1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/xattr.h>
13 #include <linux/posix_acl.h>
14 #include <linux/random.h>
15 #include <linux/sort.h>
16 #include <linux/iversion.h>
17 #include <linux/fscrypt.h>
20 #include "mds_client.h"
23 #include <linux/ceph/decode.h>
26 * Ceph inode operations
28 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
29 * setattr, etc.), xattr helpers, and helpers for assimilating
30 * metadata returned by the MDS into our cache.
32 * Also define helpers for doing asynchronous writeback, invalidation,
33 * and truncation for the benefit of those who can't afford to block
34 * (typically because they are in the message handler path).
37 static const struct inode_operations ceph_symlink_iops;
38 static const struct inode_operations ceph_encrypted_symlink_iops;
40 static void ceph_inode_work(struct work_struct *work);
43 * find or create an inode, given the ceph ino number
45 static int ceph_set_ino_cb(struct inode *inode, void *data)
47 struct ceph_inode_info *ci = ceph_inode(inode);
48 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
50 ci->i_vino = *(struct ceph_vino *)data;
51 inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
52 inode_set_iversion_raw(inode, 0);
53 percpu_counter_inc(&mdsc->metric.total_inodes);
59 * ceph_new_inode - allocate a new inode in advance of an expected create
60 * @dir: parent directory for new inode
61 * @dentry: dentry that may eventually point to new inode
62 * @mode: mode of new inode
63 * @as_ctx: pointer to inherited security context
65 * Allocate a new inode in advance of an operation to create a new inode.
66 * This allocates the inode and sets up the acl_sec_ctx with appropriate
67 * info for the new inode.
69 * Returns a pointer to the new inode or an ERR_PTR.
71 struct inode *ceph_new_inode(struct inode *dir, struct dentry *dentry,
72 umode_t *mode, struct ceph_acl_sec_ctx *as_ctx)
77 inode = new_inode(dir->i_sb);
79 return ERR_PTR(-ENOMEM);
81 inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
83 if (!S_ISLNK(*mode)) {
84 err = ceph_pre_init_acls(dir, mode, as_ctx);
90 inode->i_mode = *mode;
92 err = ceph_security_init_secctx(dentry, *mode, as_ctx);
97 * We'll skip setting fscrypt context for snapshots, leaving that for
100 if (ceph_snap(dir) != CEPH_SNAPDIR) {
101 err = ceph_fscrypt_prepare_context(dir, inode, as_ctx);
112 void ceph_as_ctx_to_req(struct ceph_mds_request *req,
113 struct ceph_acl_sec_ctx *as_ctx)
115 if (as_ctx->pagelist) {
116 req->r_pagelist = as_ctx->pagelist;
117 as_ctx->pagelist = NULL;
119 ceph_fscrypt_as_ctx_to_req(req, as_ctx);
123 * ceph_get_inode - find or create/hash a new inode
124 * @sb: superblock to search and allocate in
125 * @vino: vino to search for
126 * @newino: optional new inode to insert if one isn't found (may be NULL)
128 * Search for or insert a new inode into the hash for the given vino, and
129 * return a reference to it. If new is non-NULL, its reference is consumed.
131 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino,
132 struct inode *newino)
134 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(sb);
135 struct ceph_client *cl = mdsc->fsc->client;
138 if (ceph_vino_is_reserved(vino))
139 return ERR_PTR(-EREMOTEIO);
142 inode = inode_insert5(newino, (unsigned long)vino.ino,
143 ceph_ino_compare, ceph_set_ino_cb, &vino);
147 inode = iget5_locked(sb, (unsigned long)vino.ino,
148 ceph_ino_compare, ceph_set_ino_cb, &vino);
152 doutc(cl, "no inode found for %llx.%llx\n", vino.ino, vino.snap);
153 return ERR_PTR(-ENOMEM);
156 doutc(cl, "on %llx=%llx.%llx got %p new %d\n",
157 ceph_present_inode(inode), ceph_vinop(inode), inode,
158 !!(inode->i_state & I_NEW));
163 * get/constuct snapdir inode for a given directory
165 struct inode *ceph_get_snapdir(struct inode *parent)
167 struct ceph_client *cl = ceph_inode_to_client(parent);
168 struct ceph_vino vino = {
169 .ino = ceph_ino(parent),
170 .snap = CEPH_SNAPDIR,
172 struct inode *inode = ceph_get_inode(parent->i_sb, vino, NULL);
173 struct ceph_inode_info *ci = ceph_inode(inode);
179 if (!S_ISDIR(parent->i_mode)) {
180 pr_warn_once_client(cl, "bad snapdir parent type (mode=0%o)\n",
185 if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
186 pr_warn_once_client(cl, "bad snapdir inode type (mode=0%o)\n",
191 inode->i_mode = parent->i_mode;
192 inode->i_uid = parent->i_uid;
193 inode->i_gid = parent->i_gid;
194 inode_set_mtime_to_ts(inode, inode_get_mtime(parent));
195 inode_set_ctime_to_ts(inode, inode_get_ctime(parent));
196 inode_set_atime_to_ts(inode, inode_get_atime(parent));
198 ci->i_btime = ceph_inode(parent)->i_btime;
200 #ifdef CONFIG_FS_ENCRYPTION
201 /* if encrypted, just borrow fscrypt_auth from parent */
202 if (IS_ENCRYPTED(parent)) {
203 struct ceph_inode_info *pci = ceph_inode(parent);
205 ci->fscrypt_auth = kmemdup(pci->fscrypt_auth,
206 pci->fscrypt_auth_len,
208 if (ci->fscrypt_auth) {
209 inode->i_flags |= S_ENCRYPTED;
210 ci->fscrypt_auth_len = pci->fscrypt_auth_len;
212 doutc(cl, "Failed to alloc snapdir fscrypt_auth\n");
218 if (inode->i_state & I_NEW) {
219 inode->i_op = &ceph_snapdir_iops;
220 inode->i_fop = &ceph_snapdir_fops;
221 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
222 unlock_new_inode(inode);
227 if ((inode->i_state & I_NEW))
228 discard_new_inode(inode);
234 const struct inode_operations ceph_file_iops = {
235 .permission = ceph_permission,
236 .setattr = ceph_setattr,
237 .getattr = ceph_getattr,
238 .listxattr = ceph_listxattr,
239 .get_inode_acl = ceph_get_acl,
240 .set_acl = ceph_set_acl,
245 * We use a 'frag tree' to keep track of the MDS's directory fragments
246 * for a given inode (usually there is just a single fragment). We
247 * need to know when a child frag is delegated to a new MDS, or when
248 * it is flagged as replicated, so we can direct our requests
253 * find/create a frag in the tree
255 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
258 struct inode *inode = &ci->netfs.inode;
259 struct ceph_client *cl = ceph_inode_to_client(inode);
261 struct rb_node *parent = NULL;
262 struct ceph_inode_frag *frag;
265 p = &ci->i_fragtree.rb_node;
268 frag = rb_entry(parent, struct ceph_inode_frag, node);
269 c = ceph_frag_compare(f, frag->frag);
278 frag = kmalloc(sizeof(*frag), GFP_NOFS);
280 return ERR_PTR(-ENOMEM);
287 rb_link_node(&frag->node, parent, p);
288 rb_insert_color(&frag->node, &ci->i_fragtree);
290 doutc(cl, "added %p %llx.%llx frag %x\n", inode, ceph_vinop(inode), f);
295 * find a specific frag @f
297 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
299 struct rb_node *n = ci->i_fragtree.rb_node;
302 struct ceph_inode_frag *frag =
303 rb_entry(n, struct ceph_inode_frag, node);
304 int c = ceph_frag_compare(f, frag->frag);
316 * Choose frag containing the given value @v. If @pfrag is
317 * specified, copy the frag delegation info to the caller if
320 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
321 struct ceph_inode_frag *pfrag, int *found)
323 struct ceph_client *cl = ceph_inode_to_client(&ci->netfs.inode);
324 u32 t = ceph_frag_make(0, 0);
325 struct ceph_inode_frag *frag;
333 WARN_ON(!ceph_frag_contains_value(t, v));
334 frag = __ceph_find_frag(ci, t);
336 break; /* t is a leaf */
337 if (frag->split_by == 0) {
339 memcpy(pfrag, frag, sizeof(*pfrag));
346 nway = 1 << frag->split_by;
347 doutc(cl, "frag(%x) %x splits by %d (%d ways)\n", v, t,
348 frag->split_by, nway);
349 for (i = 0; i < nway; i++) {
350 n = ceph_frag_make_child(t, frag->split_by, i);
351 if (ceph_frag_contains_value(n, v)) {
358 doutc(cl, "frag(%x) = %x\n", v, t);
363 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
364 struct ceph_inode_frag *pfrag, int *found)
367 mutex_lock(&ci->i_fragtree_mutex);
368 ret = __ceph_choose_frag(ci, v, pfrag, found);
369 mutex_unlock(&ci->i_fragtree_mutex);
374 * Process dirfrag (delegation) info from the mds. Include leaf
375 * fragment in tree ONLY if ndist > 0. Otherwise, only
376 * branches/splits are included in i_fragtree)
378 static int ceph_fill_dirfrag(struct inode *inode,
379 struct ceph_mds_reply_dirfrag *dirinfo)
381 struct ceph_inode_info *ci = ceph_inode(inode);
382 struct ceph_client *cl = ceph_inode_to_client(inode);
383 struct ceph_inode_frag *frag;
384 u32 id = le32_to_cpu(dirinfo->frag);
385 int mds = le32_to_cpu(dirinfo->auth);
386 int ndist = le32_to_cpu(dirinfo->ndist);
391 spin_lock(&ci->i_ceph_lock);
393 diri_auth = ci->i_auth_cap->mds;
394 spin_unlock(&ci->i_ceph_lock);
396 if (mds == -1) /* CDIR_AUTH_PARENT */
399 mutex_lock(&ci->i_fragtree_mutex);
400 if (ndist == 0 && mds == diri_auth) {
401 /* no delegation info needed. */
402 frag = __ceph_find_frag(ci, id);
405 if (frag->split_by == 0) {
406 /* tree leaf, remove */
407 doutc(cl, "removed %p %llx.%llx frag %x (no ref)\n",
408 inode, ceph_vinop(inode), id);
409 rb_erase(&frag->node, &ci->i_fragtree);
412 /* tree branch, keep and clear */
413 doutc(cl, "cleared %p %llx.%llx frag %x referral\n",
414 inode, ceph_vinop(inode), id);
422 /* find/add this frag to store mds delegation info */
423 frag = __get_or_create_frag(ci, id);
425 /* this is not the end of the world; we can continue
426 with bad/inaccurate delegation info */
427 pr_err_client(cl, "ENOMEM on mds ref %p %llx.%llx fg %x\n",
428 inode, ceph_vinop(inode),
429 le32_to_cpu(dirinfo->frag));
435 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
436 for (i = 0; i < frag->ndist; i++)
437 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
438 doutc(cl, "%p %llx.%llx frag %x ndist=%d\n", inode,
439 ceph_vinop(inode), frag->frag, frag->ndist);
442 mutex_unlock(&ci->i_fragtree_mutex);
446 static int frag_tree_split_cmp(const void *l, const void *r)
448 struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
449 struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
450 return ceph_frag_compare(le32_to_cpu(ls->frag),
451 le32_to_cpu(rs->frag));
454 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
457 return f == ceph_frag_make(0, 0);
458 if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
460 return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
463 static int ceph_fill_fragtree(struct inode *inode,
464 struct ceph_frag_tree_head *fragtree,
465 struct ceph_mds_reply_dirfrag *dirinfo)
467 struct ceph_client *cl = ceph_inode_to_client(inode);
468 struct ceph_inode_info *ci = ceph_inode(inode);
469 struct ceph_inode_frag *frag, *prev_frag = NULL;
470 struct rb_node *rb_node;
471 unsigned i, split_by, nsplits;
475 mutex_lock(&ci->i_fragtree_mutex);
476 nsplits = le32_to_cpu(fragtree->nsplits);
477 if (nsplits != ci->i_fragtree_nsplits) {
479 } else if (nsplits) {
480 i = get_random_u32_below(nsplits);
481 id = le32_to_cpu(fragtree->splits[i].frag);
482 if (!__ceph_find_frag(ci, id))
484 } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
485 rb_node = rb_first(&ci->i_fragtree);
486 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
487 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
490 if (!update && dirinfo) {
491 id = le32_to_cpu(dirinfo->frag);
492 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
499 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
500 frag_tree_split_cmp, NULL);
503 doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
504 rb_node = rb_first(&ci->i_fragtree);
505 for (i = 0; i < nsplits; i++) {
506 id = le32_to_cpu(fragtree->splits[i].frag);
507 split_by = le32_to_cpu(fragtree->splits[i].by);
508 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
509 pr_err_client(cl, "%p %llx.%llx invalid split %d/%u, "
510 "frag %x split by %d\n", inode,
511 ceph_vinop(inode), i, nsplits, id, split_by);
516 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
517 if (ceph_frag_compare(frag->frag, id) >= 0) {
518 if (frag->frag != id)
521 rb_node = rb_next(rb_node);
524 rb_node = rb_next(rb_node);
525 /* delete stale split/leaf node */
526 if (frag->split_by > 0 ||
527 !is_frag_child(frag->frag, prev_frag)) {
528 rb_erase(&frag->node, &ci->i_fragtree);
529 if (frag->split_by > 0)
530 ci->i_fragtree_nsplits--;
536 frag = __get_or_create_frag(ci, id);
540 if (frag->split_by == 0)
541 ci->i_fragtree_nsplits++;
542 frag->split_by = split_by;
543 doutc(cl, " frag %x split by %d\n", frag->frag, frag->split_by);
547 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
548 rb_node = rb_next(rb_node);
549 /* delete stale split/leaf node */
550 if (frag->split_by > 0 ||
551 !is_frag_child(frag->frag, prev_frag)) {
552 rb_erase(&frag->node, &ci->i_fragtree);
553 if (frag->split_by > 0)
554 ci->i_fragtree_nsplits--;
559 mutex_unlock(&ci->i_fragtree_mutex);
564 * initialize a newly allocated inode.
566 struct inode *ceph_alloc_inode(struct super_block *sb)
568 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
569 struct ceph_inode_info *ci;
572 ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS);
576 doutc(fsc->client, "%p\n", &ci->netfs.inode);
578 /* Set parameters for the netfs library */
579 netfs_inode_init(&ci->netfs, &ceph_netfs_ops);
581 spin_lock_init(&ci->i_ceph_lock);
584 ci->i_inline_version = 0;
585 ci->i_time_warp_seq = 0;
586 ci->i_ceph_flags = 0;
587 atomic64_set(&ci->i_ordered_count, 1);
588 atomic64_set(&ci->i_release_count, 1);
589 atomic64_set(&ci->i_complete_seq[0], 0);
590 atomic64_set(&ci->i_complete_seq[1], 0);
591 ci->i_symlink = NULL;
596 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
597 memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
598 RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
600 ci->i_fragtree = RB_ROOT;
601 mutex_init(&ci->i_fragtree_mutex);
603 ci->i_xattrs.blob = NULL;
604 ci->i_xattrs.prealloc_blob = NULL;
605 ci->i_xattrs.dirty = false;
606 ci->i_xattrs.index = RB_ROOT;
607 ci->i_xattrs.count = 0;
608 ci->i_xattrs.names_size = 0;
609 ci->i_xattrs.vals_size = 0;
610 ci->i_xattrs.version = 0;
611 ci->i_xattrs.index_version = 0;
613 ci->i_caps = RB_ROOT;
614 ci->i_auth_cap = NULL;
615 ci->i_dirty_caps = 0;
616 ci->i_flushing_caps = 0;
617 INIT_LIST_HEAD(&ci->i_dirty_item);
618 INIT_LIST_HEAD(&ci->i_flushing_item);
619 ci->i_prealloc_cap_flush = NULL;
620 INIT_LIST_HEAD(&ci->i_cap_flush_list);
621 init_waitqueue_head(&ci->i_cap_wq);
622 ci->i_hold_caps_max = 0;
623 INIT_LIST_HEAD(&ci->i_cap_delay_list);
624 INIT_LIST_HEAD(&ci->i_cap_snaps);
625 ci->i_head_snapc = NULL;
628 ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
629 for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
630 ci->i_nr_by_mode[i] = 0;
632 mutex_init(&ci->i_truncate_mutex);
633 ci->i_truncate_seq = 0;
634 ci->i_truncate_size = 0;
635 ci->i_truncate_pending = 0;
636 ci->i_truncate_pagecache_size = 0;
639 ci->i_reported_size = 0;
640 ci->i_wanted_max_size = 0;
641 ci->i_requested_max_size = 0;
645 ci->i_rdcache_ref = 0;
649 ci->i_wrbuffer_ref = 0;
650 ci->i_wrbuffer_ref_head = 0;
651 atomic_set(&ci->i_filelock_ref, 0);
652 atomic_set(&ci->i_shared_gen, 1);
653 ci->i_rdcache_gen = 0;
654 ci->i_rdcache_revoking = 0;
656 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
657 INIT_LIST_HEAD(&ci->i_unsafe_iops);
658 spin_lock_init(&ci->i_unsafe_lock);
660 ci->i_snap_realm = NULL;
661 INIT_LIST_HEAD(&ci->i_snap_realm_item);
662 INIT_LIST_HEAD(&ci->i_snap_flush_item);
664 INIT_WORK(&ci->i_work, ceph_inode_work);
666 memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
667 #ifdef CONFIG_FS_ENCRYPTION
668 ci->fscrypt_auth = NULL;
669 ci->fscrypt_auth_len = 0;
671 return &ci->netfs.inode;
674 void ceph_free_inode(struct inode *inode)
676 struct ceph_inode_info *ci = ceph_inode(inode);
678 kfree(ci->i_symlink);
679 #ifdef CONFIG_FS_ENCRYPTION
680 kfree(ci->fscrypt_auth);
682 fscrypt_free_inode(inode);
683 kmem_cache_free(ceph_inode_cachep, ci);
686 void ceph_evict_inode(struct inode *inode)
688 struct ceph_inode_info *ci = ceph_inode(inode);
689 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
690 struct ceph_client *cl = ceph_inode_to_client(inode);
691 struct ceph_inode_frag *frag;
694 doutc(cl, "%p ino %llx.%llx\n", inode, ceph_vinop(inode));
696 percpu_counter_dec(&mdsc->metric.total_inodes);
698 truncate_inode_pages_final(&inode->i_data);
699 if (inode->i_state & I_PINNING_FSCACHE_WB)
700 ceph_fscache_unuse_cookie(inode, true);
703 ceph_fscache_unregister_inode_cookie(ci);
704 fscrypt_put_encryption_info(inode);
706 __ceph_remove_caps(ci);
708 if (__ceph_has_quota(ci, QUOTA_GET_ANY))
709 ceph_adjust_quota_realms_count(inode, false);
712 * we may still have a snap_realm reference if there are stray
713 * caps in i_snap_caps.
715 if (ci->i_snap_realm) {
716 if (ceph_snap(inode) == CEPH_NOSNAP) {
717 doutc(cl, " dropping residual ref to snap realm %p\n",
719 ceph_change_snap_realm(inode, NULL);
721 ceph_put_snapid_map(mdsc, ci->i_snapid_map);
722 ci->i_snap_realm = NULL;
726 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
727 frag = rb_entry(n, struct ceph_inode_frag, node);
728 rb_erase(n, &ci->i_fragtree);
731 ci->i_fragtree_nsplits = 0;
733 __ceph_destroy_xattrs(ci);
734 if (ci->i_xattrs.blob)
735 ceph_buffer_put(ci->i_xattrs.blob);
736 if (ci->i_xattrs.prealloc_blob)
737 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
739 ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
740 ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
743 static inline blkcnt_t calc_inode_blocks(u64 size)
745 return (size + (1<<9) - 1) >> 9;
749 * Helpers to fill in size, ctime, mtime, and atime. We have to be
750 * careful because either the client or MDS may have more up to date
751 * info, depending on which capabilities are held, and whether
752 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
753 * and size are monotonically increasing, except when utimes() or
754 * truncate() increments the corresponding _seq values.)
756 int ceph_fill_file_size(struct inode *inode, int issued,
757 u32 truncate_seq, u64 truncate_size, u64 size)
759 struct ceph_client *cl = ceph_inode_to_client(inode);
760 struct ceph_inode_info *ci = ceph_inode(inode);
762 loff_t isize = i_size_read(inode);
764 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
765 (truncate_seq == ci->i_truncate_seq && size > isize)) {
766 doutc(cl, "size %lld -> %llu\n", isize, size);
767 if (size > 0 && S_ISDIR(inode->i_mode)) {
768 pr_err_client(cl, "non-zero size for directory\n");
771 i_size_write(inode, size);
772 inode->i_blocks = calc_inode_blocks(size);
774 * If we're expanding, then we should be able to just update
775 * the existing cookie.
778 ceph_fscache_update(inode);
779 ci->i_reported_size = size;
780 if (truncate_seq != ci->i_truncate_seq) {
781 doutc(cl, "truncate_seq %u -> %u\n",
782 ci->i_truncate_seq, truncate_seq);
783 ci->i_truncate_seq = truncate_seq;
785 /* the MDS should have revoked these caps */
786 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_RD |
787 CEPH_CAP_FILE_LAZYIO));
789 * If we hold relevant caps, or in the case where we're
790 * not the only client referencing this file and we
791 * don't hold those caps, then we need to check whether
792 * the file is either opened or mmaped
794 if ((issued & (CEPH_CAP_FILE_CACHE|
795 CEPH_CAP_FILE_BUFFER)) ||
796 mapping_mapped(inode->i_mapping) ||
797 __ceph_is_file_opened(ci)) {
798 ci->i_truncate_pending++;
805 * It's possible that the new sizes of the two consecutive
806 * size truncations will be in the same fscrypt last block,
807 * and we need to truncate the corresponding page caches
810 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0) {
811 doutc(cl, "truncate_size %lld -> %llu, encrypted %d\n",
812 ci->i_truncate_size, truncate_size,
813 !!IS_ENCRYPTED(inode));
815 ci->i_truncate_size = truncate_size;
817 if (IS_ENCRYPTED(inode)) {
818 doutc(cl, "truncate_pagecache_size %lld -> %llu\n",
819 ci->i_truncate_pagecache_size, size);
820 ci->i_truncate_pagecache_size = size;
822 ci->i_truncate_pagecache_size = truncate_size;
828 void ceph_fill_file_time(struct inode *inode, int issued,
829 u64 time_warp_seq, struct timespec64 *ctime,
830 struct timespec64 *mtime, struct timespec64 *atime)
832 struct ceph_client *cl = ceph_inode_to_client(inode);
833 struct ceph_inode_info *ci = ceph_inode(inode);
834 struct timespec64 ictime = inode_get_ctime(inode);
837 if (issued & (CEPH_CAP_FILE_EXCL|
839 CEPH_CAP_FILE_BUFFER|
841 CEPH_CAP_XATTR_EXCL)) {
842 if (ci->i_version == 0 ||
843 timespec64_compare(ctime, &ictime) > 0) {
844 doutc(cl, "ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
845 ictime.tv_sec, ictime.tv_nsec,
846 ctime->tv_sec, ctime->tv_nsec);
847 inode_set_ctime_to_ts(inode, *ctime);
849 if (ci->i_version == 0 ||
850 ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
851 /* the MDS did a utimes() */
852 doutc(cl, "mtime %lld.%09ld -> %lld.%09ld tw %d -> %d\n",
853 inode_get_mtime_sec(inode),
854 inode_get_mtime_nsec(inode),
855 mtime->tv_sec, mtime->tv_nsec,
856 ci->i_time_warp_seq, (int)time_warp_seq);
858 inode_set_mtime_to_ts(inode, *mtime);
859 inode_set_atime_to_ts(inode, *atime);
860 ci->i_time_warp_seq = time_warp_seq;
861 } else if (time_warp_seq == ci->i_time_warp_seq) {
862 struct timespec64 ts;
864 /* nobody did utimes(); take the max */
865 ts = inode_get_mtime(inode);
866 if (timespec64_compare(mtime, &ts) > 0) {
867 doutc(cl, "mtime %lld.%09ld -> %lld.%09ld inc\n",
868 ts.tv_sec, ts.tv_nsec,
869 mtime->tv_sec, mtime->tv_nsec);
870 inode_set_mtime_to_ts(inode, *mtime);
872 ts = inode_get_atime(inode);
873 if (timespec64_compare(atime, &ts) > 0) {
874 doutc(cl, "atime %lld.%09ld -> %lld.%09ld inc\n",
875 ts.tv_sec, ts.tv_nsec,
876 atime->tv_sec, atime->tv_nsec);
877 inode_set_atime_to_ts(inode, *atime);
879 } else if (issued & CEPH_CAP_FILE_EXCL) {
880 /* we did a utimes(); ignore mds values */
885 /* we have no write|excl caps; whatever the MDS says is true */
886 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
887 inode_set_ctime_to_ts(inode, *ctime);
888 inode_set_mtime_to_ts(inode, *mtime);
889 inode_set_atime_to_ts(inode, *atime);
890 ci->i_time_warp_seq = time_warp_seq;
895 if (warn) /* time_warp_seq shouldn't go backwards */
896 doutc(cl, "%p mds time_warp_seq %llu < %u\n", inode,
897 time_warp_seq, ci->i_time_warp_seq);
900 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
901 static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
903 int enclen, u8 **decsym)
905 struct ceph_client *cl = mdsc->fsc->client;
909 sym = kmalloc(enclen + 1, GFP_NOFS);
913 declen = ceph_base64_decode(encsym, enclen, sym);
916 "can't decode symlink (%d). Content: %.*s\n",
917 declen, enclen, encsym);
921 sym[declen + 1] = '\0';
926 static int decode_encrypted_symlink(struct ceph_mds_client *mdsc,
928 int symlen, u8 **decsym)
935 * Populate an inode based on info from mds. May be called on new or
938 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
939 struct ceph_mds_reply_info_in *iinfo,
940 struct ceph_mds_reply_dirfrag *dirinfo,
941 struct ceph_mds_session *session, int cap_fmode,
942 struct ceph_cap_reservation *caps_reservation)
944 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
945 struct ceph_client *cl = mdsc->fsc->client;
946 struct ceph_mds_reply_inode *info = iinfo->in;
947 struct ceph_inode_info *ci = ceph_inode(inode);
948 int issued, new_issued, info_caps;
949 struct timespec64 mtime, atime, ctime;
950 struct ceph_buffer *xattr_blob = NULL;
951 struct ceph_buffer *old_blob = NULL;
952 struct ceph_string *pool_ns = NULL;
953 struct ceph_cap *new_cap = NULL;
956 bool queue_trunc = false;
957 bool new_version = false;
958 bool fill_inline = false;
959 umode_t mode = le32_to_cpu(info->mode);
960 dev_t rdev = le32_to_cpu(info->rdev);
962 lockdep_assert_held(&mdsc->snap_rwsem);
964 doutc(cl, "%p ino %llx.%llx v %llu had %llu\n", inode, ceph_vinop(inode),
965 le64_to_cpu(info->version), ci->i_version);
967 /* Once I_NEW is cleared, we can't change type or dev numbers */
968 if (inode->i_state & I_NEW) {
969 inode->i_mode = mode;
971 if (inode_wrong_type(inode, mode)) {
972 pr_warn_once_client(cl,
973 "inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
974 ceph_vinop(inode), inode->i_mode, mode);
978 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
979 pr_warn_once_client(cl,
980 "dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
981 ceph_vinop(inode), MAJOR(inode->i_rdev),
982 MINOR(inode->i_rdev), MAJOR(rdev),
988 info_caps = le32_to_cpu(info->cap.caps);
990 /* prealloc new cap struct */
991 if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
992 new_cap = ceph_get_cap(mdsc, caps_reservation);
998 * prealloc xattr data, if it looks like we'll need it. only
999 * if len > 4 (meaning there are actually xattrs; the first 4
1000 * bytes are the xattr count).
1002 if (iinfo->xattr_len > 4) {
1003 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
1005 pr_err_client(cl, "ENOMEM xattr blob %d bytes\n",
1009 if (iinfo->pool_ns_len > 0)
1010 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
1011 iinfo->pool_ns_len);
1013 if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
1014 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
1016 spin_lock(&ci->i_ceph_lock);
1019 * provided version will be odd if inode value is projected,
1020 * even if stable. skip the update if we have newer stable
1021 * info (ours>=theirs, e.g. due to racing mds replies), unless
1022 * we are getting projected (unstable) info (in which case the
1023 * version is odd, and we want ours>theirs).
1029 if (ci->i_version == 0 ||
1030 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1031 le64_to_cpu(info->version) > (ci->i_version & ~1)))
1034 /* Update change_attribute */
1035 inode_set_max_iversion_raw(inode, iinfo->change_attr);
1037 __ceph_caps_issued(ci, &issued);
1038 issued |= __ceph_caps_dirty(ci);
1039 new_issued = ~issued & info_caps;
1041 __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
1043 #ifdef CONFIG_FS_ENCRYPTION
1044 if (iinfo->fscrypt_auth_len &&
1045 ((inode->i_state & I_NEW) || (ci->fscrypt_auth_len == 0))) {
1046 kfree(ci->fscrypt_auth);
1047 ci->fscrypt_auth_len = iinfo->fscrypt_auth_len;
1048 ci->fscrypt_auth = iinfo->fscrypt_auth;
1049 iinfo->fscrypt_auth = NULL;
1050 iinfo->fscrypt_auth_len = 0;
1051 inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED);
1055 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
1056 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
1057 inode->i_mode = mode;
1058 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
1059 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
1060 doutc(cl, "%p %llx.%llx mode 0%o uid.gid %d.%d\n", inode,
1061 ceph_vinop(inode), inode->i_mode,
1062 from_kuid(&init_user_ns, inode->i_uid),
1063 from_kgid(&init_user_ns, inode->i_gid));
1064 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
1065 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
1068 /* directories have fl_stripe_unit set to zero */
1069 if (IS_ENCRYPTED(inode))
1070 inode->i_blkbits = CEPH_FSCRYPT_BLOCK_SHIFT;
1071 else if (le32_to_cpu(info->layout.fl_stripe_unit))
1073 fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
1075 inode->i_blkbits = CEPH_BLOCK_SHIFT;
1077 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
1078 (issued & CEPH_CAP_LINK_EXCL) == 0)
1079 set_nlink(inode, le32_to_cpu(info->nlink));
1081 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
1082 /* be careful with mtime, atime, size */
1083 ceph_decode_timespec64(&atime, &info->atime);
1084 ceph_decode_timespec64(&mtime, &info->mtime);
1085 ceph_decode_timespec64(&ctime, &info->ctime);
1086 ceph_fill_file_time(inode, issued,
1087 le32_to_cpu(info->time_warp_seq),
1088 &ctime, &mtime, &atime);
1091 if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
1092 ci->i_files = le64_to_cpu(info->files);
1093 ci->i_subdirs = le64_to_cpu(info->subdirs);
1097 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
1098 u64 size = le64_to_cpu(info->size);
1099 s64 old_pool = ci->i_layout.pool_id;
1100 struct ceph_string *old_ns;
1102 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
1103 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
1104 lockdep_is_held(&ci->i_ceph_lock));
1105 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
1107 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
1108 ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
1112 if (IS_ENCRYPTED(inode) && size &&
1113 iinfo->fscrypt_file_len == sizeof(__le64)) {
1114 u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
1116 if (size == round_up(fsize, CEPH_FSCRYPT_BLOCK_SIZE)) {
1120 "fscrypt size mismatch: size=%llu fscrypt_file=%llu, discarding fscrypt_file size.\n",
1125 queue_trunc = ceph_fill_file_size(inode, issued,
1126 le32_to_cpu(info->truncate_seq),
1127 le64_to_cpu(info->truncate_size),
1129 /* only update max_size on auth cap */
1130 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
1131 ci->i_max_size != le64_to_cpu(info->max_size)) {
1132 doutc(cl, "max_size %lld -> %llu\n",
1133 ci->i_max_size, le64_to_cpu(info->max_size));
1134 ci->i_max_size = le64_to_cpu(info->max_size);
1138 /* layout and rstat are not tracked by capability, update them if
1139 * the inode info is from auth mds */
1140 if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
1141 if (S_ISDIR(inode->i_mode)) {
1142 ci->i_dir_layout = iinfo->dir_layout;
1143 ci->i_rbytes = le64_to_cpu(info->rbytes);
1144 ci->i_rfiles = le64_to_cpu(info->rfiles);
1145 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
1146 ci->i_dir_pin = iinfo->dir_pin;
1147 ci->i_rsnaps = iinfo->rsnaps;
1148 ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
1153 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
1154 if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) &&
1155 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
1156 if (ci->i_xattrs.blob)
1157 old_blob = ci->i_xattrs.blob;
1158 ci->i_xattrs.blob = xattr_blob;
1160 memcpy(ci->i_xattrs.blob->vec.iov_base,
1161 iinfo->xattr_data, iinfo->xattr_len);
1162 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
1163 ceph_forget_all_cached_acls(inode);
1164 ceph_security_invalidate_secctx(inode);
1168 /* finally update i_version */
1169 if (le64_to_cpu(info->version) > ci->i_version)
1170 ci->i_version = le64_to_cpu(info->version);
1172 inode->i_mapping->a_ops = &ceph_aops;
1174 switch (inode->i_mode & S_IFMT) {
1179 inode->i_blkbits = PAGE_SHIFT;
1180 init_special_inode(inode, inode->i_mode, rdev);
1181 inode->i_op = &ceph_file_iops;
1184 inode->i_op = &ceph_file_iops;
1185 inode->i_fop = &ceph_file_fops;
1188 if (!ci->i_symlink) {
1189 u32 symlen = iinfo->symlink_len;
1192 spin_unlock(&ci->i_ceph_lock);
1194 if (IS_ENCRYPTED(inode)) {
1195 if (symlen != i_size_read(inode))
1197 "%p %llx.%llx BAD symlink size %lld\n",
1198 inode, ceph_vinop(inode),
1199 i_size_read(inode));
1201 err = decode_encrypted_symlink(mdsc, iinfo->symlink,
1202 symlen, (u8 **)&sym);
1205 "decoding encrypted symlink failed: %d\n",
1210 i_size_write(inode, symlen);
1211 inode->i_blocks = calc_inode_blocks(symlen);
1213 if (symlen != i_size_read(inode)) {
1215 "%p %llx.%llx BAD symlink size %lld\n",
1216 inode, ceph_vinop(inode),
1217 i_size_read(inode));
1218 i_size_write(inode, symlen);
1219 inode->i_blocks = calc_inode_blocks(symlen);
1223 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
1228 spin_lock(&ci->i_ceph_lock);
1230 ci->i_symlink = sym;
1232 kfree(sym); /* lost a race */
1235 if (IS_ENCRYPTED(inode)) {
1237 * Encrypted symlinks need to be decrypted before we can
1238 * cache their targets in i_link. Don't touch it here.
1240 inode->i_op = &ceph_encrypted_symlink_iops;
1242 inode->i_link = ci->i_symlink;
1243 inode->i_op = &ceph_symlink_iops;
1247 inode->i_op = &ceph_dir_iops;
1248 inode->i_fop = &ceph_dir_fops;
1251 pr_err_client(cl, "%p %llx.%llx BAD mode 0%o\n", inode,
1252 ceph_vinop(inode), inode->i_mode);
1255 /* were we issued a capability? */
1257 if (ceph_snap(inode) == CEPH_NOSNAP) {
1258 ceph_add_cap(inode, session,
1259 le64_to_cpu(info->cap.cap_id),
1261 le32_to_cpu(info->cap.wanted),
1262 le32_to_cpu(info->cap.seq),
1263 le32_to_cpu(info->cap.mseq),
1264 le64_to_cpu(info->cap.realm),
1265 info->cap.flags, &new_cap);
1267 /* set dir completion flag? */
1268 if (S_ISDIR(inode->i_mode) &&
1269 ci->i_files == 0 && ci->i_subdirs == 0 &&
1270 (info_caps & CEPH_CAP_FILE_SHARED) &&
1271 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1272 !__ceph_dir_is_complete(ci)) {
1273 doutc(cl, " marking %p complete (empty)\n",
1275 i_size_write(inode, 0);
1276 __ceph_dir_set_complete(ci,
1277 atomic64_read(&ci->i_release_count),
1278 atomic64_read(&ci->i_ordered_count));
1283 doutc(cl, " %p got snap_caps %s\n", inode,
1284 ceph_cap_string(info_caps));
1285 ci->i_snap_caps |= info_caps;
1289 if (iinfo->inline_version > 0 &&
1290 iinfo->inline_version >= ci->i_inline_version) {
1291 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1292 ci->i_inline_version = iinfo->inline_version;
1293 if (ceph_has_inline_data(ci) &&
1294 (locked_page || (info_caps & cache_caps)))
1298 if (cap_fmode >= 0) {
1300 pr_warn_client(cl, "mds issued no caps on %llx.%llx\n",
1302 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1305 spin_unlock(&ci->i_ceph_lock);
1307 ceph_fscache_register_inode_cookie(inode);
1310 ceph_fill_inline_data(inode, locked_page,
1311 iinfo->inline_data, iinfo->inline_len);
1314 wake_up_all(&ci->i_cap_wq);
1316 /* queue truncate if we saw i_size decrease */
1318 ceph_queue_vmtruncate(inode);
1320 /* populate frag tree */
1321 if (S_ISDIR(inode->i_mode))
1322 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1324 /* update delegation info? */
1326 ceph_fill_dirfrag(inode, dirinfo);
1331 ceph_put_cap(mdsc, new_cap);
1332 ceph_buffer_put(old_blob);
1333 ceph_buffer_put(xattr_blob);
1334 ceph_put_string(pool_ns);
1339 * caller should hold session s_mutex and dentry->d_lock.
1341 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1342 struct ceph_mds_reply_lease *lease,
1343 struct ceph_mds_session *session,
1344 unsigned long from_time,
1345 struct ceph_mds_session **old_lease_session)
1347 struct ceph_client *cl = ceph_inode_to_client(dir);
1348 struct ceph_dentry_info *di = ceph_dentry(dentry);
1349 unsigned mask = le16_to_cpu(lease->mask);
1350 long unsigned duration = le32_to_cpu(lease->duration_ms);
1351 long unsigned ttl = from_time + (duration * HZ) / 1000;
1352 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1354 doutc(cl, "%p duration %lu ms ttl %lu\n", dentry, duration, ttl);
1356 /* only track leases on regular dentries */
1357 if (ceph_snap(dir) != CEPH_NOSNAP)
1360 if (mask & CEPH_LEASE_PRIMARY_LINK)
1361 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1363 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1365 di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1366 if (!(mask & CEPH_LEASE_VALID)) {
1367 __ceph_dentry_dir_lease_touch(di);
1371 if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1372 time_before(ttl, di->time))
1373 return; /* we already have a newer lease. */
1375 if (di->lease_session && di->lease_session != session) {
1376 *old_lease_session = di->lease_session;
1377 di->lease_session = NULL;
1380 if (!di->lease_session)
1381 di->lease_session = ceph_get_mds_session(session);
1382 di->lease_gen = atomic_read(&session->s_cap_gen);
1383 di->lease_seq = le32_to_cpu(lease->seq);
1384 di->lease_renew_after = half_ttl;
1385 di->lease_renew_from = 0;
1388 __ceph_dentry_lease_touch(di);
1391 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1392 struct ceph_mds_reply_lease *lease,
1393 struct ceph_mds_session *session,
1394 unsigned long from_time)
1396 struct ceph_mds_session *old_lease_session = NULL;
1397 spin_lock(&dentry->d_lock);
1398 __update_dentry_lease(dir, dentry, lease, session, from_time,
1399 &old_lease_session);
1400 spin_unlock(&dentry->d_lock);
1401 ceph_put_mds_session(old_lease_session);
1405 * update dentry lease without having parent inode locked
1407 static void update_dentry_lease_careful(struct dentry *dentry,
1408 struct ceph_mds_reply_lease *lease,
1409 struct ceph_mds_session *session,
1410 unsigned long from_time,
1411 char *dname, u32 dname_len,
1412 struct ceph_vino *pdvino,
1413 struct ceph_vino *ptvino)
1417 struct ceph_mds_session *old_lease_session = NULL;
1419 spin_lock(&dentry->d_lock);
1420 /* make sure dentry's name matches target */
1421 if (dentry->d_name.len != dname_len ||
1422 memcmp(dentry->d_name.name, dname, dname_len))
1425 dir = d_inode(dentry->d_parent);
1426 /* make sure parent matches dvino */
1427 if (!ceph_ino_compare(dir, pdvino))
1430 /* make sure dentry's inode matches target. NULL ptvino means that
1431 * we expect a negative dentry */
1433 if (d_really_is_negative(dentry))
1435 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1438 if (d_really_is_positive(dentry))
1442 __update_dentry_lease(dir, dentry, lease, session,
1443 from_time, &old_lease_session);
1445 spin_unlock(&dentry->d_lock);
1446 ceph_put_mds_session(old_lease_session);
1450 * splice a dentry to an inode.
1451 * caller must hold directory i_rwsem for this to be safe.
1453 static int splice_dentry(struct dentry **pdn, struct inode *in)
1455 struct ceph_client *cl = ceph_inode_to_client(in);
1456 struct dentry *dn = *pdn;
1457 struct dentry *realdn;
1459 BUG_ON(d_inode(dn));
1461 if (S_ISDIR(in->i_mode)) {
1462 /* If inode is directory, d_splice_alias() below will remove
1463 * 'realdn' from its origin parent. We need to ensure that
1464 * origin parent's readdir cache will not reference 'realdn'
1466 realdn = d_find_any_alias(in);
1468 struct ceph_dentry_info *di = ceph_dentry(realdn);
1469 spin_lock(&realdn->d_lock);
1471 realdn->d_op->d_prune(realdn);
1474 di->lease_shared_gen = 0;
1477 spin_unlock(&realdn->d_lock);
1482 /* dn must be unhashed */
1483 if (!d_unhashed(dn))
1485 realdn = d_splice_alias(in, dn);
1486 if (IS_ERR(realdn)) {
1487 pr_err_client(cl, "error %ld %p inode %p ino %llx.%llx\n",
1488 PTR_ERR(realdn), dn, in, ceph_vinop(in));
1489 return PTR_ERR(realdn);
1493 doutc(cl, "dn %p (%d) spliced with %p (%d) inode %p ino %llx.%llx\n",
1494 dn, d_count(dn), realdn, d_count(realdn),
1495 d_inode(realdn), ceph_vinop(d_inode(realdn)));
1499 BUG_ON(!ceph_dentry(dn));
1500 doutc(cl, "dn %p attached to %p ino %llx.%llx\n", dn,
1501 d_inode(dn), ceph_vinop(d_inode(dn)));
1507 * Incorporate results into the local cache. This is either just
1508 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1511 * A reply may contain
1512 * a directory inode along with a dentry.
1513 * and/or a target inode
1515 * Called with snap_rwsem (read).
1517 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1519 struct ceph_mds_session *session = req->r_session;
1520 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1521 struct inode *in = NULL;
1522 struct ceph_vino tvino, dvino;
1523 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
1524 struct ceph_client *cl = fsc->client;
1527 doutc(cl, "%p is_dentry %d is_target %d\n", req,
1528 rinfo->head->is_dentry, rinfo->head->is_target);
1530 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1531 doutc(cl, "reply is empty!\n");
1532 if (rinfo->head->result == 0 && req->r_parent)
1533 ceph_invalidate_dir_request(req);
1537 if (rinfo->head->is_dentry) {
1538 struct inode *dir = req->r_parent;
1541 err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1542 rinfo->dirfrag, session, -1,
1543 &req->r_caps_reservation);
1550 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1551 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1552 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1553 bool is_nokey = false;
1555 struct dentry *dn, *parent;
1556 struct fscrypt_str oname = FSTR_INIT(NULL, 0);
1557 struct ceph_fname fname = { .dir = dir,
1558 .name = rinfo->dname,
1559 .ctext = rinfo->altname,
1560 .name_len = rinfo->dname_len,
1561 .ctext_len = rinfo->altname_len };
1563 BUG_ON(!rinfo->head->is_target);
1564 BUG_ON(req->r_dentry);
1566 parent = d_find_any_alias(dir);
1569 err = ceph_fname_alloc_buffer(dir, &oname);
1575 err = ceph_fname_to_usr(&fname, NULL, &oname, &is_nokey);
1578 ceph_fname_free_buffer(dir, &oname);
1581 dname.name = oname.name;
1582 dname.len = oname.len;
1583 dname.hash = full_name_hash(parent, dname.name, dname.len);
1584 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1585 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1587 dn = d_lookup(parent, &dname);
1588 doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
1589 parent, dname.len, dname.name, dn);
1592 dn = d_alloc(parent, &dname);
1593 doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
1594 dname.len, dname.name, dn);
1597 ceph_fname_free_buffer(dir, &oname);
1602 spin_lock(&dn->d_lock);
1603 dn->d_flags |= DCACHE_NOKEY_NAME;
1604 spin_unlock(&dn->d_lock);
1607 } else if (d_really_is_positive(dn) &&
1608 (ceph_ino(d_inode(dn)) != tvino.ino ||
1609 ceph_snap(d_inode(dn)) != tvino.snap)) {
1610 doutc(cl, " dn %p points to wrong inode %p\n",
1612 ceph_dir_clear_ordered(dir);
1617 ceph_fname_free_buffer(dir, &oname);
1624 if (rinfo->head->is_target) {
1625 /* Should be filled in by handle_reply */
1626 BUG_ON(!req->r_target_inode);
1628 in = req->r_target_inode;
1629 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1631 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1632 !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1633 rinfo->head->result == 0) ? req->r_fmode : -1,
1634 &req->r_caps_reservation);
1636 pr_err_client(cl, "badness %p %llx.%llx\n", in,
1638 req->r_target_inode = NULL;
1639 if (in->i_state & I_NEW)
1640 discard_new_inode(in);
1645 if (in->i_state & I_NEW)
1646 unlock_new_inode(in);
1650 * ignore null lease/binding on snapdir ENOENT, or else we
1651 * will have trouble splicing in the virtual snapdir later
1653 if (rinfo->head->is_dentry &&
1654 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1655 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1656 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1657 fsc->mount_options->snapdir_name,
1658 req->r_dentry->d_name.len))) {
1660 * lookup link rename : null -> possibly existing inode
1661 * mknod symlink mkdir : null -> new inode
1662 * unlink : linked -> null
1664 struct inode *dir = req->r_parent;
1665 struct dentry *dn = req->r_dentry;
1666 bool have_dir_cap, have_lease;
1670 BUG_ON(d_inode(dn->d_parent) != dir);
1672 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1673 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1675 BUG_ON(ceph_ino(dir) != dvino.ino);
1676 BUG_ON(ceph_snap(dir) != dvino.snap);
1678 /* do we have a lease on the whole dir? */
1680 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1681 CEPH_CAP_FILE_SHARED);
1683 /* do we have a dn lease? */
1684 have_lease = have_dir_cap ||
1685 le32_to_cpu(rinfo->dlease->duration_ms);
1687 doutc(cl, "no dentry lease or dir cap\n");
1690 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1691 struct inode *olddir = req->r_old_dentry_dir;
1694 doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1695 req->r_old_dentry, req->r_old_dentry, dn, dn);
1696 doutc(cl, "doing d_move %p -> %p\n", req->r_old_dentry, dn);
1698 /* d_move screws up sibling dentries' offsets */
1699 ceph_dir_clear_ordered(dir);
1700 ceph_dir_clear_ordered(olddir);
1702 d_move(req->r_old_dentry, dn);
1703 doutc(cl, " src %p '%pd' dst %p '%pd'\n",
1704 req->r_old_dentry, req->r_old_dentry, dn, dn);
1706 /* ensure target dentry is invalidated, despite
1707 rehashing bug in vfs_rename_dir */
1708 ceph_invalidate_dentry_lease(dn);
1710 doutc(cl, "dn %p gets new offset %lld\n",
1712 ceph_dentry(req->r_old_dentry)->offset);
1714 /* swap r_dentry and r_old_dentry in case that
1715 * splice_dentry() gets called later. This is safe
1716 * because no other place will use them */
1717 req->r_dentry = req->r_old_dentry;
1718 req->r_old_dentry = dn;
1723 if (!rinfo->head->is_target) {
1724 doutc(cl, "null dentry\n");
1725 if (d_really_is_positive(dn)) {
1726 doutc(cl, "d_delete %p\n", dn);
1727 ceph_dir_clear_ordered(dir);
1729 } else if (have_lease) {
1734 if (!d_unhashed(dn) && have_lease)
1735 update_dentry_lease(dir, dn,
1736 rinfo->dlease, session,
1737 req->r_request_started);
1741 /* attach proper inode */
1742 if (d_really_is_negative(dn)) {
1743 ceph_dir_clear_ordered(dir);
1745 err = splice_dentry(&req->r_dentry, in);
1748 dn = req->r_dentry; /* may have spliced */
1749 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1750 doutc(cl, " %p links to %p %llx.%llx, not %llx.%llx\n",
1751 dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1758 update_dentry_lease(dir, dn,
1759 rinfo->dlease, session,
1760 req->r_request_started);
1762 doutc(cl, " final dn %p\n", dn);
1763 } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1764 req->r_op == CEPH_MDS_OP_MKSNAP) &&
1765 test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1766 !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1767 struct inode *dir = req->r_parent;
1769 /* fill out a snapdir LOOKUPSNAP dentry */
1771 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1772 BUG_ON(!req->r_dentry);
1773 doutc(cl, " linking snapped dir %p to dn %p\n", in,
1775 ceph_dir_clear_ordered(dir);
1777 err = splice_dentry(&req->r_dentry, in);
1780 } else if (rinfo->head->is_dentry && req->r_dentry) {
1781 /* parent inode is not locked, be carefull */
1782 struct ceph_vino *ptvino = NULL;
1783 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1784 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1785 if (rinfo->head->is_target) {
1786 tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1787 tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1790 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1791 session, req->r_request_started,
1792 rinfo->dname, rinfo->dname_len,
1796 doutc(cl, "done err=%d\n", err);
1801 * Prepopulate our cache with readdir results, leases, etc.
1803 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1804 struct ceph_mds_session *session)
1806 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1807 struct ceph_client *cl = session->s_mdsc->fsc->client;
1810 for (i = 0; i < rinfo->dir_nr; i++) {
1811 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1812 struct ceph_vino vino;
1816 vino.ino = le64_to_cpu(rde->inode.in->ino);
1817 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1819 in = ceph_get_inode(req->r_dentry->d_sb, vino, NULL);
1822 doutc(cl, "badness got %d\n", err);
1825 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1826 -1, &req->r_caps_reservation);
1828 pr_err_client(cl, "inode badness on %p got %d\n", in,
1831 if (in->i_state & I_NEW) {
1833 discard_new_inode(in);
1835 } else if (in->i_state & I_NEW) {
1836 unlock_new_inode(in);
1845 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1849 put_page(ctl->page);
1854 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1855 struct ceph_readdir_cache_control *ctl,
1856 struct ceph_mds_request *req)
1858 struct ceph_client *cl = ceph_inode_to_client(dir);
1859 struct ceph_inode_info *ci = ceph_inode(dir);
1860 unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1861 unsigned idx = ctl->index % nsize;
1862 pgoff_t pgoff = ctl->index / nsize;
1864 if (!ctl->page || pgoff != page_index(ctl->page)) {
1865 ceph_readdir_cache_release(ctl);
1867 ctl->page = grab_cache_page(&dir->i_data, pgoff);
1869 ctl->page = find_lock_page(&dir->i_data, pgoff);
1872 return idx == 0 ? -ENOMEM : 0;
1874 /* reading/filling the cache are serialized by
1875 * i_rwsem, no need to use page lock */
1876 unlock_page(ctl->page);
1877 ctl->dentries = kmap(ctl->page);
1879 memset(ctl->dentries, 0, PAGE_SIZE);
1882 if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1883 req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1884 doutc(cl, "dn %p idx %d\n", dn, ctl->index);
1885 ctl->dentries[idx] = dn;
1888 doutc(cl, "disable readdir cache\n");
1894 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1895 struct ceph_mds_session *session)
1897 struct dentry *parent = req->r_dentry;
1898 struct inode *inode = d_inode(parent);
1899 struct ceph_inode_info *ci = ceph_inode(inode);
1900 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1901 struct ceph_client *cl = session->s_mdsc->fsc->client;
1905 int err = 0, skipped = 0, ret, i;
1906 u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1909 struct ceph_readdir_cache_control cache_ctl = {};
1911 if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1912 return readdir_prepopulate_inodes_only(req, session);
1914 if (rinfo->hash_order) {
1916 last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1918 strlen(req->r_path2));
1919 last_hash = ceph_frag_value(last_hash);
1920 } else if (rinfo->offset_hash) {
1921 /* mds understands offset_hash */
1922 WARN_ON_ONCE(req->r_readdir_offset != 2);
1923 last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1927 if (rinfo->dir_dir &&
1928 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1929 doutc(cl, "got new frag %x -> %x\n", frag,
1930 le32_to_cpu(rinfo->dir_dir->frag));
1931 frag = le32_to_cpu(rinfo->dir_dir->frag);
1932 if (!rinfo->hash_order)
1933 req->r_readdir_offset = 2;
1936 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1937 doutc(cl, "%d items under SNAPDIR dn %p\n",
1938 rinfo->dir_nr, parent);
1940 doutc(cl, "%d items under dn %p\n", rinfo->dir_nr, parent);
1942 ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1944 if (ceph_frag_is_leftmost(frag) &&
1945 req->r_readdir_offset == 2 &&
1946 !(rinfo->hash_order && last_hash)) {
1947 /* note dir version at start of readdir so we can
1948 * tell if any dentries get dropped */
1949 req->r_dir_release_cnt =
1950 atomic64_read(&ci->i_release_count);
1951 req->r_dir_ordered_cnt =
1952 atomic64_read(&ci->i_ordered_count);
1953 req->r_readdir_cache_idx = 0;
1957 cache_ctl.index = req->r_readdir_cache_idx;
1958 fpos_offset = req->r_readdir_offset;
1960 /* FIXME: release caps/leases if error occurs */
1961 for (i = 0; i < rinfo->dir_nr; i++) {
1962 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1963 struct ceph_vino tvino;
1965 dname.name = rde->name;
1966 dname.len = rde->name_len;
1967 dname.hash = full_name_hash(parent, dname.name, dname.len);
1969 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1970 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1972 if (rinfo->hash_order) {
1973 u32 hash = ceph_frag_value(rde->raw_hash);
1974 if (hash != last_hash)
1977 rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1979 rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1983 dn = d_lookup(parent, &dname);
1984 doutc(cl, "d_lookup on parent=%p name=%.*s got %p\n",
1985 parent, dname.len, dname.name, dn);
1988 dn = d_alloc(parent, &dname);
1989 doutc(cl, "d_alloc %p '%.*s' = %p\n", parent,
1990 dname.len, dname.name, dn);
1992 doutc(cl, "d_alloc badness\n");
1996 if (rde->is_nokey) {
1997 spin_lock(&dn->d_lock);
1998 dn->d_flags |= DCACHE_NOKEY_NAME;
1999 spin_unlock(&dn->d_lock);
2001 } else if (d_really_is_positive(dn) &&
2002 (ceph_ino(d_inode(dn)) != tvino.ino ||
2003 ceph_snap(d_inode(dn)) != tvino.snap)) {
2004 struct ceph_dentry_info *di = ceph_dentry(dn);
2005 doutc(cl, " dn %p points to wrong inode %p\n",
2008 spin_lock(&dn->d_lock);
2009 if (di->offset > 0 &&
2010 di->lease_shared_gen ==
2011 atomic_read(&ci->i_shared_gen)) {
2012 __ceph_dir_clear_ordered(ci);
2015 spin_unlock(&dn->d_lock);
2023 if (d_really_is_positive(dn)) {
2026 in = ceph_get_inode(parent->d_sb, tvino, NULL);
2028 doutc(cl, "new_inode badness\n");
2036 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
2037 -1, &req->r_caps_reservation);
2039 pr_err_client(cl, "badness on %p %llx.%llx\n", in,
2041 if (d_really_is_negative(dn)) {
2042 if (in->i_state & I_NEW) {
2044 discard_new_inode(in);
2052 if (in->i_state & I_NEW)
2053 unlock_new_inode(in);
2055 if (d_really_is_negative(dn)) {
2056 if (ceph_security_xattr_deadlock(in)) {
2057 doutc(cl, " skip splicing dn %p to inode %p"
2058 " (security xattr deadlock)\n", dn, in);
2064 err = splice_dentry(&dn, in);
2069 ceph_dentry(dn)->offset = rde->offset;
2071 update_dentry_lease(d_inode(parent), dn,
2072 rde->lease, req->r_session,
2073 req->r_request_started);
2075 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
2076 ret = fill_readdir_cache(d_inode(parent), dn,
2085 if (err == 0 && skipped == 0) {
2086 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
2087 req->r_readdir_cache_idx = cache_ctl.index;
2089 ceph_readdir_cache_release(&cache_ctl);
2090 doutc(cl, "done\n");
2094 bool ceph_inode_set_size(struct inode *inode, loff_t size)
2096 struct ceph_client *cl = ceph_inode_to_client(inode);
2097 struct ceph_inode_info *ci = ceph_inode(inode);
2100 spin_lock(&ci->i_ceph_lock);
2101 doutc(cl, "set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
2102 i_size_write(inode, size);
2103 ceph_fscache_update(inode);
2104 inode->i_blocks = calc_inode_blocks(size);
2106 ret = __ceph_should_report_size(ci);
2108 spin_unlock(&ci->i_ceph_lock);
2113 void ceph_queue_inode_work(struct inode *inode, int work_bit)
2115 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2116 struct ceph_client *cl = fsc->client;
2117 struct ceph_inode_info *ci = ceph_inode(inode);
2118 set_bit(work_bit, &ci->i_work_mask);
2121 if (queue_work(fsc->inode_wq, &ci->i_work)) {
2122 doutc(cl, "%p %llx.%llx mask=%lx\n", inode,
2123 ceph_vinop(inode), ci->i_work_mask);
2125 doutc(cl, "%p %llx.%llx already queued, mask=%lx\n",
2126 inode, ceph_vinop(inode), ci->i_work_mask);
2131 static void ceph_do_invalidate_pages(struct inode *inode)
2133 struct ceph_client *cl = ceph_inode_to_client(inode);
2134 struct ceph_inode_info *ci = ceph_inode(inode);
2138 ceph_fscache_invalidate(inode, false);
2140 mutex_lock(&ci->i_truncate_mutex);
2142 if (ceph_inode_is_shutdown(inode)) {
2143 pr_warn_ratelimited_client(cl,
2144 "%p %llx.%llx is shut down\n", inode,
2146 mapping_set_error(inode->i_mapping, -EIO);
2147 truncate_pagecache(inode, 0);
2148 mutex_unlock(&ci->i_truncate_mutex);
2152 spin_lock(&ci->i_ceph_lock);
2153 doutc(cl, "%p %llx.%llx gen %d revoking %d\n", inode,
2154 ceph_vinop(inode), ci->i_rdcache_gen, ci->i_rdcache_revoking);
2155 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
2156 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2158 spin_unlock(&ci->i_ceph_lock);
2159 mutex_unlock(&ci->i_truncate_mutex);
2162 orig_gen = ci->i_rdcache_gen;
2163 spin_unlock(&ci->i_ceph_lock);
2165 if (invalidate_inode_pages2(inode->i_mapping) < 0) {
2166 pr_err_client(cl, "invalidate_inode_pages2 %llx.%llx failed\n",
2170 spin_lock(&ci->i_ceph_lock);
2171 if (orig_gen == ci->i_rdcache_gen &&
2172 orig_gen == ci->i_rdcache_revoking) {
2173 doutc(cl, "%p %llx.%llx gen %d successful\n", inode,
2174 ceph_vinop(inode), ci->i_rdcache_gen);
2175 ci->i_rdcache_revoking--;
2178 doutc(cl, "%p %llx.%llx gen %d raced, now %d revoking %d\n",
2179 inode, ceph_vinop(inode), orig_gen, ci->i_rdcache_gen,
2180 ci->i_rdcache_revoking);
2181 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
2184 spin_unlock(&ci->i_ceph_lock);
2185 mutex_unlock(&ci->i_truncate_mutex);
2188 ceph_check_caps(ci, 0);
2192 * Make sure any pending truncation is applied before doing anything
2193 * that may depend on it.
2195 void __ceph_do_pending_vmtruncate(struct inode *inode)
2197 struct ceph_client *cl = ceph_inode_to_client(inode);
2198 struct ceph_inode_info *ci = ceph_inode(inode);
2200 int wrbuffer_refs, finish = 0;
2202 mutex_lock(&ci->i_truncate_mutex);
2204 spin_lock(&ci->i_ceph_lock);
2205 if (ci->i_truncate_pending == 0) {
2206 doutc(cl, "%p %llx.%llx none pending\n", inode,
2208 spin_unlock(&ci->i_ceph_lock);
2209 mutex_unlock(&ci->i_truncate_mutex);
2214 * make sure any dirty snapped pages are flushed before we
2215 * possibly truncate them.. so write AND block!
2217 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
2218 spin_unlock(&ci->i_ceph_lock);
2219 doutc(cl, "%p %llx.%llx flushing snaps first\n", inode,
2221 filemap_write_and_wait_range(&inode->i_data, 0,
2222 inode->i_sb->s_maxbytes);
2226 /* there should be no reader or writer */
2227 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
2229 to = ci->i_truncate_pagecache_size;
2230 wrbuffer_refs = ci->i_wrbuffer_ref;
2231 doutc(cl, "%p %llx.%llx (%d) to %lld\n", inode, ceph_vinop(inode),
2232 ci->i_truncate_pending, to);
2233 spin_unlock(&ci->i_ceph_lock);
2235 ceph_fscache_resize(inode, to);
2236 truncate_pagecache(inode, to);
2238 spin_lock(&ci->i_ceph_lock);
2239 if (to == ci->i_truncate_pagecache_size) {
2240 ci->i_truncate_pending = 0;
2243 spin_unlock(&ci->i_ceph_lock);
2247 mutex_unlock(&ci->i_truncate_mutex);
2249 if (wrbuffer_refs == 0)
2250 ceph_check_caps(ci, 0);
2252 wake_up_all(&ci->i_cap_wq);
2255 static void ceph_inode_work(struct work_struct *work)
2257 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
2259 struct inode *inode = &ci->netfs.inode;
2260 struct ceph_client *cl = ceph_inode_to_client(inode);
2262 if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
2263 doutc(cl, "writeback %p %llx.%llx\n", inode, ceph_vinop(inode));
2264 filemap_fdatawrite(&inode->i_data);
2266 if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
2267 ceph_do_invalidate_pages(inode);
2269 if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
2270 __ceph_do_pending_vmtruncate(inode);
2272 if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
2273 ceph_check_caps(ci, 0);
2275 if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
2276 ceph_flush_snaps(ci, NULL);
2281 static const char *ceph_encrypted_get_link(struct dentry *dentry,
2282 struct inode *inode,
2283 struct delayed_call *done)
2285 struct ceph_inode_info *ci = ceph_inode(inode);
2288 return ERR_PTR(-ECHILD);
2290 return fscrypt_get_symlink(inode, ci->i_symlink, i_size_read(inode),
2294 static int ceph_encrypted_symlink_getattr(struct mnt_idmap *idmap,
2295 const struct path *path,
2296 struct kstat *stat, u32 request_mask,
2297 unsigned int query_flags)
2301 ret = ceph_getattr(idmap, path, stat, request_mask, query_flags);
2304 return fscrypt_symlink_getattr(path, stat);
2310 static const struct inode_operations ceph_symlink_iops = {
2311 .get_link = simple_get_link,
2312 .setattr = ceph_setattr,
2313 .getattr = ceph_getattr,
2314 .listxattr = ceph_listxattr,
2317 static const struct inode_operations ceph_encrypted_symlink_iops = {
2318 .get_link = ceph_encrypted_get_link,
2319 .setattr = ceph_setattr,
2320 .getattr = ceph_encrypted_symlink_getattr,
2321 .listxattr = ceph_listxattr,
2325 * Transfer the encrypted last block to the MDS and the MDS
2326 * will help update it when truncating a smaller size.
2328 * We don't support a PAGE_SIZE that is smaller than the
2329 * CEPH_FSCRYPT_BLOCK_SIZE.
2331 static int fill_fscrypt_truncate(struct inode *inode,
2332 struct ceph_mds_request *req,
2335 struct ceph_client *cl = ceph_inode_to_client(inode);
2336 struct ceph_inode_info *ci = ceph_inode(inode);
2337 int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
2338 loff_t pos, orig_pos = round_down(attr->ia_size,
2339 CEPH_FSCRYPT_BLOCK_SIZE);
2340 u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
2341 struct ceph_pagelist *pagelist = NULL;
2342 struct kvec iov = {0};
2343 struct iov_iter iter;
2344 struct page *page = NULL;
2345 struct ceph_fscrypt_truncate_size_header header;
2347 int len = CEPH_FSCRYPT_BLOCK_SIZE;
2348 loff_t i_size = i_size_read(inode);
2349 int got, ret, issued;
2352 ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
2356 issued = __ceph_caps_issued(ci, NULL);
2358 doutc(cl, "size %lld -> %lld got cap refs on %s, issued %s\n",
2359 i_size, attr->ia_size, ceph_cap_string(got),
2360 ceph_cap_string(issued));
2362 /* Try to writeback the dirty pagecaches */
2363 if (issued & (CEPH_CAP_FILE_BUFFER)) {
2364 loff_t lend = orig_pos + CEPH_FSCRYPT_BLOCK_SHIFT - 1;
2366 ret = filemap_write_and_wait_range(inode->i_mapping,
2372 page = __page_cache_alloc(GFP_KERNEL);
2378 pagelist = ceph_pagelist_alloc(GFP_KERNEL);
2384 iov.iov_base = kmap_local_page(page);
2386 iov_iter_kvec(&iter, READ, &iov, 1, len);
2389 ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objver);
2393 /* Insert the header first */
2396 header.change_attr = cpu_to_le64(inode_peek_iversion_raw(inode));
2399 * Always set the block_size to CEPH_FSCRYPT_BLOCK_SIZE,
2400 * because in MDS it may need this to do the truncate.
2402 header.block_size = cpu_to_le32(CEPH_FSCRYPT_BLOCK_SIZE);
2405 * If we hit a hole here, we should just skip filling
2406 * the fscrypt for the request, because once the fscrypt
2407 * is enabled, the file will be split into many blocks
2408 * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
2409 * has a hole, the hole size should be multiple of block
2412 * If the Rados object doesn't exist, it will be set to 0.
2415 doutc(cl, "hit hole, ppos %lld < size %lld\n", pos, i_size);
2417 header.data_len = cpu_to_le32(8 + 8 + 4);
2418 header.file_offset = 0;
2421 header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
2422 header.file_offset = cpu_to_le64(orig_pos);
2424 doutc(cl, "encrypt block boff/bsize %d/%lu\n", boff,
2425 CEPH_FSCRYPT_BLOCK_SIZE);
2427 /* truncate and zero out the extra contents for the last block */
2428 memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
2430 /* encrypt the last block */
2431 ret = ceph_fscrypt_encrypt_block_inplace(inode, page,
2432 CEPH_FSCRYPT_BLOCK_SIZE,
2439 /* Insert the header */
2440 ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
2444 if (header.block_size) {
2445 /* Append the last block contents to pagelist */
2446 ret = ceph_pagelist_append(pagelist, iov.iov_base,
2447 CEPH_FSCRYPT_BLOCK_SIZE);
2451 req->r_pagelist = pagelist;
2453 doutc(cl, "%p %llx.%llx size dropping cap refs on %s\n", inode,
2454 ceph_vinop(inode), ceph_cap_string(got));
2455 ceph_put_cap_refs(ci, got);
2457 kunmap_local(iov.iov_base);
2459 __free_pages(page, 0);
2460 if (ret && pagelist)
2461 ceph_pagelist_release(pagelist);
2465 int __ceph_setattr(struct mnt_idmap *idmap, struct inode *inode,
2466 struct iattr *attr, struct ceph_iattr *cia)
2468 struct ceph_inode_info *ci = ceph_inode(inode);
2469 unsigned int ia_valid = attr->ia_valid;
2470 struct ceph_mds_request *req;
2471 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(inode->i_sb)->mdsc;
2472 struct ceph_client *cl = ceph_inode_to_client(inode);
2473 struct ceph_cap_flush *prealloc_cf;
2474 loff_t isize = i_size_read(inode);
2476 int release = 0, dirtied = 0;
2479 int inode_dirty_flags = 0;
2480 bool lock_snap_rwsem = false;
2482 int truncate_retry = 20; /* The RMW will take around 50ms */
2485 prealloc_cf = ceph_alloc_cap_flush();
2489 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2492 ceph_free_cap_flush(prealloc_cf);
2493 return PTR_ERR(req);
2496 fill_fscrypt = false;
2497 spin_lock(&ci->i_ceph_lock);
2498 issued = __ceph_caps_issued(ci, NULL);
2500 if (!ci->i_head_snapc &&
2501 (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2502 lock_snap_rwsem = true;
2503 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2504 spin_unlock(&ci->i_ceph_lock);
2505 down_read(&mdsc->snap_rwsem);
2506 spin_lock(&ci->i_ceph_lock);
2507 issued = __ceph_caps_issued(ci, NULL);
2511 doutc(cl, "%p %llx.%llx issued %s\n", inode, ceph_vinop(inode),
2512 ceph_cap_string(issued));
2513 #if IS_ENABLED(CONFIG_FS_ENCRYPTION)
2514 if (cia && cia->fscrypt_auth) {
2515 u32 len = ceph_fscrypt_auth_len(cia->fscrypt_auth);
2517 if (len > sizeof(*cia->fscrypt_auth)) {
2519 spin_unlock(&ci->i_ceph_lock);
2523 doutc(cl, "%p %llx.%llx fscrypt_auth len %u to %u)\n", inode,
2524 ceph_vinop(inode), ci->fscrypt_auth_len, len);
2526 /* It should never be re-set once set */
2527 WARN_ON_ONCE(ci->fscrypt_auth);
2529 if (issued & CEPH_CAP_AUTH_EXCL) {
2530 dirtied |= CEPH_CAP_AUTH_EXCL;
2531 kfree(ci->fscrypt_auth);
2532 ci->fscrypt_auth = (u8 *)cia->fscrypt_auth;
2533 ci->fscrypt_auth_len = len;
2534 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2535 ci->fscrypt_auth_len != len ||
2536 memcmp(ci->fscrypt_auth, cia->fscrypt_auth, len)) {
2537 req->r_fscrypt_auth = cia->fscrypt_auth;
2538 mask |= CEPH_SETATTR_FSCRYPT_AUTH;
2539 release |= CEPH_CAP_AUTH_SHARED;
2541 cia->fscrypt_auth = NULL;
2544 if (cia && cia->fscrypt_auth) {
2546 spin_unlock(&ci->i_ceph_lock);
2549 #endif /* CONFIG_FS_ENCRYPTION */
2551 if (ia_valid & ATTR_UID) {
2552 kuid_t fsuid = from_vfsuid(idmap, i_user_ns(inode), attr->ia_vfsuid);
2554 doutc(cl, "%p %llx.%llx uid %d -> %d\n", inode,
2556 from_kuid(&init_user_ns, inode->i_uid),
2557 from_kuid(&init_user_ns, attr->ia_uid));
2558 if (issued & CEPH_CAP_AUTH_EXCL) {
2559 inode->i_uid = fsuid;
2560 dirtied |= CEPH_CAP_AUTH_EXCL;
2561 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2562 !uid_eq(fsuid, inode->i_uid)) {
2563 req->r_args.setattr.uid = cpu_to_le32(
2564 from_kuid(&init_user_ns, fsuid));
2565 mask |= CEPH_SETATTR_UID;
2566 release |= CEPH_CAP_AUTH_SHARED;
2569 if (ia_valid & ATTR_GID) {
2570 kgid_t fsgid = from_vfsgid(idmap, i_user_ns(inode), attr->ia_vfsgid);
2572 doutc(cl, "%p %llx.%llx gid %d -> %d\n", inode,
2574 from_kgid(&init_user_ns, inode->i_gid),
2575 from_kgid(&init_user_ns, attr->ia_gid));
2576 if (issued & CEPH_CAP_AUTH_EXCL) {
2577 inode->i_gid = fsgid;
2578 dirtied |= CEPH_CAP_AUTH_EXCL;
2579 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2580 !gid_eq(fsgid, inode->i_gid)) {
2581 req->r_args.setattr.gid = cpu_to_le32(
2582 from_kgid(&init_user_ns, fsgid));
2583 mask |= CEPH_SETATTR_GID;
2584 release |= CEPH_CAP_AUTH_SHARED;
2587 if (ia_valid & ATTR_MODE) {
2588 doutc(cl, "%p %llx.%llx mode 0%o -> 0%o\n", inode,
2589 ceph_vinop(inode), inode->i_mode, attr->ia_mode);
2590 if (issued & CEPH_CAP_AUTH_EXCL) {
2591 inode->i_mode = attr->ia_mode;
2592 dirtied |= CEPH_CAP_AUTH_EXCL;
2593 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2594 attr->ia_mode != inode->i_mode) {
2595 inode->i_mode = attr->ia_mode;
2596 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2597 mask |= CEPH_SETATTR_MODE;
2598 release |= CEPH_CAP_AUTH_SHARED;
2602 if (ia_valid & ATTR_ATIME) {
2603 struct timespec64 atime = inode_get_atime(inode);
2605 doutc(cl, "%p %llx.%llx atime %lld.%09ld -> %lld.%09ld\n",
2606 inode, ceph_vinop(inode),
2607 atime.tv_sec, atime.tv_nsec,
2608 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2609 if (issued & CEPH_CAP_FILE_EXCL) {
2610 ci->i_time_warp_seq++;
2611 inode_set_atime_to_ts(inode, attr->ia_atime);
2612 dirtied |= CEPH_CAP_FILE_EXCL;
2613 } else if ((issued & CEPH_CAP_FILE_WR) &&
2614 timespec64_compare(&atime,
2615 &attr->ia_atime) < 0) {
2616 inode_set_atime_to_ts(inode, attr->ia_atime);
2617 dirtied |= CEPH_CAP_FILE_WR;
2618 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2619 !timespec64_equal(&atime, &attr->ia_atime)) {
2620 ceph_encode_timespec64(&req->r_args.setattr.atime,
2622 mask |= CEPH_SETATTR_ATIME;
2623 release |= CEPH_CAP_FILE_SHARED |
2624 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2627 if (ia_valid & ATTR_SIZE) {
2628 doutc(cl, "%p %llx.%llx size %lld -> %lld\n", inode,
2629 ceph_vinop(inode), isize, attr->ia_size);
2631 * Only when the new size is smaller and not aligned to
2632 * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
2634 if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
2635 (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
2636 mask |= CEPH_SETATTR_SIZE;
2637 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2638 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2639 set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2640 mask |= CEPH_SETATTR_FSCRYPT_FILE;
2641 req->r_args.setattr.size =
2642 cpu_to_le64(round_up(attr->ia_size,
2643 CEPH_FSCRYPT_BLOCK_SIZE));
2644 req->r_args.setattr.old_size =
2645 cpu_to_le64(round_up(isize,
2646 CEPH_FSCRYPT_BLOCK_SIZE));
2647 req->r_fscrypt_file = attr->ia_size;
2648 fill_fscrypt = true;
2649 } else if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2650 if (attr->ia_size > isize) {
2651 i_size_write(inode, attr->ia_size);
2652 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2653 ci->i_reported_size = attr->ia_size;
2654 dirtied |= CEPH_CAP_FILE_EXCL;
2655 ia_valid |= ATTR_MTIME;
2657 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2658 attr->ia_size != isize) {
2659 mask |= CEPH_SETATTR_SIZE;
2660 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2661 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2662 if (IS_ENCRYPTED(inode) && attr->ia_size) {
2663 set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
2664 mask |= CEPH_SETATTR_FSCRYPT_FILE;
2665 req->r_args.setattr.size =
2666 cpu_to_le64(round_up(attr->ia_size,
2667 CEPH_FSCRYPT_BLOCK_SIZE));
2668 req->r_args.setattr.old_size =
2669 cpu_to_le64(round_up(isize,
2670 CEPH_FSCRYPT_BLOCK_SIZE));
2671 req->r_fscrypt_file = attr->ia_size;
2673 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2674 req->r_args.setattr.old_size = cpu_to_le64(isize);
2675 req->r_fscrypt_file = 0;
2679 if (ia_valid & ATTR_MTIME) {
2680 struct timespec64 mtime = inode_get_mtime(inode);
2682 doutc(cl, "%p %llx.%llx mtime %lld.%09ld -> %lld.%09ld\n",
2683 inode, ceph_vinop(inode),
2684 mtime.tv_sec, mtime.tv_nsec,
2685 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2686 if (issued & CEPH_CAP_FILE_EXCL) {
2687 ci->i_time_warp_seq++;
2688 inode_set_mtime_to_ts(inode, attr->ia_mtime);
2689 dirtied |= CEPH_CAP_FILE_EXCL;
2690 } else if ((issued & CEPH_CAP_FILE_WR) &&
2691 timespec64_compare(&mtime, &attr->ia_mtime) < 0) {
2692 inode_set_mtime_to_ts(inode, attr->ia_mtime);
2693 dirtied |= CEPH_CAP_FILE_WR;
2694 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2695 !timespec64_equal(&mtime, &attr->ia_mtime)) {
2696 ceph_encode_timespec64(&req->r_args.setattr.mtime,
2698 mask |= CEPH_SETATTR_MTIME;
2699 release |= CEPH_CAP_FILE_SHARED |
2700 CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2704 /* these do nothing */
2705 if (ia_valid & ATTR_CTIME) {
2706 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2707 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2708 doutc(cl, "%p %llx.%llx ctime %lld.%09ld -> %lld.%09ld (%s)\n",
2709 inode, ceph_vinop(inode),
2710 inode_get_ctime_sec(inode),
2711 inode_get_ctime_nsec(inode),
2712 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2713 only ? "ctime only" : "ignored");
2716 * if kernel wants to dirty ctime but nothing else,
2717 * we need to choose a cap to dirty under, or do
2718 * a almost-no-op setattr
2720 if (issued & CEPH_CAP_AUTH_EXCL)
2721 dirtied |= CEPH_CAP_AUTH_EXCL;
2722 else if (issued & CEPH_CAP_FILE_EXCL)
2723 dirtied |= CEPH_CAP_FILE_EXCL;
2724 else if (issued & CEPH_CAP_XATTR_EXCL)
2725 dirtied |= CEPH_CAP_XATTR_EXCL;
2727 mask |= CEPH_SETATTR_CTIME;
2730 if (ia_valid & ATTR_FILE)
2731 doutc(cl, "%p %llx.%llx ATTR_FILE ... hrm!\n", inode,
2735 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2737 inode_set_ctime_to_ts(inode, attr->ia_ctime);
2738 inode_inc_iversion_raw(inode);
2742 spin_unlock(&ci->i_ceph_lock);
2743 if (lock_snap_rwsem) {
2744 up_read(&mdsc->snap_rwsem);
2745 lock_snap_rwsem = false;
2748 if (inode_dirty_flags)
2749 __mark_inode_dirty(inode, inode_dirty_flags);
2752 req->r_inode = inode;
2754 req->r_inode_drop = release;
2755 req->r_args.setattr.mask = cpu_to_le32(mask);
2756 req->r_num_caps = 1;
2757 req->r_stamp = attr->ia_ctime;
2759 err = fill_fscrypt_truncate(inode, req, attr);
2765 * The truncate request will return -EAGAIN when the
2766 * last block has been updated just before the MDS
2767 * successfully gets the xlock for the FILE lock. To
2768 * avoid corrupting the file contents we need to retry
2771 err = ceph_mdsc_do_request(mdsc, NULL, req);
2772 if (err == -EAGAIN && truncate_retry--) {
2773 doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote), retry it!\n",
2774 inode, ceph_vinop(inode), err,
2775 ceph_cap_string(dirtied), mask);
2776 ceph_mdsc_put_request(req);
2777 ceph_free_cap_flush(prealloc_cf);
2782 doutc(cl, "%p %llx.%llx result=%d (%s locally, %d remote)\n", inode,
2783 ceph_vinop(inode), err, ceph_cap_string(dirtied), mask);
2785 ceph_mdsc_put_request(req);
2786 ceph_free_cap_flush(prealloc_cf);
2788 if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2789 __ceph_do_pending_vmtruncate(inode);
2797 int ceph_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
2800 struct inode *inode = d_inode(dentry);
2801 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
2804 if (ceph_snap(inode) != CEPH_NOSNAP)
2807 if (ceph_inode_is_shutdown(inode))
2810 err = fscrypt_prepare_setattr(dentry, attr);
2814 err = setattr_prepare(idmap, dentry, attr);
2818 if ((attr->ia_valid & ATTR_SIZE) &&
2819 attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2822 if ((attr->ia_valid & ATTR_SIZE) &&
2823 ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2826 err = __ceph_setattr(idmap, inode, attr, NULL);
2828 if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2829 err = posix_acl_chmod(idmap, dentry, attr->ia_mode);
2834 int ceph_try_to_choose_auth_mds(struct inode *inode, int mask)
2836 int issued = ceph_caps_issued(ceph_inode(inode));
2839 * If any 'x' caps is issued we can just choose the auth MDS
2840 * instead of the random replica MDSes. Because only when the
2841 * Locker is in LOCK_EXEC state will the loner client could
2842 * get the 'x' caps. And if we send the getattr requests to
2843 * any replica MDS it must auth pin and tries to rdlock from
2844 * the auth MDS, and then the auth MDS need to do the Locker
2845 * state transition to LOCK_SYNC. And after that the lock state
2848 * This cost much when doing the Locker state transition and
2849 * usually will need to revoke caps from clients.
2851 * And for the 'Xs' caps for getxattr we will also choose the
2852 * auth MDS, because the MDS side code is buggy due to setxattr
2853 * won't notify the replica MDSes when the values changed and
2854 * the replica MDS will return the old values. Though we will
2855 * fix it in MDS code, but this still makes sense for old ceph.
2857 if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL))
2858 || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR)))
2859 return USE_AUTH_MDS;
2865 * Verify that we have a lease on the given mask. If not,
2866 * do a getattr against an mds.
2868 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2869 int mask, bool force)
2871 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
2872 struct ceph_client *cl = fsc->client;
2873 struct ceph_mds_client *mdsc = fsc->mdsc;
2874 struct ceph_mds_request *req;
2878 if (ceph_snap(inode) == CEPH_SNAPDIR) {
2879 doutc(cl, "inode %p %llx.%llx SNAPDIR\n", inode,
2884 doutc(cl, "inode %p %llx.%llx mask %s mode 0%o\n", inode,
2885 ceph_vinop(inode), ceph_cap_string(mask), inode->i_mode);
2886 if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2889 mode = ceph_try_to_choose_auth_mds(inode, mask);
2890 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2892 return PTR_ERR(req);
2893 req->r_inode = inode;
2895 req->r_num_caps = 1;
2896 req->r_args.getattr.mask = cpu_to_le32(mask);
2897 req->r_locked_page = locked_page;
2898 err = ceph_mdsc_do_request(mdsc, NULL, req);
2899 if (locked_page && err == 0) {
2900 u64 inline_version = req->r_reply_info.targeti.inline_version;
2901 if (inline_version == 0) {
2902 /* the reply is supposed to contain inline data */
2904 } else if (inline_version == CEPH_INLINE_NONE ||
2905 inline_version == 1) {
2908 err = req->r_reply_info.targeti.inline_len;
2911 ceph_mdsc_put_request(req);
2912 doutc(cl, "result=%d\n", err);
2916 int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2919 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(inode->i_sb);
2920 struct ceph_client *cl = fsc->client;
2921 struct ceph_mds_client *mdsc = fsc->mdsc;
2922 struct ceph_mds_request *req;
2923 int mode = USE_AUTH_MDS;
2926 size_t xattr_value_len;
2928 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2934 req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR;
2935 req->r_path2 = kstrdup(name, GFP_NOFS);
2936 if (!req->r_path2) {
2942 req->r_inode = inode;
2943 err = ceph_mdsc_do_request(mdsc, NULL, req);
2947 xattr_value = req->r_reply_info.xattr_info.xattr_value;
2948 xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2950 doutc(cl, "xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
2952 err = (int)xattr_value_len;
2956 if (xattr_value_len > size) {
2961 memcpy(value, xattr_value, xattr_value_len);
2963 ceph_mdsc_put_request(req);
2965 doutc(cl, "result=%d\n", err);
2971 * Check inode permissions. We verify we have a valid value for
2972 * the AUTH cap, then call the generic handler.
2974 int ceph_permission(struct mnt_idmap *idmap, struct inode *inode,
2979 if (mask & MAY_NOT_BLOCK)
2982 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2985 err = generic_permission(idmap, inode, mask);
2989 /* Craft a mask of needed caps given a set of requested statx attrs. */
2990 static int statx_to_caps(u32 want, umode_t mode)
2994 if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME|STATX_CHANGE_COOKIE))
2995 mask |= CEPH_CAP_AUTH_SHARED;
2997 if (want & (STATX_NLINK|STATX_CTIME|STATX_CHANGE_COOKIE)) {
2999 * The link count for directories depends on inode->i_subdirs,
3000 * and that is only updated when Fs caps are held.
3003 mask |= CEPH_CAP_FILE_SHARED;
3005 mask |= CEPH_CAP_LINK_SHARED;
3008 if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|STATX_BLOCKS|STATX_CHANGE_COOKIE))
3009 mask |= CEPH_CAP_FILE_SHARED;
3011 if (want & (STATX_CTIME|STATX_CHANGE_COOKIE))
3012 mask |= CEPH_CAP_XATTR_SHARED;
3018 * Get all the attributes. If we have sufficient caps for the requested attrs,
3019 * then we can avoid talking to the MDS at all.
3021 int ceph_getattr(struct mnt_idmap *idmap, const struct path *path,
3022 struct kstat *stat, u32 request_mask, unsigned int flags)
3024 struct inode *inode = d_inode(path->dentry);
3025 struct super_block *sb = inode->i_sb;
3026 struct ceph_inode_info *ci = ceph_inode(inode);
3027 u32 valid_mask = STATX_BASIC_STATS;
3030 if (ceph_inode_is_shutdown(inode))
3033 /* Skip the getattr altogether if we're asked not to sync */
3034 if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) {
3035 err = ceph_do_getattr(inode,
3036 statx_to_caps(request_mask, inode->i_mode),
3037 flags & AT_STATX_FORCE_SYNC);
3042 generic_fillattr(idmap, request_mask, inode, stat);
3043 stat->ino = ceph_present_inode(inode);
3046 * btime on newly-allocated inodes is 0, so if this is still set to
3047 * that, then assume that it's not valid.
3049 if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
3050 stat->btime = ci->i_btime;
3051 valid_mask |= STATX_BTIME;
3054 if (request_mask & STATX_CHANGE_COOKIE) {
3055 stat->change_cookie = inode_peek_iversion_raw(inode);
3056 valid_mask |= STATX_CHANGE_COOKIE;
3059 if (ceph_snap(inode) == CEPH_NOSNAP)
3060 stat->dev = sb->s_dev;
3062 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
3064 if (S_ISDIR(inode->i_mode)) {
3065 if (ceph_test_mount_opt(ceph_sb_to_fs_client(sb), RBYTES)) {
3066 stat->size = ci->i_rbytes;
3067 } else if (ceph_snap(inode) == CEPH_SNAPDIR) {
3068 struct ceph_inode_info *pci;
3069 struct ceph_snap_realm *realm;
3070 struct inode *parent;
3072 parent = ceph_lookup_inode(sb, ceph_ino(inode));
3074 return PTR_ERR(parent);
3076 pci = ceph_inode(parent);
3077 spin_lock(&pci->i_ceph_lock);
3078 realm = pci->i_snap_realm;
3080 stat->size = realm->num_snaps;
3083 spin_unlock(&pci->i_ceph_lock);
3086 stat->size = ci->i_files + ci->i_subdirs;
3089 stat->blksize = 65536;
3091 * Some applications rely on the number of st_nlink
3092 * value on directories to be either 0 (if unlinked)
3093 * or 2 + number of subdirectories.
3095 if (stat->nlink == 1)
3096 /* '.' + '..' + subdirs */
3097 stat->nlink = 1 + 1 + ci->i_subdirs;
3100 stat->attributes |= STATX_ATTR_CHANGE_MONOTONIC;
3101 if (IS_ENCRYPTED(inode))
3102 stat->attributes |= STATX_ATTR_ENCRYPTED;
3103 stat->attributes_mask |= (STATX_ATTR_CHANGE_MONOTONIC |
3104 STATX_ATTR_ENCRYPTED);
3106 stat->result_mask = request_mask & valid_mask;
3110 void ceph_inode_shutdown(struct inode *inode)
3112 struct ceph_inode_info *ci = ceph_inode(inode);
3115 bool invalidate = false;
3117 spin_lock(&ci->i_ceph_lock);
3118 ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
3119 p = rb_first(&ci->i_caps);
3121 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
3124 iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
3126 spin_unlock(&ci->i_ceph_lock);
3129 ceph_queue_invalidate(inode);