode %p mask %s mode 0%o\n",
2008 inode, ceph_cap_string(mask), inode->i_mode);
2009 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2012 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2014 return PTR_ERR(req);
2015 req->r_inode = inode;
2017 req->r_num_caps = 1;
2018 req->r_args.getattr.mask = cpu_to_le32(mask);
2019 req->r_locked_page = locked_page;
2020 err = ceph_mdsc_do_request(mdsc, NULL, req);
2021 if (locked_page && err == 0) {
2022 u64 inline_version = req->r_reply_info.targeti.inline_version;
2023 if (inline_version == 0) {
2024 /* the reply is supposed to contain inline data */
2026 } else if (inline_version == CEPH_INLINE_NONE) {
2029 err = req->r_reply_info.targeti.inline_len;
2032 ceph_mdsc_put_request(req);
2033 dout("do_getattr result=%d\n", err);
2039 * Check inode permissions. We verify we have a valid value for
2040 * the AUTH cap, then call the generic handler.
2042 int ceph_permission(struct inode *inode, int mask)
2046 if (mask & MAY_NOT_BLOCK)
2049 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2052 err = generic_permission(inode, mask);
2057 * Get all attributes. Hopefully somedata we'll have a statlite()
2058 * and can limit the fields we require to be accurate.
2060 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2063 struct inode *inode = d_inode(dentry);
2064 struct ceph_inode_info *ci = ceph_inode(inode);
2067 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2069 generic_fillattr(inode, stat);
2070 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2071 if (ceph_snap(inode) != CEPH_NOSNAP)
2072 stat->dev = ceph_snap(inode);
2075 if (S_ISDIR(inode->i_mode)) {
2076 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2078 stat->size = ci->i_rbytes;
2080 stat->size = ci->i_files + ci->i_subdirs;
2082 stat->blksize = 65536;