1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <linux/string.h>
7 #include <linux/uaccess.h>
8 #include <linux/kernel.h>
9 #include <linux/writeback.h>
10 #include <linux/vmalloc.h>
11 #include <linux/posix_acl.h>
12 #include <linux/random.h>
13 #include <linux/sort.h>
16 #include "mds_client.h"
18 #include <linux/ceph/decode.h>
21 * Ceph inode operations
23 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
24 * setattr, etc.), xattr helpers, and helpers for assimilating
25 * metadata returned by the MDS into our cache.
27 * Also define helpers for doing asynchronous writeback, invalidation,
28 * and truncation for the benefit of those who can't afford to block
29 * (typically because they are in the message handler path).
32 static const struct inode_operations ceph_symlink_iops;
34 static void ceph_invalidate_work(struct work_struct *work);
35 static void ceph_writeback_work(struct work_struct *work);
36 static void ceph_vmtruncate_work(struct work_struct *work);
39 * find or create an inode, given the ceph ino number
41 static int ceph_set_ino_cb(struct inode *inode, void *data)
43 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
44 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
48 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
51 ino_t t = ceph_vino_to_ino(vino);
53 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
55 return ERR_PTR(-ENOMEM);
56 if (inode->i_state & I_NEW) {
57 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
58 inode, ceph_vinop(inode), (u64)inode->i_ino);
59 unlock_new_inode(inode);
62 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
68 * get/constuct snapdir inode for a given directory
70 struct inode *ceph_get_snapdir(struct inode *parent)
72 struct ceph_vino vino = {
73 .ino = ceph_ino(parent),
76 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
77 struct ceph_inode_info *ci = ceph_inode(inode);
79 BUG_ON(!S_ISDIR(parent->i_mode));
82 inode->i_mode = parent->i_mode;
83 inode->i_uid = parent->i_uid;
84 inode->i_gid = parent->i_gid;
85 inode->i_op = &ceph_snapdir_iops;
86 inode->i_fop = &ceph_snapdir_fops;
87 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
92 const struct inode_operations ceph_file_iops = {
93 .permission = ceph_permission,
94 .setattr = ceph_setattr,
95 .getattr = ceph_getattr,
96 .setxattr = ceph_setxattr,
97 .getxattr = ceph_getxattr,
98 .listxattr = ceph_listxattr,
99 .removexattr = ceph_removexattr,
100 .get_acl = ceph_get_acl,
101 .set_acl = ceph_set_acl,
106 * We use a 'frag tree' to keep track of the MDS's directory fragments
107 * for a given inode (usually there is just a single fragment). We
108 * need to know when a child frag is delegated to a new MDS, or when
109 * it is flagged as replicated, so we can direct our requests
114 * find/create a frag in the tree
116 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
120 struct rb_node *parent = NULL;
121 struct ceph_inode_frag *frag;
124 p = &ci->i_fragtree.rb_node;
127 frag = rb_entry(parent, struct ceph_inode_frag, node);
128 c = ceph_frag_compare(f, frag->frag);
137 frag = kmalloc(sizeof(*frag), GFP_NOFS);
139 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
140 "frag %x\n", &ci->vfs_inode,
141 ceph_vinop(&ci->vfs_inode), f);
142 return ERR_PTR(-ENOMEM);
149 rb_link_node(&frag->node, parent, p);
150 rb_insert_color(&frag->node, &ci->i_fragtree);
152 dout("get_or_create_frag added %llx.%llx frag %x\n",
153 ceph_vinop(&ci->vfs_inode), f);
158 * find a specific frag @f
160 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
162 struct rb_node *n = ci->i_fragtree.rb_node;
165 struct ceph_inode_frag *frag =
166 rb_entry(n, struct ceph_inode_frag, node);
167 int c = ceph_frag_compare(f, frag->frag);
179 * Choose frag containing the given value @v. If @pfrag is
180 * specified, copy the frag delegation info to the caller if
183 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
184 struct ceph_inode_frag *pfrag, int *found)
186 u32 t = ceph_frag_make(0, 0);
187 struct ceph_inode_frag *frag;
195 WARN_ON(!ceph_frag_contains_value(t, v));
196 frag = __ceph_find_frag(ci, t);
198 break; /* t is a leaf */
199 if (frag->split_by == 0) {
201 memcpy(pfrag, frag, sizeof(*pfrag));
208 nway = 1 << frag->split_by;
209 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
210 frag->split_by, nway);
211 for (i = 0; i < nway; i++) {
212 n = ceph_frag_make_child(t, frag->split_by, i);
213 if (ceph_frag_contains_value(n, v)) {
220 dout("choose_frag(%x) = %x\n", v, t);
225 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
226 struct ceph_inode_frag *pfrag, int *found)
229 mutex_lock(&ci->i_fragtree_mutex);
230 ret = __ceph_choose_frag(ci, v, pfrag, found);
231 mutex_unlock(&ci->i_fragtree_mutex);
236 * Process dirfrag (delegation) info from the mds. Include leaf
237 * fragment in tree ONLY if ndist > 0. Otherwise, only
238 * branches/splits are included in i_fragtree)
240 static int ceph_fill_dirfrag(struct inode *inode,
241 struct ceph_mds_reply_dirfrag *dirinfo)
243 struct ceph_inode_info *ci = ceph_inode(inode);
244 struct ceph_inode_frag *frag;
245 u32 id = le32_to_cpu(dirinfo->frag);
246 int mds = le32_to_cpu(dirinfo->auth);
247 int ndist = le32_to_cpu(dirinfo->ndist);
252 spin_lock(&ci->i_ceph_lock);
254 diri_auth = ci->i_auth_cap->mds;
255 spin_unlock(&ci->i_ceph_lock);
257 if (mds == -1) /* CDIR_AUTH_PARENT */
260 mutex_lock(&ci->i_fragtree_mutex);
261 if (ndist == 0 && mds == diri_auth) {
262 /* no delegation info needed. */
263 frag = __ceph_find_frag(ci, id);
266 if (frag->split_by == 0) {
267 /* tree leaf, remove */
268 dout("fill_dirfrag removed %llx.%llx frag %x"
269 " (no ref)\n", ceph_vinop(inode), id);
270 rb_erase(&frag->node, &ci->i_fragtree);
273 /* tree branch, keep and clear */
274 dout("fill_dirfrag cleared %llx.%llx frag %x"
275 " referral\n", ceph_vinop(inode), id);
283 /* find/add this frag to store mds delegation info */
284 frag = __get_or_create_frag(ci, id);
286 /* this is not the end of the world; we can continue
287 with bad/inaccurate delegation info */
288 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
289 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
295 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
296 for (i = 0; i < frag->ndist; i++)
297 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
298 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
299 ceph_vinop(inode), frag->frag, frag->ndist);
302 mutex_unlock(&ci->i_fragtree_mutex);
306 static int frag_tree_split_cmp(const void *l, const void *r)
308 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
309 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
310 return ceph_frag_compare(ls->frag, rs->frag);
313 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
316 return f == ceph_frag_make(0, 0);
317 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
319 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
322 static int ceph_fill_fragtree(struct inode *inode,
323 struct ceph_frag_tree_head *fragtree,
324 struct ceph_mds_reply_dirfrag *dirinfo)
326 struct ceph_inode_info *ci = ceph_inode(inode);
327 struct ceph_inode_frag *frag, *prev_frag = NULL;
328 struct rb_node *rb_node;
329 unsigned i, split_by, nsplits;
333 mutex_lock(&ci->i_fragtree_mutex);
334 nsplits = le32_to_cpu(fragtree->nsplits);
335 if (nsplits != ci->i_fragtree_nsplits) {
337 } else if (nsplits) {
338 i = prandom_u32() % nsplits;
339 id = le32_to_cpu(fragtree->splits[i].frag);
340 if (!__ceph_find_frag(ci, id))
342 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
343 rb_node = rb_first(&ci->i_fragtree);
344 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
345 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
348 if (!update && dirinfo) {
349 id = le32_to_cpu(dirinfo->frag);
350 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
357 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
358 frag_tree_split_cmp, NULL);
361 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
362 rb_node = rb_first(&ci->i_fragtree);
363 for (i = 0; i < nsplits; i++) {
364 id = le32_to_cpu(fragtree->splits[i].frag);
365 split_by = le32_to_cpu(fragtree->splits[i].by);
366 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
367 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
368 "frag %x split by %d\n", ceph_vinop(inode),
369 i, nsplits, id, split_by);
374 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
375 if (ceph_frag_compare(frag->frag, id) >= 0) {
376 if (frag->frag != id)
379 rb_node = rb_next(rb_node);
382 rb_node = rb_next(rb_node);
383 /* delete stale split/leaf node */
384 if (frag->split_by > 0 ||
385 !is_frag_child(frag->frag, prev_frag)) {
386 rb_erase(&frag->node, &ci->i_fragtree);
387 if (frag->split_by > 0)
388 ci->i_fragtree_nsplits--;
394 frag = __get_or_create_frag(ci, id);
398 if (frag->split_by == 0)
399 ci->i_fragtree_nsplits++;
400 frag->split_by = split_by;
401 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
405 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
406 rb_node = rb_next(rb_node);
407 /* delete stale split/leaf node */
408 if (frag->split_by > 0 ||
409 !is_frag_child(frag->frag, prev_frag)) {
410 rb_erase(&frag->node, &ci->i_fragtree);
411 if (frag->split_by > 0)
412 ci->i_fragtree_nsplits--;
417 mutex_unlock(&ci->i_fragtree_mutex);
422 * initialize a newly allocated inode.
424 struct inode *ceph_alloc_inode(struct super_block *sb)
426 struct ceph_inode_info *ci;
429 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
433 dout("alloc_inode %p\n", &ci->vfs_inode);
435 spin_lock_init(&ci->i_ceph_lock);
438 ci->i_inline_version = 0;
439 ci->i_time_warp_seq = 0;
440 ci->i_ceph_flags = 0;
441 atomic64_set(&ci->i_ordered_count, 1);
442 atomic64_set(&ci->i_release_count, 1);
443 atomic64_set(&ci->i_complete_seq[0], 0);
444 atomic64_set(&ci->i_complete_seq[1], 0);
445 ci->i_symlink = NULL;
447 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
448 ci->i_pool_ns_len = 0;
450 ci->i_fragtree = RB_ROOT;
451 mutex_init(&ci->i_fragtree_mutex);
453 ci->i_xattrs.blob = NULL;
454 ci->i_xattrs.prealloc_blob = NULL;
455 ci->i_xattrs.dirty = false;
456 ci->i_xattrs.index = RB_ROOT;
457 ci->i_xattrs.count = 0;
458 ci->i_xattrs.names_size = 0;
459 ci->i_xattrs.vals_size = 0;
460 ci->i_xattrs.version = 0;
461 ci->i_xattrs.index_version = 0;
463 ci->i_caps = RB_ROOT;
464 ci->i_auth_cap = NULL;
465 ci->i_dirty_caps = 0;
466 ci->i_flushing_caps = 0;
467 INIT_LIST_HEAD(&ci->i_dirty_item);
468 INIT_LIST_HEAD(&ci->i_flushing_item);
469 ci->i_prealloc_cap_flush = NULL;
470 ci->i_cap_flush_tree = RB_ROOT;
471 init_waitqueue_head(&ci->i_cap_wq);
472 ci->i_hold_caps_min = 0;
473 ci->i_hold_caps_max = 0;
474 INIT_LIST_HEAD(&ci->i_cap_delay_list);
475 INIT_LIST_HEAD(&ci->i_cap_snaps);
476 ci->i_head_snapc = NULL;
479 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
480 ci->i_nr_by_mode[i] = 0;
482 mutex_init(&ci->i_truncate_mutex);
483 ci->i_truncate_seq = 0;
484 ci->i_truncate_size = 0;
485 ci->i_truncate_pending = 0;
488 ci->i_reported_size = 0;
489 ci->i_wanted_max_size = 0;
490 ci->i_requested_max_size = 0;
494 ci->i_rdcache_ref = 0;
497 ci->i_wrbuffer_ref = 0;
498 ci->i_wrbuffer_ref_head = 0;
499 ci->i_shared_gen = 0;
500 ci->i_rdcache_gen = 0;
501 ci->i_rdcache_revoking = 0;
503 INIT_LIST_HEAD(&ci->i_unsafe_writes);
504 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
505 INIT_LIST_HEAD(&ci->i_unsafe_iops);
506 spin_lock_init(&ci->i_unsafe_lock);
508 ci->i_snap_realm = NULL;
509 INIT_LIST_HEAD(&ci->i_snap_realm_item);
510 INIT_LIST_HEAD(&ci->i_snap_flush_item);
512 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
513 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
515 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
517 ceph_fscache_inode_init(ci);
519 return &ci->vfs_inode;
522 static void ceph_i_callback(struct rcu_head *head)
524 struct inode *inode = container_of(head, struct inode, i_rcu);
525 struct ceph_inode_info *ci = ceph_inode(inode);
527 kmem_cache_free(ceph_inode_cachep, ci);
530 void ceph_destroy_inode(struct inode *inode)
532 struct ceph_inode_info *ci = ceph_inode(inode);
533 struct ceph_inode_frag *frag;
536 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
538 ceph_fscache_unregister_inode_cookie(ci);
540 ceph_queue_caps_release(inode);
543 * we may still have a snap_realm reference if there are stray
544 * caps in i_snap_caps.
546 if (ci->i_snap_realm) {
547 struct ceph_mds_client *mdsc =
548 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
549 struct ceph_snap_realm *realm = ci->i_snap_realm;
551 dout(" dropping residual ref to snap realm %p\n", realm);
552 spin_lock(&realm->inodes_with_caps_lock);
553 list_del_init(&ci->i_snap_realm_item);
554 spin_unlock(&realm->inodes_with_caps_lock);
555 ceph_put_snap_realm(mdsc, realm);
558 kfree(ci->i_symlink);
559 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
560 frag = rb_entry(n, struct ceph_inode_frag, node);
561 rb_erase(n, &ci->i_fragtree);
564 ci->i_fragtree_nsplits = 0;
566 __ceph_destroy_xattrs(ci);
567 if (ci->i_xattrs.blob)
568 ceph_buffer_put(ci->i_xattrs.blob);
569 if (ci->i_xattrs.prealloc_blob)
570 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
572 call_rcu(&inode->i_rcu, ceph_i_callback);
575 int ceph_drop_inode(struct inode *inode)
578 * Positve dentry and corresponding inode are always accompanied
579 * in MDS reply. So no need to keep inode in the cache after
580 * dropping all its aliases.
586 * Helpers to fill in size, ctime, mtime, and atime. We have to be
587 * careful because either the client or MDS may have more up to date
588 * info, depending on which capabilities are held, and whether
589 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
590 * and size are monotonically increasing, except when utimes() or
591 * truncate() increments the corresponding _seq values.)
593 int ceph_fill_file_size(struct inode *inode, int issued,
594 u32 truncate_seq, u64 truncate_size, u64 size)
596 struct ceph_inode_info *ci = ceph_inode(inode);
599 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
600 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
601 dout("size %lld -> %llu\n", inode->i_size, size);
602 if (size > 0 && S_ISDIR(inode->i_mode)) {
603 pr_err("fill_file_size non-zero size for directory\n");
606 i_size_write(inode, size);
607 inode->i_blocks = (size + (1<<9) - 1) >> 9;
608 ci->i_reported_size = size;
609 if (truncate_seq != ci->i_truncate_seq) {
610 dout("truncate_seq %u -> %u\n",
611 ci->i_truncate_seq, truncate_seq);
612 ci->i_truncate_seq = truncate_seq;
614 /* the MDS should have revoked these caps */
615 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
618 CEPH_CAP_FILE_LAZYIO));
620 * If we hold relevant caps, or in the case where we're
621 * not the only client referencing this file and we
622 * don't hold those caps, then we need to check whether
623 * the file is either opened or mmaped
625 if ((issued & (CEPH_CAP_FILE_CACHE|
626 CEPH_CAP_FILE_BUFFER)) ||
627 mapping_mapped(inode->i_mapping) ||
628 __ceph_caps_file_wanted(ci)) {
629 ci->i_truncate_pending++;
634 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
635 ci->i_truncate_size != truncate_size) {
636 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
638 ci->i_truncate_size = truncate_size;
642 ceph_fscache_invalidate(inode);
647 void ceph_fill_file_time(struct inode *inode, int issued,
648 u64 time_warp_seq, struct timespec *ctime,
649 struct timespec *mtime, struct timespec *atime)
651 struct ceph_inode_info *ci = ceph_inode(inode);
654 if (issued & (CEPH_CAP_FILE_EXCL|
656 CEPH_CAP_FILE_BUFFER|
658 CEPH_CAP_XATTR_EXCL)) {
659 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
660 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
661 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
662 ctime->tv_sec, ctime->tv_nsec);
663 inode->i_ctime = *ctime;
665 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
666 /* the MDS did a utimes() */
667 dout("mtime %ld.%09ld -> %ld.%09ld "
669 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
670 mtime->tv_sec, mtime->tv_nsec,
671 ci->i_time_warp_seq, (int)time_warp_seq);
673 inode->i_mtime = *mtime;
674 inode->i_atime = *atime;
675 ci->i_time_warp_seq = time_warp_seq;
676 } else if (time_warp_seq == ci->i_time_warp_seq) {
677 /* nobody did utimes(); take the max */
678 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
679 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
680 inode->i_mtime.tv_sec,
681 inode->i_mtime.tv_nsec,
682 mtime->tv_sec, mtime->tv_nsec);
683 inode->i_mtime = *mtime;
685 if (timespec_compare(atime, &inode->i_atime) > 0) {
686 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
687 inode->i_atime.tv_sec,
688 inode->i_atime.tv_nsec,
689 atime->tv_sec, atime->tv_nsec);
690 inode->i_atime = *atime;
692 } else if (issued & CEPH_CAP_FILE_EXCL) {
693 /* we did a utimes(); ignore mds values */
698 /* we have no write|excl caps; whatever the MDS says is true */
699 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
700 inode->i_ctime = *ctime;
701 inode->i_mtime = *mtime;
702 inode->i_atime = *atime;
703 ci->i_time_warp_seq = time_warp_seq;
708 if (warn) /* time_warp_seq shouldn't go backwards */
709 dout("%p mds time_warp_seq %llu < %u\n",
710 inode, time_warp_seq, ci->i_time_warp_seq);
714 * Populate an inode based on info from mds. May be called on new or
717 static int fill_inode(struct inode *inode, struct page *locked_page,
718 struct ceph_mds_reply_info_in *iinfo,
719 struct ceph_mds_reply_dirfrag *dirinfo,
720 struct ceph_mds_session *session,
721 unsigned long ttl_from, int cap_fmode,
722 struct ceph_cap_reservation *caps_reservation)
724 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
725 struct ceph_mds_reply_inode *info = iinfo->in;
726 struct ceph_inode_info *ci = ceph_inode(inode);
727 int issued = 0, implemented, new_issued;
728 struct timespec mtime, atime, ctime;
729 struct ceph_buffer *xattr_blob = NULL;
730 struct ceph_cap *new_cap = NULL;
733 bool queue_trunc = false;
734 bool new_version = false;
735 bool fill_inline = false;
737 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
738 inode, ceph_vinop(inode), le64_to_cpu(info->version),
741 /* prealloc new cap struct */
742 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
743 new_cap = ceph_get_cap(mdsc, caps_reservation);
746 * prealloc xattr data, if it looks like we'll need it. only
747 * if len > 4 (meaning there are actually xattrs; the first 4
748 * bytes are the xattr count).
750 if (iinfo->xattr_len > 4) {
751 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
753 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
757 spin_lock(&ci->i_ceph_lock);
760 * provided version will be odd if inode value is projected,
761 * even if stable. skip the update if we have newer stable
762 * info (ours>=theirs, e.g. due to racing mds replies), unless
763 * we are getting projected (unstable) info (in which case the
764 * version is odd, and we want ours>theirs).
770 if (ci->i_version == 0 ||
771 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
772 le64_to_cpu(info->version) > (ci->i_version & ~1)))
775 issued = __ceph_caps_issued(ci, &implemented);
776 issued |= implemented | __ceph_caps_dirty(ci);
777 new_issued = ~issued & le32_to_cpu(info->cap.caps);
780 ci->i_version = le64_to_cpu(info->version);
782 inode->i_rdev = le32_to_cpu(info->rdev);
783 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
785 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
786 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
787 inode->i_mode = le32_to_cpu(info->mode);
788 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
789 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
790 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
791 from_kuid(&init_user_ns, inode->i_uid),
792 from_kgid(&init_user_ns, inode->i_gid));
795 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
796 (issued & CEPH_CAP_LINK_EXCL) == 0)
797 set_nlink(inode, le32_to_cpu(info->nlink));
799 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
800 /* be careful with mtime, atime, size */
801 ceph_decode_timespec(&atime, &info->atime);
802 ceph_decode_timespec(&mtime, &info->mtime);
803 ceph_decode_timespec(&ctime, &info->ctime);
804 ceph_fill_file_time(inode, issued,
805 le32_to_cpu(info->time_warp_seq),
806 &ctime, &mtime, &atime);
810 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
811 if (ci->i_layout.fl_pg_pool != info->layout.fl_pg_pool)
812 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
813 ci->i_layout = info->layout;
814 ci->i_pool_ns_len = iinfo->pool_ns_len;
816 queue_trunc = ceph_fill_file_size(inode, issued,
817 le32_to_cpu(info->truncate_seq),
818 le64_to_cpu(info->truncate_size),
819 le64_to_cpu(info->size));
820 /* only update max_size on auth cap */
821 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
822 ci->i_max_size != le64_to_cpu(info->max_size)) {
823 dout("max_size %lld -> %llu\n", ci->i_max_size,
824 le64_to_cpu(info->max_size));
825 ci->i_max_size = le64_to_cpu(info->max_size);
830 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
831 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
832 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
833 if (ci->i_xattrs.blob)
834 ceph_buffer_put(ci->i_xattrs.blob);
835 ci->i_xattrs.blob = xattr_blob;
837 memcpy(ci->i_xattrs.blob->vec.iov_base,
838 iinfo->xattr_data, iinfo->xattr_len);
839 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
840 ceph_forget_all_cached_acls(inode);
844 inode->i_mapping->a_ops = &ceph_aops;
846 switch (inode->i_mode & S_IFMT) {
851 init_special_inode(inode, inode->i_mode, inode->i_rdev);
852 inode->i_op = &ceph_file_iops;
855 inode->i_op = &ceph_file_iops;
856 inode->i_fop = &ceph_file_fops;
859 inode->i_op = &ceph_symlink_iops;
860 if (!ci->i_symlink) {
861 u32 symlen = iinfo->symlink_len;
864 spin_unlock(&ci->i_ceph_lock);
867 if (WARN_ON(symlen != i_size_read(inode)))
871 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
875 spin_lock(&ci->i_ceph_lock);
879 kfree(sym); /* lost a race */
881 inode->i_link = ci->i_symlink;
884 inode->i_op = &ceph_dir_iops;
885 inode->i_fop = &ceph_dir_fops;
887 ci->i_dir_layout = iinfo->dir_layout;
889 ci->i_files = le64_to_cpu(info->files);
890 ci->i_subdirs = le64_to_cpu(info->subdirs);
891 ci->i_rbytes = le64_to_cpu(info->rbytes);
892 ci->i_rfiles = le64_to_cpu(info->rfiles);
893 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
894 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
897 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
898 ceph_vinop(inode), inode->i_mode);
901 /* were we issued a capability? */
902 if (info->cap.caps) {
903 if (ceph_snap(inode) == CEPH_NOSNAP) {
904 unsigned caps = le32_to_cpu(info->cap.caps);
905 ceph_add_cap(inode, session,
906 le64_to_cpu(info->cap.cap_id),
908 le32_to_cpu(info->cap.wanted),
909 le32_to_cpu(info->cap.seq),
910 le32_to_cpu(info->cap.mseq),
911 le64_to_cpu(info->cap.realm),
912 info->cap.flags, &new_cap);
914 /* set dir completion flag? */
915 if (S_ISDIR(inode->i_mode) &&
916 ci->i_files == 0 && ci->i_subdirs == 0 &&
917 (caps & CEPH_CAP_FILE_SHARED) &&
918 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
919 !__ceph_dir_is_complete(ci)) {
920 dout(" marking %p complete (empty)\n", inode);
921 i_size_write(inode, 0);
922 __ceph_dir_set_complete(ci,
923 atomic64_read(&ci->i_release_count),
924 atomic64_read(&ci->i_ordered_count));
929 dout(" %p got snap_caps %s\n", inode,
930 ceph_cap_string(le32_to_cpu(info->cap.caps)));
931 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
933 __ceph_get_fmode(ci, cap_fmode);
935 } else if (cap_fmode >= 0) {
936 pr_warn("mds issued no caps on %llx.%llx\n",
938 __ceph_get_fmode(ci, cap_fmode);
941 if (iinfo->inline_version > 0 &&
942 iinfo->inline_version >= ci->i_inline_version) {
943 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
944 ci->i_inline_version = iinfo->inline_version;
945 if (ci->i_inline_version != CEPH_INLINE_NONE &&
947 (le32_to_cpu(info->cap.caps) & cache_caps)))
951 spin_unlock(&ci->i_ceph_lock);
954 ceph_fill_inline_data(inode, locked_page,
955 iinfo->inline_data, iinfo->inline_len);
958 wake_up_all(&ci->i_cap_wq);
960 /* queue truncate if we saw i_size decrease */
962 ceph_queue_vmtruncate(inode);
964 /* populate frag tree */
965 if (S_ISDIR(inode->i_mode))
966 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
968 /* update delegation info? */
970 ceph_fill_dirfrag(inode, dirinfo);
975 ceph_put_cap(mdsc, new_cap);
977 ceph_buffer_put(xattr_blob);
982 * caller should hold session s_mutex.
984 static void update_dentry_lease(struct dentry *dentry,
985 struct ceph_mds_reply_lease *lease,
986 struct ceph_mds_session *session,
987 unsigned long from_time)
989 struct ceph_dentry_info *di = ceph_dentry(dentry);
990 long unsigned duration = le32_to_cpu(lease->duration_ms);
991 long unsigned ttl = from_time + (duration * HZ) / 1000;
992 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
995 /* only track leases on regular dentries */
996 if (dentry->d_op != &ceph_dentry_ops)
999 spin_lock(&dentry->d_lock);
1000 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1001 dentry, duration, ttl);
1003 /* make lease_rdcache_gen match directory */
1004 dir = d_inode(dentry->d_parent);
1005 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1010 if (di->lease_gen == session->s_cap_gen &&
1011 time_before(ttl, dentry->d_time))
1012 goto out_unlock; /* we already have a newer lease. */
1014 if (di->lease_session && di->lease_session != session)
1017 ceph_dentry_lru_touch(dentry);
1019 if (!di->lease_session)
1020 di->lease_session = ceph_get_mds_session(session);
1021 di->lease_gen = session->s_cap_gen;
1022 di->lease_seq = le32_to_cpu(lease->seq);
1023 di->lease_renew_after = half_ttl;
1024 di->lease_renew_from = 0;
1025 dentry->d_time = ttl;
1027 spin_unlock(&dentry->d_lock);
1032 * splice a dentry to an inode.
1033 * caller must hold directory i_mutex for this to be safe.
1035 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1037 struct dentry *realdn;
1039 BUG_ON(d_inode(dn));
1041 /* dn must be unhashed */
1042 if (!d_unhashed(dn))
1044 realdn = d_splice_alias(in, dn);
1045 if (IS_ERR(realdn)) {
1046 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1047 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1048 dn = realdn; /* note realdn contains the error */
1050 } else if (realdn) {
1051 dout("dn %p (%d) spliced with %p (%d) "
1052 "inode %p ino %llx.%llx\n",
1054 realdn, d_count(realdn),
1055 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1059 BUG_ON(!ceph_dentry(dn));
1060 dout("dn %p attached to %p ino %llx.%llx\n",
1061 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1068 * Incorporate results into the local cache. This is either just
1069 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1072 * A reply may contain
1073 * a directory inode along with a dentry.
1074 * and/or a target inode
1076 * Called with snap_rwsem (read).
1078 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1079 struct ceph_mds_session *session)
1081 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1082 struct inode *in = NULL;
1083 struct ceph_vino vino;
1084 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1087 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1088 rinfo->head->is_dentry, rinfo->head->is_target);
1094 * If we resend completed ops to a recovering mds, we get no
1095 * trace. Since that is very rare, pretend this is the case
1096 * to ensure the 'no trace' handlers in the callers behave.
1098 * Fill in inodes unconditionally to avoid breaking cap
1101 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1102 pr_info("fill_trace faking empty trace on %lld %s\n",
1103 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1104 if (rinfo->head->is_dentry) {
1105 rinfo->head->is_dentry = 0;
1106 err = fill_inode(req->r_locked_dir,
1107 &rinfo->diri, rinfo->dirfrag,
1108 session, req->r_request_started, -1);
1110 if (rinfo->head->is_target) {
1111 rinfo->head->is_target = 0;
1112 ininfo = rinfo->targeti.in;
1113 vino.ino = le64_to_cpu(ininfo->ino);
1114 vino.snap = le64_to_cpu(ininfo->snapid);
1115 in = ceph_get_inode(sb, vino);
1116 err = fill_inode(in, &rinfo->targeti, NULL,
1117 session, req->r_request_started,
1124 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1125 dout("fill_trace reply is empty!\n");
1126 if (rinfo->head->result == 0 && req->r_locked_dir)
1127 ceph_invalidate_dir_request(req);
1131 if (rinfo->head->is_dentry) {
1132 struct inode *dir = req->r_locked_dir;
1135 err = fill_inode(dir, NULL,
1136 &rinfo->diri, rinfo->dirfrag,
1137 session, req->r_request_started, -1,
1138 &req->r_caps_reservation);
1145 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1147 struct dentry *dn, *parent;
1149 BUG_ON(!rinfo->head->is_target);
1150 BUG_ON(req->r_dentry);
1152 parent = d_find_any_alias(dir);
1155 dname.name = rinfo->dname;
1156 dname.len = rinfo->dname_len;
1157 dname.hash = full_name_hash(dname.name, dname.len);
1158 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1159 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1161 dn = d_lookup(parent, &dname);
1162 dout("d_lookup on parent=%p name=%.*s got %p\n",
1163 parent, dname.len, dname.name, dn);
1166 dn = d_alloc(parent, &dname);
1167 dout("d_alloc %p '%.*s' = %p\n", parent,
1168 dname.len, dname.name, dn);
1174 err = ceph_init_dentry(dn);
1180 } else if (d_really_is_positive(dn) &&
1181 (ceph_ino(d_inode(dn)) != vino.ino ||
1182 ceph_snap(d_inode(dn)) != vino.snap)) {
1183 dout(" dn %p points to wrong inode %p\n",
1195 if (rinfo->head->is_target) {
1196 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1197 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1199 in = ceph_get_inode(sb, vino);
1204 req->r_target_inode = in;
1206 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1207 session, req->r_request_started,
1208 (!req->r_aborted && rinfo->head->result == 0) ?
1210 &req->r_caps_reservation);
1212 pr_err("fill_inode badness %p %llx.%llx\n",
1213 in, ceph_vinop(in));
1219 * ignore null lease/binding on snapdir ENOENT, or else we
1220 * will have trouble splicing in the virtual snapdir later
1222 if (rinfo->head->is_dentry && !req->r_aborted &&
1223 req->r_locked_dir &&
1224 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1225 fsc->mount_options->snapdir_name,
1226 req->r_dentry->d_name.len))) {
1228 * lookup link rename : null -> possibly existing inode
1229 * mknod symlink mkdir : null -> new inode
1230 * unlink : linked -> null
1232 struct inode *dir = req->r_locked_dir;
1233 struct dentry *dn = req->r_dentry;
1234 bool have_dir_cap, have_lease;
1238 BUG_ON(d_inode(dn->d_parent) != dir);
1239 BUG_ON(ceph_ino(dir) !=
1240 le64_to_cpu(rinfo->diri.in->ino));
1241 BUG_ON(ceph_snap(dir) !=
1242 le64_to_cpu(rinfo->diri.in->snapid));
1244 /* do we have a lease on the whole dir? */
1246 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1247 CEPH_CAP_FILE_SHARED);
1249 /* do we have a dn lease? */
1250 have_lease = have_dir_cap ||
1251 le32_to_cpu(rinfo->dlease->duration_ms);
1253 dout("fill_trace no dentry lease or dir cap\n");
1256 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1257 struct inode *olddir = req->r_old_dentry_dir;
1260 dout(" src %p '%pd' dst %p '%pd'\n",
1264 dout("fill_trace doing d_move %p -> %p\n",
1265 req->r_old_dentry, dn);
1267 /* d_move screws up sibling dentries' offsets */
1268 ceph_dir_clear_ordered(dir);
1269 ceph_dir_clear_ordered(olddir);
1271 d_move(req->r_old_dentry, dn);
1272 dout(" src %p '%pd' dst %p '%pd'\n",
1277 /* ensure target dentry is invalidated, despite
1278 rehashing bug in vfs_rename_dir */
1279 ceph_invalidate_dentry_lease(dn);
1281 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1282 ceph_dentry(req->r_old_dentry)->offset);
1284 dn = req->r_old_dentry; /* use old_dentry */
1288 if (!rinfo->head->is_target) {
1289 dout("fill_trace null dentry\n");
1290 if (d_really_is_positive(dn)) {
1291 ceph_dir_clear_ordered(dir);
1292 dout("d_delete %p\n", dn);
1295 if (have_lease && d_unhashed(dn))
1297 update_dentry_lease(dn, rinfo->dlease,
1299 req->r_request_started);
1304 /* attach proper inode */
1305 if (d_really_is_negative(dn)) {
1306 ceph_dir_clear_ordered(dir);
1308 dn = splice_dentry(dn, in);
1313 req->r_dentry = dn; /* may have spliced */
1314 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1315 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1316 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1323 update_dentry_lease(dn, rinfo->dlease, session,
1324 req->r_request_started);
1325 dout(" final dn %p\n", dn);
1326 } else if (!req->r_aborted &&
1327 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1328 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1329 struct dentry *dn = req->r_dentry;
1330 struct inode *dir = req->r_locked_dir;
1332 /* fill out a snapdir LOOKUPSNAP dentry */
1335 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1336 dout(" linking snapped dir %p to dn %p\n", in, dn);
1337 ceph_dir_clear_ordered(dir);
1339 dn = splice_dentry(dn, in);
1344 req->r_dentry = dn; /* may have spliced */
1347 dout("fill_trace done err=%d\n", err);
1352 * Prepopulate our cache with readdir results, leases, etc.
1354 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1355 struct ceph_mds_session *session)
1357 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1360 for (i = 0; i < rinfo->dir_nr; i++) {
1361 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1362 struct ceph_vino vino;
1366 vino.ino = le64_to_cpu(rde->inode.in->ino);
1367 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1369 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1372 dout("new_inode badness got %d\n", err);
1375 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1376 req->r_request_started, -1,
1377 &req->r_caps_reservation);
1379 pr_err("fill_inode badness on %p got %d\n", in, rc);
1388 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1392 put_page(ctl->page);
1397 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1398 struct ceph_readdir_cache_control *ctl,
1399 struct ceph_mds_request *req)
1401 struct ceph_inode_info *ci = ceph_inode(dir);
1402 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1403 unsigned idx = ctl->index % nsize;
1404 pgoff_t pgoff = ctl->index / nsize;
1406 if (!ctl->page || pgoff != page_index(ctl->page)) {
1407 ceph_readdir_cache_release(ctl);
1409 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1411 ctl->page = find_lock_page(&dir->i_data, pgoff);
1414 return idx == 0 ? -ENOMEM : 0;
1416 /* reading/filling the cache are serialized by
1417 * i_mutex, no need to use page lock */
1418 unlock_page(ctl->page);
1419 ctl->dentries = kmap(ctl->page);
1421 memset(ctl->dentries, 0, PAGE_SIZE);
1424 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1425 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1426 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1427 ctl->dentries[idx] = dn;
1430 dout("disable readdir cache\n");
1436 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1437 struct ceph_mds_session *session)
1439 struct dentry *parent = req->r_dentry;
1440 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1441 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1445 int err = 0, skipped = 0, ret, i;
1446 struct inode *snapdir = NULL;
1447 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1448 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1451 struct ceph_readdir_cache_control cache_ctl = {};
1454 return readdir_prepopulate_inodes_only(req, session);
1456 if (rinfo->hash_order && req->r_path2) {
1457 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1458 req->r_path2, strlen(req->r_path2));
1459 last_hash = ceph_frag_value(last_hash);
1462 if (rinfo->dir_dir &&
1463 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1464 dout("readdir_prepopulate got new frag %x -> %x\n",
1465 frag, le32_to_cpu(rinfo->dir_dir->frag));
1466 frag = le32_to_cpu(rinfo->dir_dir->frag);
1467 if (!rinfo->hash_order)
1468 req->r_readdir_offset = 2;
1471 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1472 snapdir = ceph_get_snapdir(d_inode(parent));
1473 parent = d_find_alias(snapdir);
1474 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1475 rinfo->dir_nr, parent);
1477 dout("readdir_prepopulate %d items under dn %p\n",
1478 rinfo->dir_nr, parent);
1480 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1483 if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1484 /* note dir version at start of readdir so we can tell
1485 * if any dentries get dropped */
1486 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1487 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1488 req->r_readdir_cache_idx = 0;
1491 cache_ctl.index = req->r_readdir_cache_idx;
1492 fpos_offset = req->r_readdir_offset;
1494 /* FIXME: release caps/leases if error occurs */
1495 for (i = 0; i < rinfo->dir_nr; i++) {
1496 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1497 struct ceph_vino vino;
1499 dname.name = rde->name;
1500 dname.len = rde->name_len;
1501 dname.hash = full_name_hash(dname.name, dname.len);
1503 vino.ino = le64_to_cpu(rde->inode.in->ino);
1504 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1506 if (rinfo->hash_order) {
1507 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1508 rde->name, rde->name_len);
1509 hash = ceph_frag_value(hash);
1510 if (hash != last_hash)
1513 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1515 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1519 dn = d_lookup(parent, &dname);
1520 dout("d_lookup on parent=%p name=%.*s got %p\n",
1521 parent, dname.len, dname.name, dn);
1524 dn = d_alloc(parent, &dname);
1525 dout("d_alloc %p '%.*s' = %p\n", parent,
1526 dname.len, dname.name, dn);
1528 dout("d_alloc badness\n");
1532 ret = ceph_init_dentry(dn);
1538 } else if (d_really_is_positive(dn) &&
1539 (ceph_ino(d_inode(dn)) != vino.ino ||
1540 ceph_snap(d_inode(dn)) != vino.snap)) {
1541 dout(" dn %p points to wrong inode %p\n",
1549 if (d_really_is_positive(dn)) {
1552 in = ceph_get_inode(parent->d_sb, vino);
1554 dout("new_inode badness\n");
1562 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1563 req->r_request_started, -1,
1564 &req->r_caps_reservation);
1566 pr_err("fill_inode badness on %p\n", in);
1567 if (d_really_is_negative(dn))
1574 if (d_really_is_negative(dn)) {
1575 struct dentry *realdn;
1577 if (ceph_security_xattr_deadlock(in)) {
1578 dout(" skip splicing dn %p to inode %p"
1579 " (security xattr deadlock)\n", dn, in);
1585 realdn = splice_dentry(dn, in);
1586 if (IS_ERR(realdn)) {
1587 err = PTR_ERR(realdn);
1595 ceph_dentry(dn)->offset = rde->offset;
1597 update_dentry_lease(dn, rde->lease, req->r_session,
1598 req->r_request_started);
1600 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1601 ret = fill_readdir_cache(d_inode(parent), dn,
1611 if (err == 0 && skipped == 0) {
1612 req->r_did_prepopulate = true;
1613 req->r_readdir_cache_idx = cache_ctl.index;
1615 ceph_readdir_cache_release(&cache_ctl);
1620 dout("readdir_prepopulate done\n");
1624 int ceph_inode_set_size(struct inode *inode, loff_t size)
1626 struct ceph_inode_info *ci = ceph_inode(inode);
1629 spin_lock(&ci->i_ceph_lock);
1630 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1631 i_size_write(inode, size);
1632 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1634 /* tell the MDS if we are approaching max_size */
1635 if ((size << 1) >= ci->i_max_size &&
1636 (ci->i_reported_size << 1) < ci->i_max_size)
1639 spin_unlock(&ci->i_ceph_lock);
1644 * Write back inode data in a worker thread. (This can't be done
1645 * in the message handler context.)
1647 void ceph_queue_writeback(struct inode *inode)
1650 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1651 &ceph_inode(inode)->i_wb_work)) {
1652 dout("ceph_queue_writeback %p\n", inode);
1654 dout("ceph_queue_writeback %p failed\n", inode);
1659 static void ceph_writeback_work(struct work_struct *work)
1661 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1663 struct inode *inode = &ci->vfs_inode;
1665 dout("writeback %p\n", inode);
1666 filemap_fdatawrite(&inode->i_data);
1671 * queue an async invalidation
1673 void ceph_queue_invalidate(struct inode *inode)
1676 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1677 &ceph_inode(inode)->i_pg_inv_work)) {
1678 dout("ceph_queue_invalidate %p\n", inode);
1680 dout("ceph_queue_invalidate %p failed\n", inode);
1686 * Invalidate inode pages in a worker thread. (This can't be done
1687 * in the message handler context.)
1689 static void ceph_invalidate_work(struct work_struct *work)
1691 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1693 struct inode *inode = &ci->vfs_inode;
1694 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1698 mutex_lock(&ci->i_truncate_mutex);
1700 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1701 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1702 inode, ceph_ino(inode));
1703 mapping_set_error(inode->i_mapping, -EIO);
1704 truncate_pagecache(inode, 0);
1705 mutex_unlock(&ci->i_truncate_mutex);
1709 spin_lock(&ci->i_ceph_lock);
1710 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1711 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1712 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1713 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1715 spin_unlock(&ci->i_ceph_lock);
1716 mutex_unlock(&ci->i_truncate_mutex);
1719 orig_gen = ci->i_rdcache_gen;
1720 spin_unlock(&ci->i_ceph_lock);
1722 truncate_pagecache(inode, 0);
1724 spin_lock(&ci->i_ceph_lock);
1725 if (orig_gen == ci->i_rdcache_gen &&
1726 orig_gen == ci->i_rdcache_revoking) {
1727 dout("invalidate_pages %p gen %d successful\n", inode,
1729 ci->i_rdcache_revoking--;
1732 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1733 inode, orig_gen, ci->i_rdcache_gen,
1734 ci->i_rdcache_revoking);
1735 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1738 spin_unlock(&ci->i_ceph_lock);
1739 mutex_unlock(&ci->i_truncate_mutex);
1742 ceph_check_caps(ci, 0, NULL);
1748 * called by trunc_wq;
1750 * We also truncate in a separate thread as well.
1752 static void ceph_vmtruncate_work(struct work_struct *work)
1754 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1756 struct inode *inode = &ci->vfs_inode;
1758 dout("vmtruncate_work %p\n", inode);
1759 __ceph_do_pending_vmtruncate(inode);
1764 * Queue an async vmtruncate. If we fail to queue work, we will handle
1765 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1767 void ceph_queue_vmtruncate(struct inode *inode)
1769 struct ceph_inode_info *ci = ceph_inode(inode);
1773 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1774 &ci->i_vmtruncate_work)) {
1775 dout("ceph_queue_vmtruncate %p\n", inode);
1777 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1778 inode, ci->i_truncate_pending);
1784 * Make sure any pending truncation is applied before doing anything
1785 * that may depend on it.
1787 void __ceph_do_pending_vmtruncate(struct inode *inode)
1789 struct ceph_inode_info *ci = ceph_inode(inode);
1791 int wrbuffer_refs, finish = 0;
1793 mutex_lock(&ci->i_truncate_mutex);
1795 spin_lock(&ci->i_ceph_lock);
1796 if (ci->i_truncate_pending == 0) {
1797 dout("__do_pending_vmtruncate %p none pending\n", inode);
1798 spin_unlock(&ci->i_ceph_lock);
1799 mutex_unlock(&ci->i_truncate_mutex);
1804 * make sure any dirty snapped pages are flushed before we
1805 * possibly truncate them.. so write AND block!
1807 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1808 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1810 spin_unlock(&ci->i_ceph_lock);
1811 filemap_write_and_wait_range(&inode->i_data, 0,
1812 inode->i_sb->s_maxbytes);
1816 /* there should be no reader or writer */
1817 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1819 to = ci->i_truncate_size;
1820 wrbuffer_refs = ci->i_wrbuffer_ref;
1821 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1822 ci->i_truncate_pending, to);
1823 spin_unlock(&ci->i_ceph_lock);
1825 truncate_pagecache(inode, to);
1827 spin_lock(&ci->i_ceph_lock);
1828 if (to == ci->i_truncate_size) {
1829 ci->i_truncate_pending = 0;
1832 spin_unlock(&ci->i_ceph_lock);
1836 mutex_unlock(&ci->i_truncate_mutex);
1838 if (wrbuffer_refs == 0)
1839 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1841 wake_up_all(&ci->i_cap_wq);
1847 static const struct inode_operations ceph_symlink_iops = {
1848 .readlink = generic_readlink,
1849 .get_link = simple_get_link,
1850 .setattr = ceph_setattr,
1851 .getattr = ceph_getattr,
1852 .setxattr = ceph_setxattr,
1853 .getxattr = ceph_getxattr,
1854 .listxattr = ceph_listxattr,
1855 .removexattr = ceph_removexattr,
1861 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1863 struct inode *inode = d_inode(dentry);
1864 struct ceph_inode_info *ci = ceph_inode(inode);
1865 const unsigned int ia_valid = attr->ia_valid;
1866 struct ceph_mds_request *req;
1867 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1868 struct ceph_cap_flush *prealloc_cf;
1870 int release = 0, dirtied = 0;
1873 int inode_dirty_flags = 0;
1874 bool lock_snap_rwsem = false;
1876 if (ceph_snap(inode) != CEPH_NOSNAP)
1879 err = inode_change_ok(inode, attr);
1883 prealloc_cf = ceph_alloc_cap_flush();
1887 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1890 ceph_free_cap_flush(prealloc_cf);
1891 return PTR_ERR(req);
1894 spin_lock(&ci->i_ceph_lock);
1895 issued = __ceph_caps_issued(ci, NULL);
1897 if (!ci->i_head_snapc &&
1898 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1899 lock_snap_rwsem = true;
1900 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1901 spin_unlock(&ci->i_ceph_lock);
1902 down_read(&mdsc->snap_rwsem);
1903 spin_lock(&ci->i_ceph_lock);
1904 issued = __ceph_caps_issued(ci, NULL);
1908 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1910 if (ia_valid & ATTR_UID) {
1911 dout("setattr %p uid %d -> %d\n", inode,
1912 from_kuid(&init_user_ns, inode->i_uid),
1913 from_kuid(&init_user_ns, attr->ia_uid));
1914 if (issued & CEPH_CAP_AUTH_EXCL) {
1915 inode->i_uid = attr->ia_uid;
1916 dirtied |= CEPH_CAP_AUTH_EXCL;
1917 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1918 !uid_eq(attr->ia_uid, inode->i_uid)) {
1919 req->r_args.setattr.uid = cpu_to_le32(
1920 from_kuid(&init_user_ns, attr->ia_uid));
1921 mask |= CEPH_SETATTR_UID;
1922 release |= CEPH_CAP_AUTH_SHARED;
1925 if (ia_valid & ATTR_GID) {
1926 dout("setattr %p gid %d -> %d\n", inode,
1927 from_kgid(&init_user_ns, inode->i_gid),
1928 from_kgid(&init_user_ns, attr->ia_gid));
1929 if (issued & CEPH_CAP_AUTH_EXCL) {
1930 inode->i_gid = attr->ia_gid;
1931 dirtied |= CEPH_CAP_AUTH_EXCL;
1932 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1933 !gid_eq(attr->ia_gid, inode->i_gid)) {
1934 req->r_args.setattr.gid = cpu_to_le32(
1935 from_kgid(&init_user_ns, attr->ia_gid));
1936 mask |= CEPH_SETATTR_GID;
1937 release |= CEPH_CAP_AUTH_SHARED;
1940 if (ia_valid & ATTR_MODE) {
1941 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1943 if (issued & CEPH_CAP_AUTH_EXCL) {
1944 inode->i_mode = attr->ia_mode;
1945 dirtied |= CEPH_CAP_AUTH_EXCL;
1946 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1947 attr->ia_mode != inode->i_mode) {
1948 inode->i_mode = attr->ia_mode;
1949 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1950 mask |= CEPH_SETATTR_MODE;
1951 release |= CEPH_CAP_AUTH_SHARED;
1955 if (ia_valid & ATTR_ATIME) {
1956 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1957 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1958 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1959 if (issued & CEPH_CAP_FILE_EXCL) {
1960 ci->i_time_warp_seq++;
1961 inode->i_atime = attr->ia_atime;
1962 dirtied |= CEPH_CAP_FILE_EXCL;
1963 } else if ((issued & CEPH_CAP_FILE_WR) &&
1964 timespec_compare(&inode->i_atime,
1965 &attr->ia_atime) < 0) {
1966 inode->i_atime = attr->ia_atime;
1967 dirtied |= CEPH_CAP_FILE_WR;
1968 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1969 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1970 ceph_encode_timespec(&req->r_args.setattr.atime,
1972 mask |= CEPH_SETATTR_ATIME;
1973 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1977 if (ia_valid & ATTR_MTIME) {
1978 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1979 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1980 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1981 if (issued & CEPH_CAP_FILE_EXCL) {
1982 ci->i_time_warp_seq++;
1983 inode->i_mtime = attr->ia_mtime;
1984 dirtied |= CEPH_CAP_FILE_EXCL;
1985 } else if ((issued & CEPH_CAP_FILE_WR) &&
1986 timespec_compare(&inode->i_mtime,
1987 &attr->ia_mtime) < 0) {
1988 inode->i_mtime = attr->ia_mtime;
1989 dirtied |= CEPH_CAP_FILE_WR;
1990 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1991 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1992 ceph_encode_timespec(&req->r_args.setattr.mtime,
1994 mask |= CEPH_SETATTR_MTIME;
1995 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1999 if (ia_valid & ATTR_SIZE) {
2000 dout("setattr %p size %lld -> %lld\n", inode,
2001 inode->i_size, attr->ia_size);
2002 if ((issued & CEPH_CAP_FILE_EXCL) &&
2003 attr->ia_size > inode->i_size) {
2004 i_size_write(inode, attr->ia_size);
2006 (attr->ia_size + (1 << 9) - 1) >> 9;
2007 inode->i_ctime = attr->ia_ctime;
2008 ci->i_reported_size = attr->ia_size;
2009 dirtied |= CEPH_CAP_FILE_EXCL;
2010 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2011 attr->ia_size != inode->i_size) {
2012 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2013 req->r_args.setattr.old_size =
2014 cpu_to_le64(inode->i_size);
2015 mask |= CEPH_SETATTR_SIZE;
2016 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2021 /* these do nothing */
2022 if (ia_valid & ATTR_CTIME) {
2023 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2024 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2025 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2026 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2027 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2028 only ? "ctime only" : "ignored");
2029 inode->i_ctime = attr->ia_ctime;
2032 * if kernel wants to dirty ctime but nothing else,
2033 * we need to choose a cap to dirty under, or do
2034 * a almost-no-op setattr
2036 if (issued & CEPH_CAP_AUTH_EXCL)
2037 dirtied |= CEPH_CAP_AUTH_EXCL;
2038 else if (issued & CEPH_CAP_FILE_EXCL)
2039 dirtied |= CEPH_CAP_FILE_EXCL;
2040 else if (issued & CEPH_CAP_XATTR_EXCL)
2041 dirtied |= CEPH_CAP_XATTR_EXCL;
2043 mask |= CEPH_SETATTR_CTIME;
2046 if (ia_valid & ATTR_FILE)
2047 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2050 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2052 inode->i_ctime = current_fs_time(inode->i_sb);
2056 spin_unlock(&ci->i_ceph_lock);
2057 if (lock_snap_rwsem)
2058 up_read(&mdsc->snap_rwsem);
2060 if (inode_dirty_flags)
2061 __mark_inode_dirty(inode, inode_dirty_flags);
2063 if (ia_valid & ATTR_MODE) {
2064 err = posix_acl_chmod(inode, attr->ia_mode);
2070 req->r_inode = inode;
2072 req->r_inode_drop = release;
2073 req->r_args.setattr.mask = cpu_to_le32(mask);
2074 req->r_num_caps = 1;
2075 err = ceph_mdsc_do_request(mdsc, NULL, req);
2077 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2078 ceph_cap_string(dirtied), mask);
2080 ceph_mdsc_put_request(req);
2081 if (mask & CEPH_SETATTR_SIZE)
2082 __ceph_do_pending_vmtruncate(inode);
2083 ceph_free_cap_flush(prealloc_cf);
2086 ceph_mdsc_put_request(req);
2087 ceph_free_cap_flush(prealloc_cf);
2092 * Verify that we have a lease on the given mask. If not,
2093 * do a getattr against an mds.
2095 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2096 int mask, bool force)
2098 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2099 struct ceph_mds_client *mdsc = fsc->mdsc;
2100 struct ceph_mds_request *req;
2103 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2104 dout("do_getattr inode %p SNAPDIR\n", inode);
2108 dout("do_getattr inode %p mask %s mode 0%o\n",
2109 inode, ceph_cap_string(mask), inode->i_mode);
2110 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2113 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2115 return PTR_ERR(req);
2116 req->r_inode = inode;
2118 req->r_num_caps = 1;
2119 req->r_args.getattr.mask = cpu_to_le32(mask);
2120 req->r_locked_page = locked_page;
2121 err = ceph_mdsc_do_request(mdsc, NULL, req);
2122 if (locked_page && err == 0) {
2123 u64 inline_version = req->r_reply_info.targeti.inline_version;
2124 if (inline_version == 0) {
2125 /* the reply is supposed to contain inline data */
2127 } else if (inline_version == CEPH_INLINE_NONE) {
2130 err = req->r_reply_info.targeti.inline_len;
2133 ceph_mdsc_put_request(req);
2134 dout("do_getattr result=%d\n", err);
2140 * Check inode permissions. We verify we have a valid value for
2141 * the AUTH cap, then call the generic handler.
2143 int ceph_permission(struct inode *inode, int mask)
2147 if (mask & MAY_NOT_BLOCK)
2150 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2153 err = generic_permission(inode, mask);
2158 * Get all attributes. Hopefully somedata we'll have a statlite()
2159 * and can limit the fields we require to be accurate.
2161 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2164 struct inode *inode = d_inode(dentry);
2165 struct ceph_inode_info *ci = ceph_inode(inode);
2168 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2170 generic_fillattr(inode, stat);
2171 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2172 if (ceph_snap(inode) != CEPH_NOSNAP)
2173 stat->dev = ceph_snap(inode);
2176 if (S_ISDIR(inode->i_mode)) {
2177 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2179 stat->size = ci->i_rbytes;
2181 stat->size = ci->i_files + ci->i_subdirs;
2183 stat->blksize = 65536;