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 .listxattr = ceph_listxattr,
98 .get_acl = ceph_get_acl,
99 .set_acl = ceph_set_acl,
104 * We use a 'frag tree' to keep track of the MDS's directory fragments
105 * for a given inode (usually there is just a single fragment). We
106 * need to know when a child frag is delegated to a new MDS, or when
107 * it is flagged as replicated, so we can direct our requests
112 * find/create a frag in the tree
114 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
118 struct rb_node *parent = NULL;
119 struct ceph_inode_frag *frag;
122 p = &ci->i_fragtree.rb_node;
125 frag = rb_entry(parent, struct ceph_inode_frag, node);
126 c = ceph_frag_compare(f, frag->frag);
135 frag = kmalloc(sizeof(*frag), GFP_NOFS);
137 return ERR_PTR(-ENOMEM);
144 rb_link_node(&frag->node, parent, p);
145 rb_insert_color(&frag->node, &ci->i_fragtree);
147 dout("get_or_create_frag added %llx.%llx frag %x\n",
148 ceph_vinop(&ci->vfs_inode), f);
153 * find a specific frag @f
155 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
157 struct rb_node *n = ci->i_fragtree.rb_node;
160 struct ceph_inode_frag *frag =
161 rb_entry(n, struct ceph_inode_frag, node);
162 int c = ceph_frag_compare(f, frag->frag);
174 * Choose frag containing the given value @v. If @pfrag is
175 * specified, copy the frag delegation info to the caller if
178 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
179 struct ceph_inode_frag *pfrag, int *found)
181 u32 t = ceph_frag_make(0, 0);
182 struct ceph_inode_frag *frag;
190 WARN_ON(!ceph_frag_contains_value(t, v));
191 frag = __ceph_find_frag(ci, t);
193 break; /* t is a leaf */
194 if (frag->split_by == 0) {
196 memcpy(pfrag, frag, sizeof(*pfrag));
203 nway = 1 << frag->split_by;
204 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
205 frag->split_by, nway);
206 for (i = 0; i < nway; i++) {
207 n = ceph_frag_make_child(t, frag->split_by, i);
208 if (ceph_frag_contains_value(n, v)) {
215 dout("choose_frag(%x) = %x\n", v, t);
220 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
221 struct ceph_inode_frag *pfrag, int *found)
224 mutex_lock(&ci->i_fragtree_mutex);
225 ret = __ceph_choose_frag(ci, v, pfrag, found);
226 mutex_unlock(&ci->i_fragtree_mutex);
231 * Process dirfrag (delegation) info from the mds. Include leaf
232 * fragment in tree ONLY if ndist > 0. Otherwise, only
233 * branches/splits are included in i_fragtree)
235 static int ceph_fill_dirfrag(struct inode *inode,
236 struct ceph_mds_reply_dirfrag *dirinfo)
238 struct ceph_inode_info *ci = ceph_inode(inode);
239 struct ceph_inode_frag *frag;
240 u32 id = le32_to_cpu(dirinfo->frag);
241 int mds = le32_to_cpu(dirinfo->auth);
242 int ndist = le32_to_cpu(dirinfo->ndist);
247 spin_lock(&ci->i_ceph_lock);
249 diri_auth = ci->i_auth_cap->mds;
250 spin_unlock(&ci->i_ceph_lock);
252 if (mds == -1) /* CDIR_AUTH_PARENT */
255 mutex_lock(&ci->i_fragtree_mutex);
256 if (ndist == 0 && mds == diri_auth) {
257 /* no delegation info needed. */
258 frag = __ceph_find_frag(ci, id);
261 if (frag->split_by == 0) {
262 /* tree leaf, remove */
263 dout("fill_dirfrag removed %llx.%llx frag %x"
264 " (no ref)\n", ceph_vinop(inode), id);
265 rb_erase(&frag->node, &ci->i_fragtree);
268 /* tree branch, keep and clear */
269 dout("fill_dirfrag cleared %llx.%llx frag %x"
270 " referral\n", ceph_vinop(inode), id);
278 /* find/add this frag to store mds delegation info */
279 frag = __get_or_create_frag(ci, id);
281 /* this is not the end of the world; we can continue
282 with bad/inaccurate delegation info */
283 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
284 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
290 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
291 for (i = 0; i < frag->ndist; i++)
292 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
293 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
294 ceph_vinop(inode), frag->frag, frag->ndist);
297 mutex_unlock(&ci->i_fragtree_mutex);
301 static int frag_tree_split_cmp(const void *l, const void *r)
303 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
304 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
305 return ceph_frag_compare(le32_to_cpu(ls->frag),
306 le32_to_cpu(rs->frag));
309 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
312 return f == ceph_frag_make(0, 0);
313 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
315 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
318 static int ceph_fill_fragtree(struct inode *inode,
319 struct ceph_frag_tree_head *fragtree,
320 struct ceph_mds_reply_dirfrag *dirinfo)
322 struct ceph_inode_info *ci = ceph_inode(inode);
323 struct ceph_inode_frag *frag, *prev_frag = NULL;
324 struct rb_node *rb_node;
325 unsigned i, split_by, nsplits;
329 mutex_lock(&ci->i_fragtree_mutex);
330 nsplits = le32_to_cpu(fragtree->nsplits);
331 if (nsplits != ci->i_fragtree_nsplits) {
333 } else if (nsplits) {
334 i = prandom_u32() % nsplits;
335 id = le32_to_cpu(fragtree->splits[i].frag);
336 if (!__ceph_find_frag(ci, id))
338 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
339 rb_node = rb_first(&ci->i_fragtree);
340 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
341 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
344 if (!update && dirinfo) {
345 id = le32_to_cpu(dirinfo->frag);
346 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
353 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
354 frag_tree_split_cmp, NULL);
357 dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
358 rb_node = rb_first(&ci->i_fragtree);
359 for (i = 0; i < nsplits; i++) {
360 id = le32_to_cpu(fragtree->splits[i].frag);
361 split_by = le32_to_cpu(fragtree->splits[i].by);
362 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
363 pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
364 "frag %x split by %d\n", ceph_vinop(inode),
365 i, nsplits, id, split_by);
370 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
371 if (ceph_frag_compare(frag->frag, id) >= 0) {
372 if (frag->frag != id)
375 rb_node = rb_next(rb_node);
378 rb_node = rb_next(rb_node);
379 /* delete stale split/leaf node */
380 if (frag->split_by > 0 ||
381 !is_frag_child(frag->frag, prev_frag)) {
382 rb_erase(&frag->node, &ci->i_fragtree);
383 if (frag->split_by > 0)
384 ci->i_fragtree_nsplits--;
390 frag = __get_or_create_frag(ci, id);
394 if (frag->split_by == 0)
395 ci->i_fragtree_nsplits++;
396 frag->split_by = split_by;
397 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
401 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
402 rb_node = rb_next(rb_node);
403 /* delete stale split/leaf node */
404 if (frag->split_by > 0 ||
405 !is_frag_child(frag->frag, prev_frag)) {
406 rb_erase(&frag->node, &ci->i_fragtree);
407 if (frag->split_by > 0)
408 ci->i_fragtree_nsplits--;
413 mutex_unlock(&ci->i_fragtree_mutex);
418 * initialize a newly allocated inode.
420 struct inode *ceph_alloc_inode(struct super_block *sb)
422 struct ceph_inode_info *ci;
425 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
429 dout("alloc_inode %p\n", &ci->vfs_inode);
431 spin_lock_init(&ci->i_ceph_lock);
434 ci->i_inline_version = 0;
435 ci->i_time_warp_seq = 0;
436 ci->i_ceph_flags = 0;
437 atomic64_set(&ci->i_ordered_count, 1);
438 atomic64_set(&ci->i_release_count, 1);
439 atomic64_set(&ci->i_complete_seq[0], 0);
440 atomic64_set(&ci->i_complete_seq[1], 0);
441 ci->i_symlink = NULL;
443 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
444 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
446 ci->i_fragtree = RB_ROOT;
447 mutex_init(&ci->i_fragtree_mutex);
449 ci->i_xattrs.blob = NULL;
450 ci->i_xattrs.prealloc_blob = NULL;
451 ci->i_xattrs.dirty = false;
452 ci->i_xattrs.index = RB_ROOT;
453 ci->i_xattrs.count = 0;
454 ci->i_xattrs.names_size = 0;
455 ci->i_xattrs.vals_size = 0;
456 ci->i_xattrs.version = 0;
457 ci->i_xattrs.index_version = 0;
459 ci->i_caps = RB_ROOT;
460 ci->i_auth_cap = NULL;
461 ci->i_dirty_caps = 0;
462 ci->i_flushing_caps = 0;
463 INIT_LIST_HEAD(&ci->i_dirty_item);
464 INIT_LIST_HEAD(&ci->i_flushing_item);
465 ci->i_prealloc_cap_flush = NULL;
466 INIT_LIST_HEAD(&ci->i_cap_flush_list);
467 init_waitqueue_head(&ci->i_cap_wq);
468 ci->i_hold_caps_min = 0;
469 ci->i_hold_caps_max = 0;
470 INIT_LIST_HEAD(&ci->i_cap_delay_list);
471 INIT_LIST_HEAD(&ci->i_cap_snaps);
472 ci->i_head_snapc = NULL;
475 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
476 ci->i_nr_by_mode[i] = 0;
478 mutex_init(&ci->i_truncate_mutex);
479 ci->i_truncate_seq = 0;
480 ci->i_truncate_size = 0;
481 ci->i_truncate_pending = 0;
484 ci->i_reported_size = 0;
485 ci->i_wanted_max_size = 0;
486 ci->i_requested_max_size = 0;
490 ci->i_rdcache_ref = 0;
493 ci->i_wrbuffer_ref = 0;
494 ci->i_wrbuffer_ref_head = 0;
495 ci->i_shared_gen = 0;
496 ci->i_rdcache_gen = 0;
497 ci->i_rdcache_revoking = 0;
499 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
500 INIT_LIST_HEAD(&ci->i_unsafe_iops);
501 spin_lock_init(&ci->i_unsafe_lock);
503 ci->i_snap_realm = NULL;
504 INIT_LIST_HEAD(&ci->i_snap_realm_item);
505 INIT_LIST_HEAD(&ci->i_snap_flush_item);
507 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
508 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
510 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
512 ceph_fscache_inode_init(ci);
514 return &ci->vfs_inode;
517 static void ceph_i_callback(struct rcu_head *head)
519 struct inode *inode = container_of(head, struct inode, i_rcu);
520 struct ceph_inode_info *ci = ceph_inode(inode);
522 kmem_cache_free(ceph_inode_cachep, ci);
525 void ceph_destroy_inode(struct inode *inode)
527 struct ceph_inode_info *ci = ceph_inode(inode);
528 struct ceph_inode_frag *frag;
531 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
533 ceph_fscache_unregister_inode_cookie(ci);
535 ceph_queue_caps_release(inode);
538 * we may still have a snap_realm reference if there are stray
539 * caps in i_snap_caps.
541 if (ci->i_snap_realm) {
542 struct ceph_mds_client *mdsc =
543 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
544 struct ceph_snap_realm *realm = ci->i_snap_realm;
546 dout(" dropping residual ref to snap realm %p\n", realm);
547 spin_lock(&realm->inodes_with_caps_lock);
548 list_del_init(&ci->i_snap_realm_item);
549 spin_unlock(&realm->inodes_with_caps_lock);
550 ceph_put_snap_realm(mdsc, realm);
553 kfree(ci->i_symlink);
554 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
555 frag = rb_entry(n, struct ceph_inode_frag, node);
556 rb_erase(n, &ci->i_fragtree);
559 ci->i_fragtree_nsplits = 0;
561 __ceph_destroy_xattrs(ci);
562 if (ci->i_xattrs.blob)
563 ceph_buffer_put(ci->i_xattrs.blob);
564 if (ci->i_xattrs.prealloc_blob)
565 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
567 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
569 call_rcu(&inode->i_rcu, ceph_i_callback);
572 int ceph_drop_inode(struct inode *inode)
575 * Positve dentry and corresponding inode are always accompanied
576 * in MDS reply. So no need to keep inode in the cache after
577 * dropping all its aliases.
582 static inline blkcnt_t calc_inode_blocks(u64 size)
584 return (size + (1<<9) - 1) >> 9;
588 * Helpers to fill in size, ctime, mtime, and atime. We have to be
589 * careful because either the client or MDS may have more up to date
590 * info, depending on which capabilities are held, and whether
591 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
592 * and size are monotonically increasing, except when utimes() or
593 * truncate() increments the corresponding _seq values.)
595 int ceph_fill_file_size(struct inode *inode, int issued,
596 u32 truncate_seq, u64 truncate_size, u64 size)
598 struct ceph_inode_info *ci = ceph_inode(inode);
601 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
602 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
603 dout("size %lld -> %llu\n", inode->i_size, size);
604 if (size > 0 && S_ISDIR(inode->i_mode)) {
605 pr_err("fill_file_size non-zero size for directory\n");
608 i_size_write(inode, size);
609 inode->i_blocks = calc_inode_blocks(size);
610 ci->i_reported_size = size;
611 if (truncate_seq != ci->i_truncate_seq) {
612 dout("truncate_seq %u -> %u\n",
613 ci->i_truncate_seq, truncate_seq);
614 ci->i_truncate_seq = truncate_seq;
616 /* the MDS should have revoked these caps */
617 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
620 CEPH_CAP_FILE_LAZYIO));
622 * If we hold relevant caps, or in the case where we're
623 * not the only client referencing this file and we
624 * don't hold those caps, then we need to check whether
625 * the file is either opened or mmaped
627 if ((issued & (CEPH_CAP_FILE_CACHE|
628 CEPH_CAP_FILE_BUFFER)) ||
629 mapping_mapped(inode->i_mapping) ||
630 __ceph_caps_file_wanted(ci)) {
631 ci->i_truncate_pending++;
636 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
637 ci->i_truncate_size != truncate_size) {
638 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
640 ci->i_truncate_size = truncate_size;
644 ceph_fscache_invalidate(inode);
649 void ceph_fill_file_time(struct inode *inode, int issued,
650 u64 time_warp_seq, struct timespec *ctime,
651 struct timespec *mtime, struct timespec *atime)
653 struct ceph_inode_info *ci = ceph_inode(inode);
656 if (issued & (CEPH_CAP_FILE_EXCL|
658 CEPH_CAP_FILE_BUFFER|
660 CEPH_CAP_XATTR_EXCL)) {
661 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
662 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
663 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
664 ctime->tv_sec, ctime->tv_nsec);
665 inode->i_ctime = *ctime;
667 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
668 /* the MDS did a utimes() */
669 dout("mtime %ld.%09ld -> %ld.%09ld "
671 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
672 mtime->tv_sec, mtime->tv_nsec,
673 ci->i_time_warp_seq, (int)time_warp_seq);
675 inode->i_mtime = *mtime;
676 inode->i_atime = *atime;
677 ci->i_time_warp_seq = time_warp_seq;
678 } else if (time_warp_seq == ci->i_time_warp_seq) {
679 /* nobody did utimes(); take the max */
680 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
681 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
682 inode->i_mtime.tv_sec,
683 inode->i_mtime.tv_nsec,
684 mtime->tv_sec, mtime->tv_nsec);
685 inode->i_mtime = *mtime;
687 if (timespec_compare(atime, &inode->i_atime) > 0) {
688 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
689 inode->i_atime.tv_sec,
690 inode->i_atime.tv_nsec,
691 atime->tv_sec, atime->tv_nsec);
692 inode->i_atime = *atime;
694 } else if (issued & CEPH_CAP_FILE_EXCL) {
695 /* we did a utimes(); ignore mds values */
700 /* we have no write|excl caps; whatever the MDS says is true */
701 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
702 inode->i_ctime = *ctime;
703 inode->i_mtime = *mtime;
704 inode->i_atime = *atime;
705 ci->i_time_warp_seq = time_warp_seq;
710 if (warn) /* time_warp_seq shouldn't go backwards */
711 dout("%p mds time_warp_seq %llu < %u\n",
712 inode, time_warp_seq, ci->i_time_warp_seq);
716 * Populate an inode based on info from mds. May be called on new or
719 static int fill_inode(struct inode *inode, struct page *locked_page,
720 struct ceph_mds_reply_info_in *iinfo,
721 struct ceph_mds_reply_dirfrag *dirinfo,
722 struct ceph_mds_session *session,
723 unsigned long ttl_from, int cap_fmode,
724 struct ceph_cap_reservation *caps_reservation)
726 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
727 struct ceph_mds_reply_inode *info = iinfo->in;
728 struct ceph_inode_info *ci = ceph_inode(inode);
729 int issued = 0, implemented, new_issued;
730 struct timespec mtime, atime, ctime;
731 struct ceph_buffer *xattr_blob = NULL;
732 struct ceph_string *pool_ns = NULL;
733 struct ceph_cap *new_cap = NULL;
736 bool queue_trunc = false;
737 bool new_version = false;
738 bool fill_inline = false;
740 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
741 inode, ceph_vinop(inode), le64_to_cpu(info->version),
744 /* prealloc new cap struct */
745 if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
746 new_cap = ceph_get_cap(mdsc, caps_reservation);
749 * prealloc xattr data, if it looks like we'll need it. only
750 * if len > 4 (meaning there are actually xattrs; the first 4
751 * bytes are the xattr count).
753 if (iinfo->xattr_len > 4) {
754 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
756 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
760 if (iinfo->pool_ns_len > 0)
761 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
764 spin_lock(&ci->i_ceph_lock);
767 * provided version will be odd if inode value is projected,
768 * even if stable. skip the update if we have newer stable
769 * info (ours>=theirs, e.g. due to racing mds replies), unless
770 * we are getting projected (unstable) info (in which case the
771 * version is odd, and we want ours>theirs).
777 if (ci->i_version == 0 ||
778 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
779 le64_to_cpu(info->version) > (ci->i_version & ~1)))
782 issued = __ceph_caps_issued(ci, &implemented);
783 issued |= implemented | __ceph_caps_dirty(ci);
784 new_issued = ~issued & le32_to_cpu(info->cap.caps);
787 ci->i_version = le64_to_cpu(info->version);
789 inode->i_rdev = le32_to_cpu(info->rdev);
790 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
792 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
793 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
794 inode->i_mode = le32_to_cpu(info->mode);
795 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
796 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
797 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
798 from_kuid(&init_user_ns, inode->i_uid),
799 from_kgid(&init_user_ns, inode->i_gid));
802 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
803 (issued & CEPH_CAP_LINK_EXCL) == 0)
804 set_nlink(inode, le32_to_cpu(info->nlink));
806 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
807 /* be careful with mtime, atime, size */
808 ceph_decode_timespec(&atime, &info->atime);
809 ceph_decode_timespec(&mtime, &info->mtime);
810 ceph_decode_timespec(&ctime, &info->ctime);
811 ceph_fill_file_time(inode, issued,
812 le32_to_cpu(info->time_warp_seq),
813 &ctime, &mtime, &atime);
817 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
818 s64 old_pool = ci->i_layout.pool_id;
819 struct ceph_string *old_ns;
821 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
822 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
823 lockdep_is_held(&ci->i_ceph_lock));
824 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
826 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
827 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
831 queue_trunc = ceph_fill_file_size(inode, issued,
832 le32_to_cpu(info->truncate_seq),
833 le64_to_cpu(info->truncate_size),
834 le64_to_cpu(info->size));
835 /* only update max_size on auth cap */
836 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
837 ci->i_max_size != le64_to_cpu(info->max_size)) {
838 dout("max_size %lld -> %llu\n", ci->i_max_size,
839 le64_to_cpu(info->max_size));
840 ci->i_max_size = le64_to_cpu(info->max_size);
845 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
846 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
847 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
848 if (ci->i_xattrs.blob)
849 ceph_buffer_put(ci->i_xattrs.blob);
850 ci->i_xattrs.blob = xattr_blob;
852 memcpy(ci->i_xattrs.blob->vec.iov_base,
853 iinfo->xattr_data, iinfo->xattr_len);
854 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
855 ceph_forget_all_cached_acls(inode);
859 inode->i_mapping->a_ops = &ceph_aops;
861 switch (inode->i_mode & S_IFMT) {
866 init_special_inode(inode, inode->i_mode, inode->i_rdev);
867 inode->i_op = &ceph_file_iops;
870 inode->i_op = &ceph_file_iops;
871 inode->i_fop = &ceph_file_fops;
874 inode->i_op = &ceph_symlink_iops;
875 if (!ci->i_symlink) {
876 u32 symlen = iinfo->symlink_len;
879 spin_unlock(&ci->i_ceph_lock);
881 if (symlen != i_size_read(inode)) {
882 pr_err("fill_inode %llx.%llx BAD symlink "
883 "size %lld\n", ceph_vinop(inode),
885 i_size_write(inode, symlen);
886 inode->i_blocks = calc_inode_blocks(symlen);
890 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
894 spin_lock(&ci->i_ceph_lock);
898 kfree(sym); /* lost a race */
900 inode->i_link = ci->i_symlink;
903 inode->i_op = &ceph_dir_iops;
904 inode->i_fop = &ceph_dir_fops;
906 ci->i_dir_layout = iinfo->dir_layout;
908 ci->i_files = le64_to_cpu(info->files);
909 ci->i_subdirs = le64_to_cpu(info->subdirs);
910 ci->i_rbytes = le64_to_cpu(info->rbytes);
911 ci->i_rfiles = le64_to_cpu(info->rfiles);
912 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
913 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
916 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
917 ceph_vinop(inode), inode->i_mode);
920 /* were we issued a capability? */
921 if (info->cap.caps) {
922 if (ceph_snap(inode) == CEPH_NOSNAP) {
923 unsigned caps = le32_to_cpu(info->cap.caps);
924 ceph_add_cap(inode, session,
925 le64_to_cpu(info->cap.cap_id),
927 le32_to_cpu(info->cap.wanted),
928 le32_to_cpu(info->cap.seq),
929 le32_to_cpu(info->cap.mseq),
930 le64_to_cpu(info->cap.realm),
931 info->cap.flags, &new_cap);
933 /* set dir completion flag? */
934 if (S_ISDIR(inode->i_mode) &&
935 ci->i_files == 0 && ci->i_subdirs == 0 &&
936 (caps & CEPH_CAP_FILE_SHARED) &&
937 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
938 !__ceph_dir_is_complete(ci)) {
939 dout(" marking %p complete (empty)\n", inode);
940 i_size_write(inode, 0);
941 __ceph_dir_set_complete(ci,
942 atomic64_read(&ci->i_release_count),
943 atomic64_read(&ci->i_ordered_count));
948 dout(" %p got snap_caps %s\n", inode,
949 ceph_cap_string(le32_to_cpu(info->cap.caps)));
950 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
952 __ceph_get_fmode(ci, cap_fmode);
954 } else if (cap_fmode >= 0) {
955 pr_warn("mds issued no caps on %llx.%llx\n",
957 __ceph_get_fmode(ci, cap_fmode);
960 if (iinfo->inline_version > 0 &&
961 iinfo->inline_version >= ci->i_inline_version) {
962 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
963 ci->i_inline_version = iinfo->inline_version;
964 if (ci->i_inline_version != CEPH_INLINE_NONE &&
966 (le32_to_cpu(info->cap.caps) & cache_caps)))
970 spin_unlock(&ci->i_ceph_lock);
973 ceph_fill_inline_data(inode, locked_page,
974 iinfo->inline_data, iinfo->inline_len);
977 wake_up_all(&ci->i_cap_wq);
979 /* queue truncate if we saw i_size decrease */
981 ceph_queue_vmtruncate(inode);
983 /* populate frag tree */
984 if (S_ISDIR(inode->i_mode))
985 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
987 /* update delegation info? */
989 ceph_fill_dirfrag(inode, dirinfo);
994 ceph_put_cap(mdsc, new_cap);
996 ceph_buffer_put(xattr_blob);
997 ceph_put_string(pool_ns);
1002 * caller should hold session s_mutex.
1004 static void update_dentry_lease(struct dentry *dentry,
1005 struct ceph_mds_reply_lease *lease,
1006 struct ceph_mds_session *session,
1007 unsigned long from_time,
1008 struct ceph_vino *tgt_vino,
1009 struct ceph_vino *dir_vino)
1011 struct ceph_dentry_info *di = ceph_dentry(dentry);
1012 long unsigned duration = le32_to_cpu(lease->duration_ms);
1013 long unsigned ttl = from_time + (duration * HZ) / 1000;
1014 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1016 struct ceph_mds_session *old_lease_session = NULL;
1019 * Make sure dentry's inode matches tgt_vino. NULL tgt_vino means that
1020 * we expect a negative dentry.
1022 if (!tgt_vino && d_really_is_positive(dentry))
1025 if (tgt_vino && (d_really_is_negative(dentry) ||
1026 !ceph_ino_compare(d_inode(dentry), tgt_vino)))
1029 spin_lock(&dentry->d_lock);
1030 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1031 dentry, duration, ttl);
1033 dir = d_inode(dentry->d_parent);
1035 /* make sure parent matches dir_vino */
1036 if (!ceph_ino_compare(dir, dir_vino))
1039 /* only track leases on regular dentries */
1040 if (ceph_snap(dir) != CEPH_NOSNAP)
1043 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
1048 if (di->lease_gen == session->s_cap_gen &&
1049 time_before(ttl, di->time))
1050 goto out_unlock; /* we already have a newer lease. */
1052 if (di->lease_session && di->lease_session != session) {
1053 old_lease_session = di->lease_session;
1054 di->lease_session = NULL;
1057 ceph_dentry_lru_touch(dentry);
1059 if (!di->lease_session)
1060 di->lease_session = ceph_get_mds_session(session);
1061 di->lease_gen = session->s_cap_gen;
1062 di->lease_seq = le32_to_cpu(lease->seq);
1063 di->lease_renew_after = half_ttl;
1064 di->lease_renew_from = 0;
1067 spin_unlock(&dentry->d_lock);
1068 if (old_lease_session)
1069 ceph_put_mds_session(old_lease_session);
1073 * splice a dentry to an inode.
1074 * caller must hold directory i_mutex for this to be safe.
1076 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in)
1078 struct dentry *realdn;
1080 BUG_ON(d_inode(dn));
1082 /* dn must be unhashed */
1083 if (!d_unhashed(dn))
1085 realdn = d_splice_alias(in, dn);
1086 if (IS_ERR(realdn)) {
1087 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1088 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1089 dn = realdn; /* note realdn contains the error */
1091 } else if (realdn) {
1092 dout("dn %p (%d) spliced with %p (%d) "
1093 "inode %p ino %llx.%llx\n",
1095 realdn, d_count(realdn),
1096 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1100 BUG_ON(!ceph_dentry(dn));
1101 dout("dn %p attached to %p ino %llx.%llx\n",
1102 dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1109 * Incorporate results into the local cache. This is either just
1110 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1113 * A reply may contain
1114 * a directory inode along with a dentry.
1115 * and/or a target inode
1117 * Called with snap_rwsem (read).
1119 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1121 struct ceph_mds_session *session = req->r_session;
1122 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1123 struct inode *in = NULL;
1124 struct ceph_vino tvino, dvino;
1125 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1128 dout("fill_trace %p is_dentry %d is_target %d\n", req,
1129 rinfo->head->is_dentry, rinfo->head->is_target);
1131 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1132 dout("fill_trace reply is empty!\n");
1133 if (rinfo->head->result == 0 && req->r_parent)
1134 ceph_invalidate_dir_request(req);
1138 if (rinfo->head->is_dentry) {
1139 struct inode *dir = req->r_parent;
1142 err = fill_inode(dir, NULL,
1143 &rinfo->diri, rinfo->dirfrag,
1144 session, req->r_request_started, -1,
1145 &req->r_caps_reservation);
1152 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1154 struct dentry *dn, *parent;
1156 BUG_ON(!rinfo->head->is_target);
1157 BUG_ON(req->r_dentry);
1159 parent = d_find_any_alias(dir);
1162 dname.name = rinfo->dname;
1163 dname.len = rinfo->dname_len;
1164 dname.hash = full_name_hash(parent, dname.name, dname.len);
1165 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1166 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1168 dn = d_lookup(parent, &dname);
1169 dout("d_lookup on parent=%p name=%.*s got %p\n",
1170 parent, dname.len, dname.name, dn);
1173 dn = d_alloc(parent, &dname);
1174 dout("d_alloc %p '%.*s' = %p\n", parent,
1175 dname.len, dname.name, dn);
1182 } else if (d_really_is_positive(dn) &&
1183 (ceph_ino(d_inode(dn)) != tvino.ino ||
1184 ceph_snap(d_inode(dn)) != tvino.snap)) {
1185 dout(" dn %p points to wrong inode %p\n",
1197 if (rinfo->head->is_target) {
1198 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1199 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1201 in = ceph_get_inode(sb, tvino);
1206 req->r_target_inode = in;
1208 err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1209 session, req->r_request_started,
1210 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1211 rinfo->head->result == 0) ? req->r_fmode : -1,
1212 &req->r_caps_reservation);
1214 pr_err("fill_inode badness %p %llx.%llx\n",
1215 in, ceph_vinop(in));
1221 * ignore null lease/binding on snapdir ENOENT, or else we
1222 * will have trouble splicing in the virtual snapdir later
1224 if (rinfo->head->is_dentry &&
1225 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1226 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1227 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1228 fsc->mount_options->snapdir_name,
1229 req->r_dentry->d_name.len))) {
1231 * lookup link rename : null -> possibly existing inode
1232 * mknod symlink mkdir : null -> new inode
1233 * unlink : linked -> null
1235 struct inode *dir = req->r_parent;
1236 struct dentry *dn = req->r_dentry;
1237 bool have_dir_cap, have_lease;
1241 BUG_ON(d_inode(dn->d_parent) != dir);
1243 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1244 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1246 BUG_ON(ceph_ino(dir) != dvino.ino);
1247 BUG_ON(ceph_snap(dir) != dvino.snap);
1249 /* do we have a lease on the whole dir? */
1251 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1252 CEPH_CAP_FILE_SHARED);
1254 /* do we have a dn lease? */
1255 have_lease = have_dir_cap ||
1256 le32_to_cpu(rinfo->dlease->duration_ms);
1258 dout("fill_trace no dentry lease or dir cap\n");
1261 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1262 struct inode *olddir = req->r_old_dentry_dir;
1265 dout(" src %p '%pd' dst %p '%pd'\n",
1269 dout("fill_trace doing d_move %p -> %p\n",
1270 req->r_old_dentry, dn);
1272 /* d_move screws up sibling dentries' offsets */
1273 ceph_dir_clear_ordered(dir);
1274 ceph_dir_clear_ordered(olddir);
1276 d_move(req->r_old_dentry, dn);
1277 dout(" src %p '%pd' dst %p '%pd'\n",
1282 /* ensure target dentry is invalidated, despite
1283 rehashing bug in vfs_rename_dir */
1284 ceph_invalidate_dentry_lease(dn);
1286 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1287 ceph_dentry(req->r_old_dentry)->offset);
1289 dn = req->r_old_dentry; /* use old_dentry */
1293 if (!rinfo->head->is_target) {
1294 dout("fill_trace null dentry\n");
1295 if (d_really_is_positive(dn)) {
1296 ceph_dir_clear_ordered(dir);
1297 dout("d_delete %p\n", dn);
1299 } else if (have_lease) {
1302 update_dentry_lease(dn, rinfo->dlease,
1304 req->r_request_started,
1310 /* attach proper inode */
1311 if (d_really_is_negative(dn)) {
1312 ceph_dir_clear_ordered(dir);
1314 dn = splice_dentry(dn, in);
1319 req->r_dentry = dn; /* may have spliced */
1320 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1321 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1322 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1329 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1330 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1331 update_dentry_lease(dn, rinfo->dlease, session,
1332 req->r_request_started,
1335 dout(" final dn %p\n", dn);
1336 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1337 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1338 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1339 struct dentry *dn = req->r_dentry;
1340 struct inode *dir = req->r_parent;
1342 /* fill out a snapdir LOOKUPSNAP dentry */
1345 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1346 dout(" linking snapped dir %p to dn %p\n", in, dn);
1347 ceph_dir_clear_ordered(dir);
1349 dn = splice_dentry(dn, in);
1354 req->r_dentry = dn; /* may have spliced */
1355 } else if (rinfo->head->is_dentry) {
1356 struct ceph_vino *ptvino = NULL;
1358 if ((le32_to_cpu(rinfo->diri.in->cap.caps) & CEPH_CAP_FILE_SHARED) ||
1359 le32_to_cpu(rinfo->dlease->duration_ms)) {
1360 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1361 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1363 if (rinfo->head->is_target) {
1364 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1365 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1369 update_dentry_lease(req->r_dentry, rinfo->dlease,
1370 session, req->r_request_started, ptvino,
1373 dout("%s: no dentry lease or dir cap\n", __func__);
1377 dout("fill_trace done err=%d\n", err);
1382 * Prepopulate our cache with readdir results, leases, etc.
1384 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1385 struct ceph_mds_session *session)
1387 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1390 for (i = 0; i < rinfo->dir_nr; i++) {
1391 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1392 struct ceph_vino vino;
1396 vino.ino = le64_to_cpu(rde->inode.in->ino);
1397 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1399 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1402 dout("new_inode badness got %d\n", err);
1405 rc = fill_inode(in, NULL, &rde->inode, NULL, session,
1406 req->r_request_started, -1,
1407 &req->r_caps_reservation);
1409 pr_err("fill_inode badness on %p got %d\n", in, rc);
1418 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1422 put_page(ctl->page);
1427 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1428 struct ceph_readdir_cache_control *ctl,
1429 struct ceph_mds_request *req)
1431 struct ceph_inode_info *ci = ceph_inode(dir);
1432 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1433 unsigned idx = ctl->index % nsize;
1434 pgoff_t pgoff = ctl->index / nsize;
1436 if (!ctl->page || pgoff != page_index(ctl->page)) {
1437 ceph_readdir_cache_release(ctl);
1439 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1441 ctl->page = find_lock_page(&dir->i_data, pgoff);
1444 return idx == 0 ? -ENOMEM : 0;
1446 /* reading/filling the cache are serialized by
1447 * i_mutex, no need to use page lock */
1448 unlock_page(ctl->page);
1449 ctl->dentries = kmap(ctl->page);
1451 memset(ctl->dentries, 0, PAGE_SIZE);
1454 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1455 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1456 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1457 ctl->dentries[idx] = dn;
1460 dout("disable readdir cache\n");
1466 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1467 struct ceph_mds_session *session)
1469 struct dentry *parent = req->r_dentry;
1470 struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1471 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1475 int err = 0, skipped = 0, ret, i;
1476 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1477 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1480 struct ceph_readdir_cache_control cache_ctl = {};
1482 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1483 return readdir_prepopulate_inodes_only(req, session);
1485 if (rinfo->hash_order) {
1487 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1489 strlen(req->r_path2));
1490 last_hash = ceph_frag_value(last_hash);
1491 } else if (rinfo->offset_hash) {
1492 /* mds understands offset_hash */
1493 WARN_ON_ONCE(req->r_readdir_offset != 2);
1494 last_hash = le32_to_cpu(rhead->args.readdir.offset_hash);
1498 if (rinfo->dir_dir &&
1499 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1500 dout("readdir_prepopulate got new frag %x -> %x\n",
1501 frag, le32_to_cpu(rinfo->dir_dir->frag));
1502 frag = le32_to_cpu(rinfo->dir_dir->frag);
1503 if (!rinfo->hash_order)
1504 req->r_readdir_offset = 2;
1507 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1508 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1509 rinfo->dir_nr, parent);
1511 dout("readdir_prepopulate %d items under dn %p\n",
1512 rinfo->dir_nr, parent);
1514 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1516 if (ceph_frag_is_leftmost(frag) &&
1517 req->r_readdir_offset == 2 &&
1518 !(rinfo->hash_order && last_hash)) {
1519 /* note dir version at start of readdir so we can
1520 * tell if any dentries get dropped */
1521 req->r_dir_release_cnt =
1522 atomic64_read(&ci->i_release_count);
1523 req->r_dir_ordered_cnt =
1524 atomic64_read(&ci->i_ordered_count);
1525 req->r_readdir_cache_idx = 0;
1529 cache_ctl.index = req->r_readdir_cache_idx;
1530 fpos_offset = req->r_readdir_offset;
1532 /* FIXME: release caps/leases if error occurs */
1533 for (i = 0; i < rinfo->dir_nr; i++) {
1534 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1535 struct ceph_vino tvino, dvino;
1537 dname.name = rde->name;
1538 dname.len = rde->name_len;
1539 dname.hash = full_name_hash(parent, dname.name, dname.len);
1541 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1542 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1544 if (rinfo->hash_order) {
1545 u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1546 rde->name, rde->name_len);
1547 hash = ceph_frag_value(hash);
1548 if (hash != last_hash)
1551 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1553 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1557 dn = d_lookup(parent, &dname);
1558 dout("d_lookup on parent=%p name=%.*s got %p\n",
1559 parent, dname.len, dname.name, dn);
1562 dn = d_alloc(parent, &dname);
1563 dout("d_alloc %p '%.*s' = %p\n", parent,
1564 dname.len, dname.name, dn);
1566 dout("d_alloc badness\n");
1570 } else if (d_really_is_positive(dn) &&
1571 (ceph_ino(d_inode(dn)) != tvino.ino ||
1572 ceph_snap(d_inode(dn)) != tvino.snap)) {
1573 dout(" dn %p points to wrong inode %p\n",
1581 if (d_really_is_positive(dn)) {
1584 in = ceph_get_inode(parent->d_sb, tvino);
1586 dout("new_inode badness\n");
1594 ret = fill_inode(in, NULL, &rde->inode, NULL, session,
1595 req->r_request_started, -1,
1596 &req->r_caps_reservation);
1598 pr_err("fill_inode badness on %p\n", in);
1599 if (d_really_is_negative(dn))
1606 if (d_really_is_negative(dn)) {
1607 struct dentry *realdn;
1609 if (ceph_security_xattr_deadlock(in)) {
1610 dout(" skip splicing dn %p to inode %p"
1611 " (security xattr deadlock)\n", dn, in);
1617 realdn = splice_dentry(dn, in);
1618 if (IS_ERR(realdn)) {
1619 err = PTR_ERR(realdn);
1627 ceph_dentry(dn)->offset = rde->offset;
1629 dvino = ceph_vino(d_inode(parent));
1630 update_dentry_lease(dn, rde->lease, req->r_session,
1631 req->r_request_started, &tvino, &dvino);
1633 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1634 ret = fill_readdir_cache(d_inode(parent), dn,
1644 if (err == 0 && skipped == 0) {
1645 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1646 req->r_readdir_cache_idx = cache_ctl.index;
1648 ceph_readdir_cache_release(&cache_ctl);
1649 dout("readdir_prepopulate done\n");
1653 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1655 struct ceph_inode_info *ci = ceph_inode(inode);
1658 spin_lock(&ci->i_ceph_lock);
1659 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1660 i_size_write(inode, size);
1661 inode->i_blocks = calc_inode_blocks(size);
1663 ret = __ceph_should_report_size(ci);
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 (READ_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 .get_link = simple_get_link,
1877 .setattr = ceph_setattr,
1878 .getattr = ceph_getattr,
1879 .listxattr = ceph_listxattr,
1882 int __ceph_setattr(struct inode *inode, struct iattr *attr)
1884 struct ceph_inode_info *ci = ceph_inode(inode);
1885 const unsigned int ia_valid = attr->ia_valid;
1886 struct ceph_mds_request *req;
1887 struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
1888 struct ceph_cap_flush *prealloc_cf;
1890 int release = 0, dirtied = 0;
1893 int inode_dirty_flags = 0;
1894 bool lock_snap_rwsem = false;
1896 prealloc_cf = ceph_alloc_cap_flush();
1900 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1903 ceph_free_cap_flush(prealloc_cf);
1904 return PTR_ERR(req);
1907 spin_lock(&ci->i_ceph_lock);
1908 issued = __ceph_caps_issued(ci, NULL);
1910 if (!ci->i_head_snapc &&
1911 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
1912 lock_snap_rwsem = true;
1913 if (!down_read_trylock(&mdsc->snap_rwsem)) {
1914 spin_unlock(&ci->i_ceph_lock);
1915 down_read(&mdsc->snap_rwsem);
1916 spin_lock(&ci->i_ceph_lock);
1917 issued = __ceph_caps_issued(ci, NULL);
1921 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1923 if (ia_valid & ATTR_UID) {
1924 dout("setattr %p uid %d -> %d\n", inode,
1925 from_kuid(&init_user_ns, inode->i_uid),
1926 from_kuid(&init_user_ns, attr->ia_uid));
1927 if (issued & CEPH_CAP_AUTH_EXCL) {
1928 inode->i_uid = attr->ia_uid;
1929 dirtied |= CEPH_CAP_AUTH_EXCL;
1930 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1931 !uid_eq(attr->ia_uid, inode->i_uid)) {
1932 req->r_args.setattr.uid = cpu_to_le32(
1933 from_kuid(&init_user_ns, attr->ia_uid));
1934 mask |= CEPH_SETATTR_UID;
1935 release |= CEPH_CAP_AUTH_SHARED;
1938 if (ia_valid & ATTR_GID) {
1939 dout("setattr %p gid %d -> %d\n", inode,
1940 from_kgid(&init_user_ns, inode->i_gid),
1941 from_kgid(&init_user_ns, attr->ia_gid));
1942 if (issued & CEPH_CAP_AUTH_EXCL) {
1943 inode->i_gid = attr->ia_gid;
1944 dirtied |= CEPH_CAP_AUTH_EXCL;
1945 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1946 !gid_eq(attr->ia_gid, inode->i_gid)) {
1947 req->r_args.setattr.gid = cpu_to_le32(
1948 from_kgid(&init_user_ns, attr->ia_gid));
1949 mask |= CEPH_SETATTR_GID;
1950 release |= CEPH_CAP_AUTH_SHARED;
1953 if (ia_valid & ATTR_MODE) {
1954 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1956 if (issued & CEPH_CAP_AUTH_EXCL) {
1957 inode->i_mode = attr->ia_mode;
1958 dirtied |= CEPH_CAP_AUTH_EXCL;
1959 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1960 attr->ia_mode != inode->i_mode) {
1961 inode->i_mode = attr->ia_mode;
1962 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1963 mask |= CEPH_SETATTR_MODE;
1964 release |= CEPH_CAP_AUTH_SHARED;
1968 if (ia_valid & ATTR_ATIME) {
1969 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1970 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1971 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1972 if (issued & CEPH_CAP_FILE_EXCL) {
1973 ci->i_time_warp_seq++;
1974 inode->i_atime = attr->ia_atime;
1975 dirtied |= CEPH_CAP_FILE_EXCL;
1976 } else if ((issued & CEPH_CAP_FILE_WR) &&
1977 timespec_compare(&inode->i_atime,
1978 &attr->ia_atime) < 0) {
1979 inode->i_atime = attr->ia_atime;
1980 dirtied |= CEPH_CAP_FILE_WR;
1981 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1982 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1983 ceph_encode_timespec(&req->r_args.setattr.atime,
1985 mask |= CEPH_SETATTR_ATIME;
1986 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1990 if (ia_valid & ATTR_MTIME) {
1991 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1992 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1993 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1994 if (issued & CEPH_CAP_FILE_EXCL) {
1995 ci->i_time_warp_seq++;
1996 inode->i_mtime = attr->ia_mtime;
1997 dirtied |= CEPH_CAP_FILE_EXCL;
1998 } else if ((issued & CEPH_CAP_FILE_WR) &&
1999 timespec_compare(&inode->i_mtime,
2000 &attr->ia_mtime) < 0) {
2001 inode->i_mtime = attr->ia_mtime;
2002 dirtied |= CEPH_CAP_FILE_WR;
2003 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2004 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
2005 ceph_encode_timespec(&req->r_args.setattr.mtime,
2007 mask |= CEPH_SETATTR_MTIME;
2008 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2012 if (ia_valid & ATTR_SIZE) {
2013 dout("setattr %p size %lld -> %lld\n", inode,
2014 inode->i_size, attr->ia_size);
2015 if ((issued & CEPH_CAP_FILE_EXCL) &&
2016 attr->ia_size > inode->i_size) {
2017 i_size_write(inode, attr->ia_size);
2018 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2019 ci->i_reported_size = attr->ia_size;
2020 dirtied |= CEPH_CAP_FILE_EXCL;
2021 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2022 attr->ia_size != inode->i_size) {
2023 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2024 req->r_args.setattr.old_size =
2025 cpu_to_le64(inode->i_size);
2026 mask |= CEPH_SETATTR_SIZE;
2027 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
2032 /* these do nothing */
2033 if (ia_valid & ATTR_CTIME) {
2034 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2035 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2036 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
2037 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2038 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2039 only ? "ctime only" : "ignored");
2042 * if kernel wants to dirty ctime but nothing else,
2043 * we need to choose a cap to dirty under, or do
2044 * a almost-no-op setattr
2046 if (issued & CEPH_CAP_AUTH_EXCL)
2047 dirtied |= CEPH_CAP_AUTH_EXCL;
2048 else if (issued & CEPH_CAP_FILE_EXCL)
2049 dirtied |= CEPH_CAP_FILE_EXCL;
2050 else if (issued & CEPH_CAP_XATTR_EXCL)
2051 dirtied |= CEPH_CAP_XATTR_EXCL;
2053 mask |= CEPH_SETATTR_CTIME;
2056 if (ia_valid & ATTR_FILE)
2057 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2060 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2062 inode->i_ctime = attr->ia_ctime;
2066 spin_unlock(&ci->i_ceph_lock);
2067 if (lock_snap_rwsem)
2068 up_read(&mdsc->snap_rwsem);
2070 if (inode_dirty_flags)
2071 __mark_inode_dirty(inode, inode_dirty_flags);
2075 req->r_inode = inode;
2077 req->r_inode_drop = release;
2078 req->r_args.setattr.mask = cpu_to_le32(mask);
2079 req->r_num_caps = 1;
2080 req->r_stamp = attr->ia_ctime;
2081 err = ceph_mdsc_do_request(mdsc, NULL, req);
2083 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2084 ceph_cap_string(dirtied), mask);
2086 ceph_mdsc_put_request(req);
2087 ceph_free_cap_flush(prealloc_cf);
2089 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2090 __ceph_do_pending_vmtruncate(inode);
2098 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
2100 struct inode *inode = d_inode(dentry);
2103 if (ceph_snap(inode) != CEPH_NOSNAP)
2106 err = setattr_prepare(dentry, attr);
2110 err = __ceph_setattr(inode, attr);
2112 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2113 err = posix_acl_chmod(inode, attr->ia_mode);
2119 * Verify that we have a lease on the given mask. If not,
2120 * do a getattr against an mds.
2122 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2123 int mask, bool force)
2125 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2126 struct ceph_mds_client *mdsc = fsc->mdsc;
2127 struct ceph_mds_request *req;
2130 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2131 dout("do_getattr inode %p SNAPDIR\n", inode);
2135 dout("do_getattr inode %p mask %s mode 0%o\n",
2136 inode, ceph_cap_string(mask), inode->i_mode);
2137 if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
2140 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
2142 return PTR_ERR(req);
2143 req->r_inode = inode;
2145 req->r_num_caps = 1;
2146 req->r_args.getattr.mask = cpu_to_le32(mask);
2147 req->r_locked_page = locked_page;
2148 err = ceph_mdsc_do_request(mdsc, NULL, req);
2149 if (locked_page && err == 0) {
2150 u64 inline_version = req->r_reply_info.targeti.inline_version;
2151 if (inline_version == 0) {
2152 /* the reply is supposed to contain inline data */
2154 } else if (inline_version == CEPH_INLINE_NONE) {
2157 err = req->r_reply_info.targeti.inline_len;
2160 ceph_mdsc_put_request(req);
2161 dout("do_getattr result=%d\n", err);
2167 * Check inode permissions. We verify we have a valid value for
2168 * the AUTH cap, then call the generic handler.
2170 int ceph_permission(struct inode *inode, int mask)
2174 if (mask & MAY_NOT_BLOCK)
2177 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2180 err = generic_permission(inode, mask);
2185 * Get all attributes. Hopefully somedata we'll have a statlite()
2186 * and can limit the fields we require to be accurate.
2188 int ceph_getattr(const struct path *path, struct kstat *stat,
2189 u32 request_mask, unsigned int flags)
2191 struct inode *inode = d_inode(path->dentry);
2192 struct ceph_inode_info *ci = ceph_inode(inode);
2195 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
2197 generic_fillattr(inode, stat);
2198 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2199 if (ceph_snap(inode) != CEPH_NOSNAP)
2200 stat->dev = ceph_snap(inode);
2203 if (S_ISDIR(inode->i_mode)) {
2204 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2206 stat->size = ci->i_rbytes;
2208 stat->size = ci->i_files + ci->i_subdirs;
2210 stat->blksize = 65536;