1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/module.h>
5 #include <linux/slab.h>
6 #include <linux/string.h>
7 #include <linux/uaccess.h>
8 #include <linux/kernel.h>
9 #include <linux/writeback.h>
10 #include <linux/vmalloc.h>
11 #include <linux/xattr.h>
12 #include <linux/posix_acl.h>
13 #include <linux/random.h>
14 #include <linux/sort.h>
17 #include "mds_client.h"
19 #include <linux/ceph/decode.h>
22 * Ceph inode operations
24 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
25 * setattr, etc.), xattr helpers, and helpers for assimilating
26 * metadata returned by the MDS into our cache.
28 * Also define helpers for doing asynchronous writeback, invalidation,
29 * and truncation for the benefit of those who can't afford to block
30 * (typically because they are in the message handler path).
33 static const struct inode_operations ceph_symlink_iops;
35 static void ceph_invalidate_work(struct work_struct *work);
36 static void ceph_writeback_work(struct work_struct *work);
37 static void ceph_vmtruncate_work(struct work_struct *work);
40 * find or create an inode, given the ceph ino number
42 static int ceph_set_ino_cb(struct inode *inode, void *data)
44 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
45 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
49 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
52 ino_t t = ceph_vino_to_ino(vino);
54 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
56 return ERR_PTR(-ENOMEM);
57 if (inode->i_state & I_NEW) {
58 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
59 inode, ceph_vinop(inode), (u64)inode->i_ino);
60 unlock_new_inode(inode);
63 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
69 * get/constuct snapdir inode for a given directory
71 struct inode *ceph_get_snapdir(struct inode *parent)
73 struct ceph_vino vino = {
74 .ino = ceph_ino(parent),
77 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
78 struct ceph_inode_info *ci = ceph_inode(inode);
80 BUG_ON(!S_ISDIR(parent->i_mode));
83 inode->i_mode = parent->i_mode;
84 inode->i_uid = parent->i_uid;
85 inode->i_gid = parent->i_gid;
86 inode->i_op = &ceph_snapdir_iops;
87 inode->i_fop = &ceph_snapdir_fops;
88 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
93 const struct inode_operations ceph_file_iops = {
94 .permission = ceph_permission,
95 .setattr = ceph_setattr,
96 .getattr = ceph_getattr,
97 .setxattr = generic_setxattr,
98 .getxattr = generic_getxattr,
99 .listxattr = ceph_listxattr,
100 .removexattr = generic_removexattr,
101 .get_acl = ceph_get_acl,
102 .set_acl = ceph_set_acl,
107 * We use a 'frag tree' to keep track of the MDS's directory fragments
108 * for a given inode (usually there is just a single fragment). We
109 * need to know when a child frag is delegated to a new MDS, or when
110 * it is flagged as replicated, so we can direct our requests
115 * find/create a frag in the tree
117 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
121 struct rb_node *parent = NULL;
122 struct ceph_inode_frag *frag;
125 p = &ci->i_fragtree.rb_node;
128 frag = rb_entry(parent, struct ceph_inode_frag, node);
129 c = ceph_frag_compare(f, frag->frag);
138 frag = kmalloc(sizeof(*frag), GFP_NOFS);
140 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
141 "frag %x\n", &ci->vfs_inode,
142 ceph_vinop(&ci->vfs_inode), f);
143 return ERR_PTR(-ENOMEM);
150 rb_link_node(&frag->node, parent, p);
151 rb_insert_color(&frag->node, &ci->i_fragtree);
153 dout("get_or_create_frag added %llx.%llx frag %x\n",
154 ceph_vinop(&ci->vfs_inode), f);
159 * find a specific frag @f
161 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
163 struct rb_node *n = ci->i_fragtree.rb_node;
166 struct ceph_inode_frag *frag =
167 rb_entry(n, struct ceph_inode_frag, node);
168 int c = ceph_frag_compare(f, frag->frag);
180 * Choose frag containing the given value @v. If @pfrag is
181 * specified, copy the frag delegation info to the caller if
184 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
185 struct ceph_inode_frag *pfrag, int *found)
187 u32 t = ceph_frag_make(0, 0);
188 struct ceph_inode_frag *frag;
196 WARN_ON(!ceph_frag_contains_value(t, v));
197 frag = __ceph_find_frag(ci, t);
199 break; /* t is a leaf */
200 if (frag->split_by == 0) {
202 memcpy(pfrag, frag, sizeof(*pfrag));
209 nway = 1 << frag->split_by;
210 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
211 frag->split_by, nway);
212 for (i = 0; i < nway; i++) {
213 n = ceph_frag_make_child(t, frag->split_by, i);
214 if (ceph_frag_contains_value(n, v)) {
221 dout("choose_frag(%x) = %x\n", v, t);
226 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
227 struct ceph_inode_frag *pfrag, int *found)
230 mutex_lock(&ci->i_fragtree_mutex);
231 ret = __ceph_choose_frag(ci, v, pfrag, found);
232 mutex_unlock(&ci->i_fragtree_mutex);
237 * Process dirfrag (delegation) info from the mds. Include leaf
238 * fragment in tree ONLY if ndist > 0. Otherwise, only
239 * branches/splits are included in i_fragtree)
241 static int ceph_fill_dirfrag(struct inode *inode,
242 struct ceph_mds_reply_dirfrag *dirinfo)
244 struct ceph_inode_info *ci = ceph_inode(inode);
245 struct ceph_inode_frag *frag;
246 u32 id = le32_to_cpu(dirinfo->frag);
247 int mds = le32_to_cpu(dirinfo->auth);
248 int ndist = le32_to_cpu(dirinfo->ndist);
253 spin_lock(&ci->i_ceph_lock);
255 diri_auth = ci->i_auth_cap->mds;
256 spin_unlock(&ci->i_ceph_lock);
258 if (mds == -1) /* CDIR_AUTH_PARENT */
261 mutex_lock(&ci->i_fragtree_mutex);
262 if (ndist == 0 && mds == diri_auth) {
263 /* no delegation info needed. */
264 frag = __ceph_find_frag(ci, id);
267 if (frag->split_by == 0) {
268 /* tree leaf, remove */
269 dout("fill_dirfrag removed %llx.%llx frag %x"
270 " (no ref)\n", ceph_vinop(inode), id);
271 rb_erase(&frag->node, &ci->i_fragtree);
274 /* tree branch, keep and clear */
275 dout("fill_dirfrag cleared %llx.%llx frag %x"
276 " referral\n", ceph_vinop(inode), id);
284 /* find/add this frag to store mds delegation info */
285 frag = __get_or_create_frag(ci, id);
287 /* this is not the end of the world; we can continue
288 with bad/inaccurate delegation info */
289 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
290 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
296 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
297 for (i = 0; i < frag->ndist; i++)
298 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
299 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
300 ceph_vinop(inode), frag->frag, frag->ndist);
303 mutex_unlock(&ci->i_fragtree_mutex);
307 static int frag_tree_split_cmp(const void *l, const void *r)
309 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
310 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
311 return ceph_frag_compare(ls->frag, rs->frag);
314 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
317 return f == ceph_frag_make(0, 0);
318 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
320 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
323 static int ceph_fill_fragtree(struct inode *inode,
324 struct ceph_frag_tree_head *fragtree,
325 struct ceph_mds_reply_dirfrag *dirinfo)
327 struct ceph_inode_info *ci = ceph_inode(inode);
328 struct ceph_inode_frag *frag, *prev_frag = NULL;
329 struct rb_node *rb_node;
330 unsigned i, split_by, nsplits;
334 mutex_lock(&ci->i_fragtree_mutex);
335 nsplits = le32_to_cpu(fragtree->nsplits);
336 if (nsplits != ci->i_fragtree_nsplits) {
338 } else if (nsplits) {
339 i = prandom_u32() % nsplits;
340 id = le32_to_cpu(fragtree->splits[i].frag);
341 if (!__ceph_find_frag(ci, id))
343 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
344 rb_node = rb_first(&ci->i_fragtree);
345 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
346 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
349 if (!update && dirinfo) {
350 id = le32_to_cpu(dirinfo->frag);
351 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
358 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
359 frag_tree_split_cmp, NULL);
362 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
363 rb_node = rb_first(&ci->i_fragtree);
364 for (i = 0; i < nsplits; i++) {
365 id = le32_to_cpu(fragtree->splits[i].frag);
366 split_by = le32_to_cpu(fragtree->splits[i].by);
367 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
368 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
369 "frag %x split by %d\n", ceph_vinop(inode),
370 i, nsplits, id, split_by);
375 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
376 if (ceph_frag_compare(frag->frag, id) >= 0) {
377 if (frag->frag != id)
380 rb_node = rb_next(rb_node);
383 rb_node = rb_next(rb_node);
384 /* delete stale split/leaf node */
385 if (frag->split_by > 0 ||
386 !is_frag_child(frag->frag, prev_frag)) {
387 rb_erase(&frag->node, &ci->i_fragtree);
388 if (frag->split_by > 0)
389 ci->i_fragtree_nsplits--;
395 frag = __get_or_create_frag(ci, id);
399 if (frag->split_by == 0)
400 ci->i_fragtree_nsplits++;
401 frag->split_by = split_by;
402 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
406 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
407 rb_node = rb_next(rb_node);
408 /* delete stale split/leaf node */
409 if (frag->split_by > 0 ||
410 !is_frag_child(frag->frag, prev_frag)) {
411 rb_erase(&frag->node, &ci->i_fragtree);
412 if (frag->split_by > 0)
413 ci->i_fragtree_nsplits--;
418 mutex_unlock(&ci->i_fragtree_mutex);
423 * initialize a newly allocated inode.
425 struct inode *ceph_alloc_inode(struct super_block *sb)
427 struct ceph_inode_info *ci;
430 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
434 dout("alloc_inode %p\n", &ci->vfs_inode);
436 spin_lock_init(&ci->i_ceph_lock);
439 ci->i_inline_version = 0;
440 ci->i_time_warp_seq = 0;
441 ci->i_ceph_flags = 0;
442 atomic64_set(&ci->i_ordered_count, 1);
443 atomic64_set(&ci->i_release_count, 1);
444 atomic64_set(&ci->i_complete_seq[0], 0);
445 atomic64_set(&ci->i_complete_seq[1], 0);
446 ci->i_symlink = NULL;
448 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
449 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
451 ci->i_fragtree = RB_ROOT;
452 mutex_init(&ci->i_fragtree_mutex);
454 ci->i_xattrs.blob = NULL;
455 ci->i_xattrs.prealloc_blob = NULL;
456 ci->i_xattrs.dirty = false;
457 ci->i_xattrs.index = RB_ROOT;
458 ci->i_xattrs.count = 0;
459 ci->i_xattrs.names_size = 0;
460 ci->i_xattrs.vals_size = 0;
461 ci->i_xattrs.version = 0;
462 ci->i_xattrs.index_version = 0;
464 ci->i_caps = RB_ROOT;
465 ci->i_auth_cap = NULL;
466 ci->i_dirty_caps = 0;
467 ci->i_flushing_caps = 0;
468 INIT_LIST_HEAD(&ci->i_dirty_item);
469 INIT_LIST_HEAD(&ci->i_flushing_item);
470 ci->i_prealloc_cap_flush = NULL;
471 ci->i_cap_flush_tree = RB_ROOT;
472 init_waitqueue_head(&ci->i_cap_wq);
473 ci->i_hold_caps_min = 0;
474 ci->i_hold_caps_max = 0;
475 INIT_LIST_HEAD(&ci->i_cap_delay_list);
476 INIT_LIST_HEAD(&ci->i_cap_snaps);
477 ci->i_head_snapc = NULL;
480 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
481 ci->i_nr_by_mode[i] = 0;
483 mutex_init(&ci->i_truncate_mutex);
484 ci->i_truncate_seq = 0;
485 ci->i_truncate_size = 0;
486 ci->i_truncate_pending = 0;
489 ci->i_reported_size = 0;
490 ci->i_wanted_max_size = 0;
491 ci->i_requested_max_size = 0;
495 ci->i_rdcache_ref = 0;
498 ci->i_wrbuffer_ref = 0;
499 ci->i_wrbuffer_ref_head = 0;
500 ci->i_shared_gen = 0;
501 ci->i_rdcache_gen = 0;
502 ci->i_rdcache_revoking = 0;
504 INIT_LIST_HEAD(&ci->i_unsafe_writes);
505 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
506 INIT_LIST_HEAD(&ci->i_unsafe_iops);
507 spin_lock_init(&ci->i_unsafe_lock);
509 ci->i_snap_realm = NULL;
510 INIT_LIST_HEAD(&ci->i_snap_realm_item);
511 INIT_LIST_HEAD(&ci->i_snap_flush_item);
513 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
514 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
516 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
518 ceph_fscache_inode_init(ci);
520 return &ci->vfs_inode;
523 static void ceph_i_callback(struct rcu_head *head)
525 struct inode *inode = container_of(head, struct inode, i_rcu);
526 struct ceph_inode_info *ci = ceph_inode(inode);
528 kmem_cache_free(ceph_inode_cachep, ci);
531 void ceph_destroy_inode(struct inode *inode)
533 struct ceph_inode_info *ci = ceph_inode(inode);
534 struct ceph_inode_frag *frag;
537 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
539 ceph_fscache_unregister_inode_cookie(ci);
541 ceph_queue_caps_release(inode);
544 * we may still have a snap_realm reference if there are stray
545 * caps in i_snap_caps.
547 if (ci->i_snap_realm) {
548 struct ceph_mds_client *mdsc =
549 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
550 struct ceph_snap_realm *realm = ci->i_snap_realm;
552 dout(" dropping residual ref to snap realm %p\n", realm);
553 spin_lock(&realm->inodes_with_caps_lock);
554 list_del_init(&ci->i_snap_realm_item);
555 spin_unlock(&realm->inodes_with_caps_lock);
556 ceph_put_snap_realm(mdsc, realm);
559 kfree(ci->i_symlink);
560 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
561 frag = rb_entry(n, struct ceph_inode_frag, node);
562 rb_erase(n, &ci->i_fragtree);
565 ci->i_fragtree_nsplits = 0;
567 __ceph_destroy_xattrs(ci);
568 if (ci->i_xattrs.blob)
569 ceph_buffer_put(ci->i_xattrs.blob);
570 if (ci->i_xattrs.prealloc_blob)
571 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
573 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
575 call_rcu(&inode->i_rcu, ceph_i_callback);
578 int ceph_drop_inode(struct inode *inode)
581 * Positve dentry and corresponding inode are always accompanied
582 * in MDS reply. So no need to keep inode in the cache after
583 * dropping all its aliases.
588 static inline blkcnt_t calc_inode_blocks(u64 size)
590 return (size + (1<<9) - 1) >> 9;
594 * Helpers to fill in size, ctime, mtime, and atime. We have to be
595 * careful because either the client or MDS may have more up to date
596 * info, depending on which capabilities are held, and whether
597 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
598 * and size are monotonically increasing, except when utimes() or
599 * truncate() increments the corresponding _seq values.)
601 int ceph_fill_file_size(struct inode *inode, int issued,
602 u32 truncate_seq, u64 truncate_size, u64 size)
604 struct ceph_inode_info *ci = ceph_inode(inode);
607 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
608 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
609 dout("size %lld -> %llu\n", inode->i_size, size);
610 if (size > 0 && S_ISDIR(inode->i_mode)) {
611 pr_err("fill_file_size non-zero size for directory\n");
614 i_size_write(inode, size);
615 inode->i_blocks = calc_inode_blocks(size);
616 ci->i_reported_size = size;
617 if (truncate_seq != ci->i_truncate_seq) {
618 dout("truncate_seq %u -> %u\n",
619 ci->i_truncate_seq, truncate_seq);
620 ci->i_truncate_seq = truncate_seq;
622 /* the MDS should have revoked these caps */
623 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
626 CEPH_CAP_FILE_LAZYIO));
628 * If we hold relevant caps, or in the case where we're
629 * not the only client referencing this file and we
630 * don't hold those caps, then we need to check whether
631 * the file is either opened or mmaped
633 if ((issued & (CEPH_CAP_FILE_CACHE|
634 CEPH_CAP_FILE_BUFFER)) ||
635 mapping_mapped(inode->i_mapping) ||
636 __ceph_caps_file_wanted(ci)) {
637 ci->i_truncate_pending++;
642 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
643 ci->i_truncate_size != truncate_size) {
644 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
646 ci->i_truncate_size = truncate_size;
650 ceph_fscache_invalidate(inode);
655 void ceph_fill_file_time(struct inode *inode, int issued,
656 u64 time_warp_seq, struct timespec *ctime,
657 struct timespec *mtime, struct timespec *atime)
659 struct ceph_inode_info *ci = ceph_inode(inode);
662 if (issued & (CEPH_CAP_FILE_EXCL|
664 CEPH_CAP_FILE_BUFFER|
666 CEPH_CAP_XATTR_EXCL)) {
667 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
668 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
669 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
670 ctime->tv_sec, ctime->tv_nsec);
671 inode->i_ctime = *ctime;
673 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
674 /* the MDS did a utimes() */
675 dout("mtime %ld.%09ld -> %ld.%09ld "
677 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
678 mtime->tv_sec, mtime->tv_nsec,
679 ci->i_time_warp_seq, (int)time_warp_seq);
681 inode->i_mtime = *mtime;
682 inode->i_atime = *atime;
683 ci->i_time_warp_seq = time_warp_seq;
684 } else if (time_warp_seq == ci->i_time_warp_seq) {
685 /* nobody did utimes(); take the max */
686 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
687 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
688 inode->i_mtime.tv_sec,
689 inode->i_mtime.tv_nsec,
690 mtime->tv_sec, mtime->tv_nsec);
691 inode->i_mtime = *mtime;
693 if (timespec_compare(atime, &inode->i_atime) > 0) {
694 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
695 inode->i_atime.tv_sec,
696 inode->i_atime.tv_nsec,
697 atime->tv_sec, atime->tv_nsec);
698 inode->i_atime = *atime;
700 } else if (issued & CEPH_CAP_FILE_EXCL) {
701 /* we did a utimes(); ignore mds values */
706 /* we have no write|excl caps; whatever the MDS says is true */
707 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
708 inode->i_ctime = *ctime;
709 inode->i_mtime = *mtime;
710 inode->i_atime = *atime;
711 ci->i_time_warp_seq = time_warp_seq;
716 if (warn) /* time_warp_seq shouldn't go backwards */
717 dout("%p mds time_warp_seq %llu < %u\n",
718 inode, time_warp_seq, ci->i_time_warp_seq);
722 * Populate an inode based on info from mds. May be called on new or
725 static int fill_inode(struct inode *inode, struct page *locked_page,
726 struct ceph_mds_reply_info_in *iinfo,
727 struct ceph_mds_reply_dirfrag *dirinfo,
728 struct ceph_mds_session *session,
729 unsigned long ttl_from, int cap_fmode,
730 struct ceph_cap_reservation *caps_reservation)
732 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
733 struct ceph_mds_reply_inode *info = iinfo->in;
734 struct ceph_inode_info *ci = ceph_inode(inode);
735 int issued = 0, implemented, new_issued;
736 struct timespec mtime, atime, ctime;
737 struct ceph_buffer *xattr_blob = NULL;
738 struct ceph_string *pool_ns = NULL;
739 struct ceph_cap *new_cap = NULL;
742 bool queue_trunc = false;
743 bool new_version = false;
744 bool fill_inline = false;
746 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
747 inode, ceph_vinop(inode), le64_to_cpu(info->version),
750 /* prealloc new cap struct */
751 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
752 new_cap = ceph_get_cap(mdsc, caps_reservation);
755 * prealloc xattr data, if it looks like we'll need it. only
756 * if len > 4 (meaning there are actually xattrs; the first 4
757 * bytes are the xattr count).
759 if (iinfo->xattr_len > 4) {
760 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
762 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
766 if (iinfo->pool_ns_len > 0)
767 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
770 spin_lock(&ci->i_ceph_lock);
773 * provided version will be odd if inode value is projected,
774 * even if stable. skip the update if we have newer stable
775 * info (ours>=theirs, e.g. due to racing mds replies), unless
776 * we are getting projected (unstable) info (in which case the
777 * version is odd, and we want ours>theirs).
783 if (ci->i_version == 0 ||
784 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
785 le64_to_cpu(info->version) > (ci->i_version & ~1)))
788 issued = __ceph_caps_issued(ci, &implemented);
789 issued |= implemented | __ceph_caps_dirty(ci);
790 new_issued = ~issued & le32_to_cpu(info->cap.caps);
793 ci->i_version = le64_to_cpu(info->version);
795 inode->i_rdev = le32_to_cpu(info->rdev);
796 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
798 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
799 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
800 inode->i_mode = le32_to_cpu(info->mode);
801 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
802 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
803 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
804 from_kuid(&init_user_ns, inode->i_uid),
805 from_kgid(&init_user_ns, inode->i_gid));
808 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
809 (issued & CEPH_CAP_LINK_EXCL) == 0)
810 set_nlink(inode, le32_to_cpu(info->nlink));
812 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
813 /* be careful with mtime, atime, size */
814 ceph_decode_timespec(&atime, &info->atime);
815 ceph_decode_timespec(&mtime, &info->mtime);
816 ceph_decode_timespec(&ctime, &info->ctime);
817 ceph_fill_file_time(inode, issued,
818 le32_to_cpu(info->time_warp_seq),
819 &ctime, &mtime, &atime);
823 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
824 s64 old_pool = ci->i_layout.pool_id;
825 struct ceph_string *old_ns;
827 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
828 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
829 lockdep_is_held(&ci->i_ceph_lock));
830 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
832 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
833 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
837 queue_trunc = ceph_fill_file_size(inode, issued,
838 le32_to_cpu(info->truncate_seq),
839 le64_to_cpu(info->truncate_size),
840 le64_to_cpu(info->size));
841 /* only update max_size on auth cap */
842 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
843 ci->i_max_size != le64_to_cpu(info->max_size)) {
844 dout("max_size %lld -> %llu\n", ci->i_max_size,
845 le64_to_cpu(info->max_size));
846 ci->i_max_size = le64_to_cpu(info->max_size);
851 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
852 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
853 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
854 if (ci->i_xattrs.blob)
855 ceph_buffer_put(ci->i_xattrs.blob);
856 ci->i_xattrs.blob = xattr_blob;
858 memcpy(ci->i_xattrs.blob->vec.iov_base,
859 iinfo->xattr_data, iinfo->xattr_len);
860 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
861 ceph_forget_all_cached_acls(inode);
865 inode->i_mapping->a_ops = &ceph_aops;
867 switch (inode->i_mode & S_IFMT) {
872 init_special_inode(inode, inode->i_mode, inode->i_rdev);
873 inode->i_op = &ceph_file_iops;
876 inode->i_op = &ceph_file_iops;
877 inode->i_fop = &ceph_file_fops;
880 inode->i_op = &ceph_symlink_iops;
881 if (!ci->i_symlink) {
882 u32 symlen = iinfo->symlink_len;
885 spin_unlock(&ci->i_ceph_lock);
887 if (symlen != i_size_read(inode)) {
888 pr_err("fill_inode %llx.%llx BAD symlink "
889 "size %lld\n", ceph_vinop(inode),
891 i_size_write(inode, symlen);
892 inode->i_blocks = calc_inode_blocks(symlen);
896 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
900 spin_lock(&ci->i_ceph_lock);
904 kfree(sym); /* lost a race */
906 inode->i_link = ci->i_symlink;
909 inode->i_op = &ceph_dir_iops;
910 inode->i_fop = &ceph_dir_fops;
912 ci->i_dir_layout = iinfo->dir_layout;
914 ci->i_files = le64_to_cpu(info->files);
915 ci->i_subdirs = le64_to_cpu(info->subdirs);
916 ci->i_rbytes = le64_to_cpu(info->rbytes);
917 ci->i_rfiles = le64_to_cpu(info->rfiles);
918 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
919 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
922 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
923 ceph_vinop(inode), inode->i_mode);
926 /* were we issued a capability? */
927 if (info->cap.caps) {
928 if (ceph_snap(inode) == CEPH_NOSNAP) {
929 unsigned caps = le32_to_cpu(info->cap.caps);
930 ceph_add_cap(inode, session,
931 le64_to_cpu(info->cap.cap_id),
933 le32_to_cpu(info->cap.wanted),
934 le32_to_cpu(info->cap.seq),
935 le32_to_cpu(info->cap.mseq),
936 le64_to_cpu(info->cap.realm),
937 info->cap.flags, &new_cap);
939 /* set dir completion flag? */
940 if (S_ISDIR(inode->i_mode) &&
941 ci->i_files == 0 && ci->i_subdirs == 0 &&
942 (caps & CEPH_CAP_FILE_SHARED) &&
943 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
944 !__ceph_dir_is_complete(ci)) {
945 dout(" marking %p complete (empty)\n", inode);
946 i_size_write(inode, 0);
947 __ceph_dir_set_complete(ci,
948 atomic64_read(&ci->i_release_count),
949 atomic64_read(&ci->i_ordered_count));
954 dout(" %p got snap_caps %s\n", inode,
955 ceph_cap_string(le32_to_cpu(info->cap.caps)));
956 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
958 __ceph_get_fmode(ci, cap_fmode);
960 } else if (cap_fmode >= 0) {
961 pr_warn("mds issued no caps on %llx.%llx\n",
963 __ceph_get_fmode(ci, cap_fmode);
966 if (iinfo->inline_version > 0 &&
967 iinfo->inline_version >= ci->i_inline_version) {
968 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
969 ci->i_inline_version = iinfo->inline_version;
970 if (ci->i_inline_version != CEPH_INLINE_NONE &&
972 (le32_to_cpu(info->cap.caps) & cache_caps)))
976 spin_unlock(&ci->i_ceph_lock);
979 ceph_fill_inline_data(inode, locked_page,
980 iinfo->inline_data, iinfo->inline_len);
983 wake_up_all(&ci->i_cap_wq);
985 /* queue truncate if we saw i_size decrease */
987 ceph_queue_vmtruncate(inode);
989 /* populate frag tree */
990 if (S_ISDIR(inode->i_mode))
991 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
993 /* update delegation info? */
995 ceph_fill_dirfrag(inode, dirinfo);
1000 ceph_put_cap(mdsc, new_cap);
1002 ceph_buffer_put(xattr_blob);
1003 ceph_put_string(pool_ns);
1008 * caller should hold session s_mutex.
1010 static void update_dentry_lease(struct dentry *dentry,
1011 struct ceph_mds_reply_lease *lease,
1012 struct ceph_mds_session *session,
1013 unsigned long from_time)
1015 struct ceph_dentry_info *di = ceph_dentry(dentry);
1016 long unsigned duration = le32_to_cpu(lease->duration_ms);
1017 long unsigned ttl = from_time + (duration * HZ) / 1000;
1018 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1021 /* only track leases on regular dentries */
1022 if (dentry->d_op != &ceph_dentry_ops)
1025 spin_lock(&dentry->d_lock);
1026 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1027 dentry, duration, ttl);
1029 /* make lease_rdcache_gen match directory */
1030 dir = d_inode(dentry->d_parent);
1031 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1036 if (di->lease_gen == session->s_cap_gen &&
1037 time_before(ttl, dentry->d_time))
1038 goto out_unlock; /* we already have a newer lease. */
1040 if (di->lease_session && di->lease_session != session)
1043 ceph_dentry_lru_touch(dentry);
1045 if (!di->lease_session)
1046 di->lease_session = ceph_get_mds_session(session);
1047 di->lease_gen = session->s_cap_gen;
1048 di->lease_seq = le32_to_cpu(lease->seq);
1049 di->lease_renew_after = half_ttl;
1050 di->lease_renew_from = 0;
1051 dentry->d_time = ttl;
1053 spin_unlock(&dentry->d_lock);
1058 * splice a dentry to an inode.
1059 * caller must hold directory i_mutex for this to be safe.
1061 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1063 struct dentry *realdn;
1065 BUG_ON(d_inode(dn));
1067 /* dn must be unhashed */
1068 if (!d_unhashed(dn))
1070 realdn = d_splice_alias(in, dn);
1071 if (IS_ERR(realdn)) {
1072 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1073 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1074 dn = realdn; /* note realdn contains the error */
1076 } else if (realdn) {
1077 dout("dn %p (%d) spliced with %p (%d) "
1078 "inode %p ino %llx.%llx\n",
1080 realdn, d_count(realdn),
1081 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1085 BUG_ON(!ceph_dentry(dn));
1086 dout("dn %p attached to %p ino %llx.%llx\n",
1087 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1094 * Incorporate results into the local cache. This is either just
1095 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1098 * A reply may contain
1099 * a directory inode along with a dentry.
1100 * and/or a target inode
1102 * Called with snap_rwsem (read).
1104 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1105 struct ceph_mds_session *session)
1107 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1108 struct inode *in = NULL;
1109 struct ceph_vino vino;
1110 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1113 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1114 rinfo->head->is_dentry, rinfo->head->is_target);
1120 * If we resend completed ops to a recovering mds, we get no
1121 * trace. Since that is very rare, pretend this is the case
1122 * to ensure the 'no trace' handlers in the callers behave.
1124 * Fill in inodes unconditionally to avoid breaking cap
1127 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1128 pr_info("fill_trace faking empty trace on %lld %s\n",
1129 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1130 if (rinfo->head->is_dentry) {
1131 rinfo->head->is_dentry = 0;
1132 err = fill_inode(req->r_locked_dir,
1133 &rinfo->diri, rinfo->dirfrag,
1134 session, req->r_request_started, -1);
1136 if (rinfo->head->is_target) {
1137 rinfo->head->is_target = 0;
1138 ininfo = rinfo->targeti.in;
1139 vino.ino = le64_to_cpu(ininfo->ino);
1140 vino.snap = le64_to_cpu(ininfo->snapid);
1141 in = ceph_get_inode(sb, vino);
1142 err = fill_inode(in, &rinfo->targeti, NULL,
1143 session, req->r_request_started,
1150 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1151 dout("fill_trace reply is empty!\n");
1152 if (rinfo->head->result == 0 && req->r_locked_dir)
1153 ceph_invalidate_dir_request(req);
1157 if (rinfo->head->is_dentry) {
1158 struct inode *dir = req->r_locked_dir;
1161 err = fill_inode(dir, NULL,
1162 &rinfo->diri, rinfo->dirfrag,
1163 session, req->r_request_started, -1,
1164 &req->r_caps_reservation);
1171 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1173 struct dentry *dn, *parent;
1175 BUG_ON(!rinfo->head->is_target);
1176 BUG_ON(req->r_dentry);
1178 parent = d_find_any_alias(dir);
1181 dname.name = rinfo->dname;
1182 dname.len = rinfo->dname_len;
1183 dname.hash = full_name_hash(dname.name, dname.len);
1184 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1185 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1187 dn = d_lookup(parent, &dname);
1188 dout("d_lookup on parent=%p name=%.*s got %p\n",
1189 parent, dname.len, dname.name, dn);
1192 dn = d_alloc(parent, &dname);
1193 dout("d_alloc %p '%.*s' = %p\n", parent,
1194 dname.len, dname.name, dn);
1200 err = ceph_init_dentry(dn);
1206 } else if (d_really_is_positive(dn) &&
1207 (ceph_ino(d_inode(dn)) != vino.ino ||
1208 ceph_snap(d_inode(dn)) != vino.snap)) {
1209 dout(" dn %p points to wrong inode %p\n",
1221 if (rinfo->head->is_target) {
1222 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1223 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1225 in = ceph_get_inode(sb, vino);
1230 req->r_target_inode = in;
1232 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1233 session, req->r_request_started,
1234 (!req->r_aborted && rinfo->head->result == 0) ?
1236 &req->r_caps_reservation);
1238 pr_err("fill_inode badness %p %llx.%llx\n",
1239 in, ceph_vinop(in));
1245 * ignore null lease/binding on snapdir ENOENT, or else we
1246 * will have trouble splicing in the virtual snapdir later
1248 if (rinfo->head->is_dentry && !req->r_aborted &&
1249 req->r_locked_dir &&
1250 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1251 fsc->mount_options->snapdir_name,
1252 req->r_dentry->d_name.len))) {
1254 * lookup link rename : null -> possibly existing inode
1255 * mknod symlink mkdir : null -> new inode
1256 * unlink : linked -> null
1258 struct inode *dir = req->r_locked_dir;
1259 struct dentry *dn = req->r_dentry;
1260 bool have_dir_cap, have_lease;
1264 BUG_ON(d_inode(dn->d_parent) != dir);
1265 BUG_ON(ceph_ino(dir) !=
1266 le64_to_cpu(rinfo->diri.in->ino));
1267 BUG_ON(ceph_snap(dir) !=
1268 le64_to_cpu(rinfo->diri.in->snapid));
1270 /* do we have a lease on the whole dir? */
1272 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1273 CEPH_CAP_FILE_SHARED);
1275 /* do we have a dn lease? */
1276 have_lease = have_dir_cap ||
1277 le32_to_cpu(rinfo->dlease->duration_ms);
1279 dout("fill_trace no dentry lease or dir cap\n");
1282 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1283 struct inode *olddir = req->r_old_dentry_dir;
1286 dout(" src %p '%pd' dst %p '%pd'\n",
1290 dout("fill_trace doing d_move %p -> %p\n",
1291 req->r_old_dentry, dn);
1293 /* d_move screws up sibling dentries' offsets */
1294 ceph_dir_clear_ordered(dir);
1295 ceph_dir_clear_ordered(olddir);
1297 d_move(req->r_old_dentry, dn);
1298 dout(" src %p '%pd' dst %p '%pd'\n",
1303 /* ensure target dentry is invalidated, despite
1304 rehashing bug in vfs_rename_dir */
1305 ceph_invalidate_dentry_lease(dn);
1307 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1308 ceph_dentry(req->r_old_dentry)->offset);
1310 dn = req->r_old_dentry; /* use old_dentry */
1314 if (!rinfo->head->is_target) {
1315 dout("fill_trace null dentry\n");
1316 if (d_really_is_positive(dn)) {
1317 ceph_dir_clear_ordered(dir);
1318 dout("d_delete %p\n", dn);
1321 if (have_lease && d_unhashed(dn))
1323 update_dentry_lease(dn, rinfo->dlease,
1325 req->r_request_started);
1330 /* attach proper inode */
1331 if (d_really_is_negative(dn)) {
1332 ceph_dir_clear_ordered(dir);
1334 dn = splice_dentry(dn, in);
1339 req->r_dentry = dn; /* may have spliced */
1340 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1341 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1342 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1349 update_dentry_lease(dn, rinfo->dlease, session,
1350 req->r_request_started);
1351 dout(" final dn %p\n", dn);
1352 } else if (!req->r_aborted &&
1353 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1354 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1355 struct dentry *dn = req->r_dentry;
1356 struct inode *dir = req->r_locked_dir;
1358 /* fill out a snapdir LOOKUPSNAP dentry */
1361 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1362 dout(" linking snapped dir %p to dn %p\n", in, dn);
1363 ceph_dir_clear_ordered(dir);
1365 dn = splice_dentry(dn, in);
1370 req->r_dentry = dn; /* may have spliced */
1373 dout("fill_trace done err=%d\n", err);
1378 * Prepopulate our cache with readdir results, leases, etc.
1380 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1381 struct ceph_mds_session *session)
1383 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1386 for (i = 0; i < rinfo->dir_nr; i++) {
1387 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1388 struct ceph_vino vino;
1392 vino.ino = le64_to_cpu(rde->inode.in->ino);
1393 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1395 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1398 dout("new_inode badness got %d\n", err);
1401 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1402 req->r_request_started, -1,
1403 &req->r_caps_reservation);
1405 pr_err("fill_inode badness on %p got %d\n", in, rc);
1414 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1418 put_page(ctl->page);
1423 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1424 struct ceph_readdir_cache_control *ctl,
1425 struct ceph_mds_request *req)
1427 struct ceph_inode_info *ci = ceph_inode(dir);
1428 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1429 unsigned idx = ctl->index % nsize;
1430 pgoff_t pgoff = ctl->index / nsize;
1432 if (!ctl->page || pgoff != page_index(ctl->page)) {
1433 ceph_readdir_cache_release(ctl);
1435 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1437 ctl->page = find_lock_page(&dir->i_data, pgoff);
1440 return idx == 0 ? -ENOMEM : 0;
1442 /* reading/filling the cache are serialized by
1443 * i_mutex, no need to use page lock */
1444 unlock_page(ctl->page);
1445 ctl->dentries = kmap(ctl->page);
1447 memset(ctl->dentries, 0, PAGE_SIZE);
1450 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1451 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1452 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1453 ctl->dentries[idx] = dn;
1456 dout("disable readdir cache\n");
1462 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1463 struct ceph_mds_session *session)
1465 struct dentry *parent = req->r_dentry;
1466 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1467 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1471 int err = 0, skipped = 0, ret, i;
1472 struct inode *snapdir = NULL;
1473 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1474 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1477 struct ceph_readdir_cache_control cache_ctl = {};
1480 return readdir_prepopulate_inodes_only(req, session);
1482 if (rinfo->hash_order && req->r_path2) {
1483 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1484 req->r_path2, strlen(req->r_path2));
1485 last_hash = ceph_frag_value(last_hash);
1488 if (rinfo->dir_dir &&
1489 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1490 dout("readdir_prepopulate got new frag %x -> %x\n",
1491 frag, le32_to_cpu(rinfo->dir_dir->frag));
1492 frag = le32_to_cpu(rinfo->dir_dir->frag);
1493 if (!rinfo->hash_order)
1494 req->r_readdir_offset = 2;
1497 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1498 snapdir = ceph_get_snapdir(d_inode(parent));
1499 parent = d_find_alias(snapdir);
1500 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1501 rinfo->dir_nr, parent);
1503 dout("readdir_prepopulate %d items under dn %p\n",
1504 rinfo->dir_nr, parent);
1506 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1509 if (ceph_frag_is_leftmost(frag) && req->r_readdir_offset == 2) {
1510 /* note dir version at start of readdir so we can tell
1511 * if any dentries get dropped */
1512 req->r_dir_release_cnt = atomic64_read(&ci->i_release_count);
1513 req->r_dir_ordered_cnt = atomic64_read(&ci->i_ordered_count);
1514 req->r_readdir_cache_idx = 0;
1517 cache_ctl.index = req->r_readdir_cache_idx;
1518 fpos_offset = req->r_readdir_offset;
1520 /* FIXME: release caps/leases if error occurs */
1521 for (i = 0; i < rinfo->dir_nr; i++) {
1522 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1523 struct ceph_vino vino;
1525 dname.name = rde->name;
1526 dname.len = rde->name_len;
1527 dname.hash = full_name_hash(dname.name, dname.len);
1529 vino.ino = le64_to_cpu(rde->inode.in->ino);
1530 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1532 if (rinfo->hash_order) {
1533 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1534 rde->name, rde->name_len);
1535 hash = ceph_frag_value(hash);
1536 if (hash != last_hash)
1539 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1541 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1545 dn = d_lookup(parent, &dname);
1546 dout("d_lookup on parent=%p name=%.*s got %p\n",
1547 parent, dname.len, dname.name, dn);
1550 dn = d_alloc(parent, &dname);
1551 dout("d_alloc %p '%.*s' = %p\n", parent,
1552 dname.len, dname.name, dn);
1554 dout("d_alloc badness\n");
1558 ret = ceph_init_dentry(dn);
1564 } else if (d_really_is_positive(dn) &&
1565 (ceph_ino(d_inode(dn)) != vino.ino ||
1566 ceph_snap(d_inode(dn)) != vino.snap)) {
1567 dout(" dn %p points to wrong inode %p\n",
1575 if (d_really_is_positive(dn)) {
1578 in = ceph_get_inode(parent->d_sb, vino);
1580 dout("new_inode badness\n");
1588 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1589 req->r_request_started, -1,
1590 &req->r_caps_reservation);
1592 pr_err("fill_inode badness on %p\n", in);
1593 if (d_really_is_negative(dn))
1600 if (d_really_is_negative(dn)) {
1601 struct dentry *realdn;
1603 if (ceph_security_xattr_deadlock(in)) {
1604 dout(" skip splicing dn %p to inode %p"
1605 " (security xattr deadlock)\n", dn, in);
1611 realdn = splice_dentry(dn, in);
1612 if (IS_ERR(realdn)) {
1613 err = PTR_ERR(realdn);
1621 ceph_dentry(dn)->offset = rde->offset;
1623 update_dentry_lease(dn, rde->lease, req->r_session,
1624 req->r_request_started);
1626 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1627 ret = fill_readdir_cache(d_inode(parent), dn,
1637 if (err == 0 && skipped == 0) {
1638 req->r_did_prepopulate = true;
1639 req->r_readdir_cache_idx = cache_ctl.index;
1641 ceph_readdir_cache_release(&cache_ctl);
1646 dout("readdir_prepopulate done\n");
1650 int ceph_inode_set_size(struct inode *inode, loff_t size)
1652 struct ceph_inode_info *ci = ceph_inode(inode);
1655 spin_lock(&ci->i_ceph_lock);
1656 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1657 i_size_write(inode, size);
1658 inode->i_blocks = calc_inode_blocks(size);
1660 /* tell the MDS if we are approaching max_size */
1661 if ((size << 1) >= ci->i_max_size &&
1662 (ci->i_reported_size << 1) < ci->i_max_size)
1665 spin_unlock(&ci->i_ceph_lock);
1670 * Write back inode data in a worker thread. (This can't be done
1671 * in the message handler context.)
1673 void ceph_queue_writeback(struct inode *inode)
1676 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1677 &ceph_inode(inode)->i_wb_work)) {
1678 dout("ceph_queue_writeback %p\n", inode);
1680 dout("ceph_queue_writeback %p failed\n", inode);
1685 static void ceph_writeback_work(struct work_struct *work)
1687 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1689 struct inode *inode = &ci->vfs_inode;
1691 dout("writeback %p\n", inode);
1692 filemap_fdatawrite(&inode->i_data);
1697 * queue an async invalidation
1699 void ceph_queue_invalidate(struct inode *inode)
1702 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1703 &ceph_inode(inode)->i_pg_inv_work)) {
1704 dout("ceph_queue_invalidate %p\n", inode);
1706 dout("ceph_queue_invalidate %p failed\n", inode);
1712 * Invalidate inode pages in a worker thread. (This can't be done
1713 * in the message handler context.)
1715 static void ceph_invalidate_work(struct work_struct *work)
1717 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1719 struct inode *inode = &ci->vfs_inode;
1720 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1724 mutex_lock(&ci->i_truncate_mutex);
1726 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1727 pr_warn_ratelimited("invalidate_pages %p %lld forced umount\n",
1728 inode, ceph_ino(inode));
1729 mapping_set_error(inode->i_mapping, -EIO);
1730 truncate_pagecache(inode, 0);
1731 mutex_unlock(&ci->i_truncate_mutex);
1735 spin_lock(&ci->i_ceph_lock);
1736 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1737 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1738 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1739 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1741 spin_unlock(&ci->i_ceph_lock);
1742 mutex_unlock(&ci->i_truncate_mutex);
1745 orig_gen = ci->i_rdcache_gen;
1746 spin_unlock(&ci->i_ceph_lock);
1748 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1749 pr_err("invalidate_pages %p fails\n", inode);
1752 spin_lock(&ci->i_ceph_lock);
1753 if (orig_gen == ci->i_rdcache_gen &&
1754 orig_gen == ci->i_rdcache_revoking) {
1755 dout("invalidate_pages %p gen %d successful\n", inode,
1757 ci->i_rdcache_revoking--;
1760 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1761 inode, orig_gen, ci->i_rdcache_gen,
1762 ci->i_rdcache_revoking);
1763 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1766 spin_unlock(&ci->i_ceph_lock);
1767 mutex_unlock(&ci->i_truncate_mutex);
1770 ceph_check_caps(ci, 0, NULL);
1776 * called by trunc_wq;
1778 * We also truncate in a separate thread as well.
1780 static void ceph_vmtruncate_work(struct work_struct *work)
1782 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1784 struct inode *inode = &ci->vfs_inode;
1786 dout("vmtruncate_work %p\n", inode);
1787 __ceph_do_pending_vmtruncate(inode);
1792 * Queue an async vmtruncate. If we fail to queue work, we will handle
1793 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1795 void ceph_queue_vmtruncate(struct inode *inode)
1797 struct ceph_inode_info *ci = ceph_inode(inode);
1801 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1802 &ci->i_vmtruncate_work)) {
1803 dout("ceph_queue_vmtruncate %p\n", inode);
1805 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1806 inode, ci->i_truncate_pending);
1812 * Make sure any pending truncation is applied before doing anything
1813 * that may depend on it.
1815 void __ceph_do_pending_vmtruncate(struct inode *inode)
1817 struct ceph_inode_info *ci = ceph_inode(inode);
1819 int wrbuffer_refs, finish = 0;
1821 mutex_lock(&ci->i_truncate_mutex);
1823 spin_lock(&ci->i_ceph_lock);
1824 if (ci->i_truncate_pending == 0) {
1825 dout("__do_pending_vmtruncate %p none pending\n", inode);
1826 spin_unlock(&ci->i_ceph_lock);
1827 mutex_unlock(&ci->i_truncate_mutex);
1832 * make sure any dirty snapped pages are flushed before we
1833 * possibly truncate them.. so write AND block!
1835 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1836 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1838 spin_unlock(&ci->i_ceph_lock);
1839 filemap_write_and_wait_range(&inode->i_data, 0,
1840 inode->i_sb->s_maxbytes);
1844 /* there should be no reader or writer */
1845 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1847 to = ci->i_truncate_size;
1848 wrbuffer_refs = ci->i_wrbuffer_ref;
1849 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1850 ci->i_truncate_pending, to);
1851 spin_unlock(&ci->i_ceph_lock);
1853 truncate_pagecache(inode, to);
1855 spin_lock(&ci->i_ceph_lock);
1856 if (to == ci->i_truncate_size) {
1857 ci->i_truncate_pending = 0;
1860 spin_unlock(&ci->i_ceph_lock);
1864 mutex_unlock(&ci->i_truncate_mutex);
1866 if (wrbuffer_refs == 0)
1867 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1869 wake_up_all(&ci->i_cap_wq);
1875 static const struct inode_operations ceph_symlink_iops = {
1876 .readlink = generic_readlink,
1877 .get_link = simple_get_link,
1878 .setattr = ceph_setattr,
1879 .getattr = ceph_getattr,
1880 .setxattr = generic_setxattr,
1881 .getxattr = generic_getxattr,
1882 .listxattr = ceph_listxattr,
1883 .removexattr = generic_removexattr,
1886 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1888 struct ceph_inode_info *ci = ceph_inode(inode);
1889 const unsigned int ia_valid = attr->ia_valid;
1890 struct ceph_mds_request *req;
1891 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1892 struct ceph_cap_flush *prealloc_cf;
1894 int release = 0, dirtied = 0;
1897 int inode_dirty_flags = 0;
1898 bool lock_snap_rwsem = false;
1900 if (ceph_snap(inode) != CEPH_NOSNAP)
1903 err = inode_change_ok(inode, attr);
1907 prealloc_cf = ceph_alloc_cap_flush();
1911 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1914 ceph_free_cap_flush(prealloc_cf);
1915 return PTR_ERR(req);
1918 spin_lock(&ci->i_ceph_lock);
1919 issued = __ceph_caps_issued(ci, NULL);
1921 if (!ci->i_head_snapc &&
1922 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1923 lock_snap_rwsem = true;
1924 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1925 spin_unlock(&ci->i_ceph_lock);
1926 down_read(&mdsc->snap_rwsem);
1927 spin_lock(&ci->i_ceph_lock);
1928 issued = __ceph_caps_issued(ci, NULL);
1932 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1934 if (ia_valid & ATTR_UID) {
1935 dout("setattr %p uid %d -> %d\n", inode,
1936 from_kuid(&init_user_ns, inode->i_uid),
1937 from_kuid(&init_user_ns, attr->ia_uid));
1938 if (issued & CEPH_CAP_AUTH_EXCL) {
1939 inode->i_uid = attr->ia_uid;
1940 dirtied |= CEPH_CAP_AUTH_EXCL;
1941 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1942 !uid_eq(attr->ia_uid, inode->i_uid)) {
1943 req->r_args.setattr.uid = cpu_to_le32(
1944 from_kuid(&init_user_ns, attr->ia_uid));
1945 mask |= CEPH_SETATTR_UID;
1946 release |= CEPH_CAP_AUTH_SHARED;
1949 if (ia_valid & ATTR_GID) {
1950 dout("setattr %p gid %d -> %d\n", inode,
1951 from_kgid(&init_user_ns, inode->i_gid),
1952 from_kgid(&init_user_ns, attr->ia_gid));
1953 if (issued & CEPH_CAP_AUTH_EXCL) {
1954 inode->i_gid = attr->ia_gid;
1955 dirtied |= CEPH_CAP_AUTH_EXCL;
1956 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1957 !gid_eq(attr->ia_gid, inode->i_gid)) {
1958 req->r_args.setattr.gid = cpu_to_le32(
1959 from_kgid(&init_user_ns, attr->ia_gid));
1960 mask |= CEPH_SETATTR_GID;
1961 release |= CEPH_CAP_AUTH_SHARED;
1964 if (ia_valid & ATTR_MODE) {
1965 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1967 if (issued & CEPH_CAP_AUTH_EXCL) {
1968 inode->i_mode = attr->ia_mode;
1969 dirtied |= CEPH_CAP_AUTH_EXCL;
1970 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1971 attr->ia_mode != inode->i_mode) {
1972 inode->i_mode = attr->ia_mode;
1973 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1974 mask |= CEPH_SETATTR_MODE;
1975 release |= CEPH_CAP_AUTH_SHARED;
1979 if (ia_valid & ATTR_ATIME) {
1980 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1981 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1982 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1983 if (issued & CEPH_CAP_FILE_EXCL) {
1984 ci->i_time_warp_seq++;
1985 inode->i_atime = attr->ia_atime;
1986 dirtied |= CEPH_CAP_FILE_EXCL;
1987 } else if ((issued & CEPH_CAP_FILE_WR) &&
1988 timespec_compare(&inode->i_atime,
1989 &attr->ia_atime) < 0) {
1990 inode->i_atime = attr->ia_atime;
1991 dirtied |= CEPH_CAP_FILE_WR;
1992 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1993 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1994 ceph_encode_timespec(&req->r_args.setattr.atime,
1996 mask |= CEPH_SETATTR_ATIME;
1997 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
2001 if (ia_valid & ATTR_MTIME) {
2002 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
2003 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2004 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2005 if (issued & CEPH_CAP_FILE_EXCL) {
2006 ci->i_time_warp_seq++;
2007 inode->i_mtime = attr->ia_mtime;
2008 dirtied |= CEPH_CAP_FILE_EXCL;
2009 } else if ((issued & CEPH_CAP_FILE_WR) &&
2010 timespec_compare(&inode->i_mtime,
2011 &attr->ia_mtime) < 0) {
2012 inode->i_mtime = attr->ia_mtime;
2013 dirtied |= CEPH_CAP_FILE_WR;
2014 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2015 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2016 ceph_encode_timespec(&req->r_args.setattr.mtime,
2018 mask |= CEPH_SETATTR_MTIME;
2019 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2023 if (ia_valid & ATTR_SIZE) {
2024 dout("setattr %p size %lld -> %lld\n", inode,
2025 inode->i_size, attr->ia_size);
2026 if ((issued & CEPH_CAP_FILE_EXCL) &&
2027 attr->ia_size > inode->i_size) {
2028 i_size_write(inode, attr->ia_size);
2029 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2030 inode->i_ctime = attr->ia_ctime;
2031 ci->i_reported_size = attr->ia_size;
2032 dirtied |= CEPH_CAP_FILE_EXCL;
2033 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2034 attr->ia_size != inode->i_size) {
2035 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2036 req->r_args.setattr.old_size =
2037 cpu_to_le64(inode->i_size);
2038 mask |= CEPH_SETATTR_SIZE;
2039 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2044 /* these do nothing */
2045 if (ia_valid & ATTR_CTIME) {
2046 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2047 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2048 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2049 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2050 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2051 only ? "ctime only" : "ignored");
2052 inode->i_ctime = attr->ia_ctime;
2055 * if kernel wants to dirty ctime but nothing else,
2056 * we need to choose a cap to dirty under, or do
2057 * a almost-no-op setattr
2059 if (issued & CEPH_CAP_AUTH_EXCL)
2060 dirtied |= CEPH_CAP_AUTH_EXCL;
2061 else if (issued & CEPH_CAP_FILE_EXCL)
2062 dirtied |= CEPH_CAP_FILE_EXCL;
2063 else if (issued & CEPH_CAP_XATTR_EXCL)
2064 dirtied |= CEPH_CAP_XATTR_EXCL;
2066 mask |= CEPH_SETATTR_CTIME;
2069 if (ia_valid & ATTR_FILE)
2070 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2073 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2075 inode->i_ctime = current_fs_time(inode->i_sb);
2079 spin_unlock(&ci->i_ceph_lock);
2080 if (lock_snap_rwsem)
2081 up_read(&mdsc->snap_rwsem);
2083 if (inode_dirty_flags)
2084 __mark_inode_dirty(inode, inode_dirty_flags);
2086 if (ia_valid & ATTR_MODE) {
2087 err = posix_acl_chmod(inode, attr->ia_mode);
2093 req->r_inode = inode;
2095 req->r_inode_drop = release;
2096 req->r_args.setattr.mask = cpu_to_le32(mask);
2097 req->r_num_caps = 1;
2098 err = ceph_mdsc_do_request(mdsc, NULL, req);
2100 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2101 ceph_cap_string(dirtied), mask);
2103 ceph_mdsc_put_request(req);
2104 if (mask & CEPH_SETATTR_SIZE)
2105 __ceph_do_pending_vmtruncate(inode);
2106 ceph_free_cap_flush(prealloc_cf);
2109 ceph_mdsc_put_request(req);
2110 ceph_free_cap_flush(prealloc_cf);
2117 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2119 return __ceph_setattr(d_inode(dentry), attr);
2123 * Verify that we have a lease on the given mask. If not,
2124 * do a getattr against an mds.
2126 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2127 int mask, bool force)
2129 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2130 struct ceph_mds_client *mdsc = fsc->mdsc;
2131 struct ceph_mds_request *req;
2134 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2135 dout("do_getattr inode %p SNAPDIR\n", inode);
2139 dout("do_getattr inode %p mask %s mode 0%o\n",
2140 inode, ceph_cap_string(mask), inode->i_mode);
2141 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2144 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2146 return PTR_ERR(req);
2147 req->r_inode = inode;
2149 req->r_num_caps = 1;
2150 req->r_args.getattr.mask = cpu_to_le32(mask);
2151 req->r_locked_page = locked_page;
2152 err = ceph_mdsc_do_request(mdsc, NULL, req);
2153 if (locked_page && err == 0) {
2154 u64 inline_version = req->r_reply_info.targeti.inline_version;
2155 if (inline_version == 0) {
2156 /* the reply is supposed to contain inline data */
2158 } else if (inline_version == CEPH_INLINE_NONE) {
2161 err = req->r_reply_info.targeti.inline_len;
2164 ceph_mdsc_put_request(req);
2165 dout("do_getattr result=%d\n", err);
2171 * Check inode permissions. We verify we have a valid value for
2172 * the AUTH cap, then call the generic handler.
2174 int ceph_permission(struct inode *inode, int mask)
2178 if (mask & MAY_NOT_BLOCK)
2181 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2184 err = generic_permission(inode, mask);
2189 * Get all attributes. Hopefully somedata we'll have a statlite()
2190 * and can limit the fields we require to be accurate.
2192 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
2195 struct inode *inode = d_inode(dentry);
2196 struct ceph_inode_info *ci = ceph_inode(inode);
2199 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2201 generic_fillattr(inode, stat);
2202 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2203 if (ceph_snap(inode) != CEPH_NOSNAP)
2204 stat->dev = ceph_snap(inode);
2207 if (S_ISDIR(inode->i_mode)) {
2208 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2210 stat->size = ci->i_rbytes;
2212 stat->size = ci->i_files + ci->i_subdirs;
2214 stat->blksize = 65536;