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/namei.h>
10 #include <linux/writeback.h>
11 #include <linux/vmalloc.h>
12 #include <linux/posix_acl.h>
15 #include "mds_client.h"
17 #include <linux/ceph/decode.h>
20 * Ceph inode operations
22 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
23 * setattr, etc.), xattr helpers, and helpers for assimilating
24 * metadata returned by the MDS into our cache.
26 * Also define helpers for doing asynchronous writeback, invalidation,
27 * and truncation for the benefit of those who can't afford to block
28 * (typically because they are in the message handler path).
31 static const struct inode_operations ceph_symlink_iops;
33 static void ceph_invalidate_work(struct work_struct *work);
34 static void ceph_writeback_work(struct work_struct *work);
35 static void ceph_vmtruncate_work(struct work_struct *work);
38 * find or create an inode, given the ceph ino number
40 static int ceph_set_ino_cb(struct inode *inode, void *data)
42 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
43 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
47 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
50 ino_t t = ceph_vino_to_ino(vino);
52 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
54 return ERR_PTR(-ENOMEM);
55 if (inode->i_state & I_NEW) {
56 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
57 inode, ceph_vinop(inode), (u64)inode->i_ino);
58 unlock_new_inode(inode);
61 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
67 * get/constuct snapdir inode for a given directory
69 struct inode *ceph_get_snapdir(struct inode *parent)
71 struct ceph_vino vino = {
72 .ino = ceph_ino(parent),
75 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
76 struct ceph_inode_info *ci = ceph_inode(inode);
78 BUG_ON(!S_ISDIR(parent->i_mode));
81 inode->i_mode = parent->i_mode;
82 inode->i_uid = parent->i_uid;
83 inode->i_gid = parent->i_gid;
84 inode->i_op = &ceph_dir_iops;
85 inode->i_fop = &ceph_dir_fops;
86 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
91 const struct inode_operations ceph_file_iops = {
92 .permission = ceph_permission,
93 .setattr = ceph_setattr,
94 .getattr = ceph_getattr,
95 .setxattr = ceph_setxattr,
96 .getxattr = ceph_getxattr,
97 .listxattr = ceph_listxattr,
98 .removexattr = ceph_removexattr,
99 .get_acl = ceph_get_acl,
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 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
138 "frag %x\n", &ci->vfs_inode,
139 ceph_vinop(&ci->vfs_inode), f);
140 return ERR_PTR(-ENOMEM);
147 rb_link_node(&frag->node, parent, p);
148 rb_insert_color(&frag->node, &ci->i_fragtree);
150 dout("get_or_create_frag added %llx.%llx frag %x\n",
151 ceph_vinop(&ci->vfs_inode), f);
156 * find a specific frag @f
158 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
160 struct rb_node *n = ci->i_fragtree.rb_node;
163 struct ceph_inode_frag *frag =
164 rb_entry(n, struct ceph_inode_frag, node);
165 int c = ceph_frag_compare(f, frag->frag);
177 * Choose frag containing the given value @v. If @pfrag is
178 * specified, copy the frag delegation info to the caller if
181 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
182 struct ceph_inode_frag *pfrag,
185 u32 t = ceph_frag_make(0, 0);
186 struct ceph_inode_frag *frag;
193 mutex_lock(&ci->i_fragtree_mutex);
195 WARN_ON(!ceph_frag_contains_value(t, v));
196 frag = __ceph_find_frag(ci, t);
198 break; /* t is a leaf */
199 if (frag->split_by == 0) {
201 memcpy(pfrag, frag, sizeof(*pfrag));
208 nway = 1 << frag->split_by;
209 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
210 frag->split_by, nway);
211 for (i = 0; i < nway; i++) {
212 n = ceph_frag_make_child(t, frag->split_by, i);
213 if (ceph_frag_contains_value(n, v)) {
220 dout("choose_frag(%x) = %x\n", v, t);
222 mutex_unlock(&ci->i_fragtree_mutex);
227 * Process dirfrag (delegation) info from the mds. Include leaf
228 * fragment in tree ONLY if ndist > 0. Otherwise, only
229 * branches/splits are included in i_fragtree)
231 static int ceph_fill_dirfrag(struct inode *inode,
232 struct ceph_mds_reply_dirfrag *dirinfo)
234 struct ceph_inode_info *ci = ceph_inode(inode);
235 struct ceph_inode_frag *frag;
236 u32 id = le32_to_cpu(dirinfo->frag);
237 int mds = le32_to_cpu(dirinfo->auth);
238 int ndist = le32_to_cpu(dirinfo->ndist);
242 mutex_lock(&ci->i_fragtree_mutex);
244 /* no delegation info needed. */
245 frag = __ceph_find_frag(ci, id);
248 if (frag->split_by == 0) {
249 /* tree leaf, remove */
250 dout("fill_dirfrag removed %llx.%llx frag %x"
251 " (no ref)\n", ceph_vinop(inode), id);
252 rb_erase(&frag->node, &ci->i_fragtree);
255 /* tree branch, keep and clear */
256 dout("fill_dirfrag cleared %llx.%llx frag %x"
257 " referral\n", ceph_vinop(inode), id);
265 /* find/add this frag to store mds delegation info */
266 frag = __get_or_create_frag(ci, id);
268 /* this is not the end of the world; we can continue
269 with bad/inaccurate delegation info */
270 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
271 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
277 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
278 for (i = 0; i < frag->ndist; i++)
279 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
280 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
281 ceph_vinop(inode), frag->frag, frag->ndist);
284 mutex_unlock(&ci->i_fragtree_mutex);
290 * initialize a newly allocated inode.
292 struct inode *ceph_alloc_inode(struct super_block *sb)
294 struct ceph_inode_info *ci;
297 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
301 dout("alloc_inode %p\n", &ci->vfs_inode);
303 spin_lock_init(&ci->i_ceph_lock);
306 ci->i_time_warp_seq = 0;
307 ci->i_ceph_flags = 0;
308 atomic_set(&ci->i_release_count, 1);
309 atomic_set(&ci->i_complete_count, 0);
310 ci->i_symlink = NULL;
312 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
314 ci->i_fragtree = RB_ROOT;
315 mutex_init(&ci->i_fragtree_mutex);
317 ci->i_xattrs.blob = NULL;
318 ci->i_xattrs.prealloc_blob = NULL;
319 ci->i_xattrs.dirty = false;
320 ci->i_xattrs.index = RB_ROOT;
321 ci->i_xattrs.count = 0;
322 ci->i_xattrs.names_size = 0;
323 ci->i_xattrs.vals_size = 0;
324 ci->i_xattrs.version = 0;
325 ci->i_xattrs.index_version = 0;
327 ci->i_caps = RB_ROOT;
328 ci->i_auth_cap = NULL;
329 ci->i_dirty_caps = 0;
330 ci->i_flushing_caps = 0;
331 INIT_LIST_HEAD(&ci->i_dirty_item);
332 INIT_LIST_HEAD(&ci->i_flushing_item);
333 ci->i_cap_flush_seq = 0;
334 ci->i_cap_flush_last_tid = 0;
335 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
336 init_waitqueue_head(&ci->i_cap_wq);
337 ci->i_hold_caps_min = 0;
338 ci->i_hold_caps_max = 0;
339 INIT_LIST_HEAD(&ci->i_cap_delay_list);
340 INIT_LIST_HEAD(&ci->i_cap_snaps);
341 ci->i_head_snapc = NULL;
343 ci->i_cap_exporting_issued = 0;
345 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
346 ci->i_nr_by_mode[i] = 0;
348 mutex_init(&ci->i_truncate_mutex);
349 ci->i_truncate_seq = 0;
350 ci->i_truncate_size = 0;
351 ci->i_truncate_pending = 0;
354 ci->i_reported_size = 0;
355 ci->i_wanted_max_size = 0;
356 ci->i_requested_max_size = 0;
360 ci->i_rdcache_ref = 0;
363 ci->i_wrbuffer_ref = 0;
364 ci->i_wrbuffer_ref_head = 0;
365 ci->i_shared_gen = 0;
366 ci->i_rdcache_gen = 0;
367 ci->i_rdcache_revoking = 0;
369 INIT_LIST_HEAD(&ci->i_unsafe_writes);
370 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
371 spin_lock_init(&ci->i_unsafe_lock);
373 ci->i_snap_realm = NULL;
374 INIT_LIST_HEAD(&ci->i_snap_realm_item);
375 INIT_LIST_HEAD(&ci->i_snap_flush_item);
377 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
378 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
380 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
382 ceph_fscache_inode_init(ci);
384 return &ci->vfs_inode;
387 static void ceph_i_callback(struct rcu_head *head)
389 struct inode *inode = container_of(head, struct inode, i_rcu);
390 struct ceph_inode_info *ci = ceph_inode(inode);
392 kmem_cache_free(ceph_inode_cachep, ci);
395 void ceph_destroy_inode(struct inode *inode)
397 struct ceph_inode_info *ci = ceph_inode(inode);
398 struct ceph_inode_frag *frag;
401 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
403 ceph_fscache_unregister_inode_cookie(ci);
405 ceph_queue_caps_release(inode);
408 * we may still have a snap_realm reference if there are stray
409 * caps in i_cap_exporting_issued or i_snap_caps.
411 if (ci->i_snap_realm) {
412 struct ceph_mds_client *mdsc =
413 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
414 struct ceph_snap_realm *realm = ci->i_snap_realm;
416 dout(" dropping residual ref to snap realm %p\n", realm);
417 spin_lock(&realm->inodes_with_caps_lock);
418 list_del_init(&ci->i_snap_realm_item);
419 spin_unlock(&realm->inodes_with_caps_lock);
420 ceph_put_snap_realm(mdsc, realm);
423 kfree(ci->i_symlink);
424 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
425 frag = rb_entry(n, struct ceph_inode_frag, node);
426 rb_erase(n, &ci->i_fragtree);
430 __ceph_destroy_xattrs(ci);
431 if (ci->i_xattrs.blob)
432 ceph_buffer_put(ci->i_xattrs.blob);
433 if (ci->i_xattrs.prealloc_blob)
434 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
436 call_rcu(&inode->i_rcu, ceph_i_callback);
439 int ceph_drop_inode(struct inode *inode)
442 * Positve dentry and corresponding inode are always accompanied
443 * in MDS reply. So no need to keep inode in the cache after
444 * dropping all its aliases.
450 * Helpers to fill in size, ctime, mtime, and atime. We have to be
451 * careful because either the client or MDS may have more up to date
452 * info, depending on which capabilities are held, and whether
453 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
454 * and size are monotonically increasing, except when utimes() or
455 * truncate() increments the corresponding _seq values.)
457 int ceph_fill_file_size(struct inode *inode, int issued,
458 u32 truncate_seq, u64 truncate_size, u64 size)
460 struct ceph_inode_info *ci = ceph_inode(inode);
463 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
464 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
465 dout("size %lld -> %llu\n", inode->i_size, size);
466 inode->i_size = size;
467 inode->i_blocks = (size + (1<<9) - 1) >> 9;
468 ci->i_reported_size = size;
469 if (truncate_seq != ci->i_truncate_seq) {
470 dout("truncate_seq %u -> %u\n",
471 ci->i_truncate_seq, truncate_seq);
472 ci->i_truncate_seq = truncate_seq;
474 /* the MDS should have revoked these caps */
475 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
478 CEPH_CAP_FILE_LAZYIO));
480 * If we hold relevant caps, or in the case where we're
481 * not the only client referencing this file and we
482 * don't hold those caps, then we need to check whether
483 * the file is either opened or mmaped
485 if ((issued & (CEPH_CAP_FILE_CACHE|
486 CEPH_CAP_FILE_BUFFER)) ||
487 mapping_mapped(inode->i_mapping) ||
488 __ceph_caps_file_wanted(ci)) {
489 ci->i_truncate_pending++;
494 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
495 ci->i_truncate_size != truncate_size) {
496 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
498 ci->i_truncate_size = truncate_size;
502 ceph_fscache_invalidate(inode);
507 void ceph_fill_file_time(struct inode *inode, int issued,
508 u64 time_warp_seq, struct timespec *ctime,
509 struct timespec *mtime, struct timespec *atime)
511 struct ceph_inode_info *ci = ceph_inode(inode);
514 if (issued & (CEPH_CAP_FILE_EXCL|
516 CEPH_CAP_FILE_BUFFER|
518 CEPH_CAP_XATTR_EXCL)) {
519 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
520 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
521 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
522 ctime->tv_sec, ctime->tv_nsec);
523 inode->i_ctime = *ctime;
525 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
526 /* the MDS did a utimes() */
527 dout("mtime %ld.%09ld -> %ld.%09ld "
529 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
530 mtime->tv_sec, mtime->tv_nsec,
531 ci->i_time_warp_seq, (int)time_warp_seq);
533 inode->i_mtime = *mtime;
534 inode->i_atime = *atime;
535 ci->i_time_warp_seq = time_warp_seq;
536 } else if (time_warp_seq == ci->i_time_warp_seq) {
537 /* nobody did utimes(); take the max */
538 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
539 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
540 inode->i_mtime.tv_sec,
541 inode->i_mtime.tv_nsec,
542 mtime->tv_sec, mtime->tv_nsec);
543 inode->i_mtime = *mtime;
545 if (timespec_compare(atime, &inode->i_atime) > 0) {
546 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
547 inode->i_atime.tv_sec,
548 inode->i_atime.tv_nsec,
549 atime->tv_sec, atime->tv_nsec);
550 inode->i_atime = *atime;
552 } else if (issued & CEPH_CAP_FILE_EXCL) {
553 /* we did a utimes(); ignore mds values */
558 /* we have no write|excl caps; whatever the MDS says is true */
559 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
560 inode->i_ctime = *ctime;
561 inode->i_mtime = *mtime;
562 inode->i_atime = *atime;
563 ci->i_time_warp_seq = time_warp_seq;
568 if (warn) /* time_warp_seq shouldn't go backwards */
569 dout("%p mds time_warp_seq %llu < %u\n",
570 inode, time_warp_seq, ci->i_time_warp_seq);
574 * Populate an inode based on info from mds. May be called on new or
577 static int fill_inode(struct inode *inode,
578 struct ceph_mds_reply_info_in *iinfo,
579 struct ceph_mds_reply_dirfrag *dirinfo,
580 struct ceph_mds_session *session,
581 unsigned long ttl_from, int cap_fmode,
582 struct ceph_cap_reservation *caps_reservation)
584 struct ceph_mds_reply_inode *info = iinfo->in;
585 struct ceph_inode_info *ci = ceph_inode(inode);
587 int issued = 0, implemented;
588 struct timespec mtime, atime, ctime;
590 struct ceph_inode_frag *frag;
591 struct rb_node *rb_node;
592 struct ceph_buffer *xattr_blob = NULL;
596 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
597 inode, ceph_vinop(inode), le64_to_cpu(info->version),
601 * prealloc xattr data, if it looks like we'll need it. only
602 * if len > 4 (meaning there are actually xattrs; the first 4
603 * bytes are the xattr count).
605 if (iinfo->xattr_len > 4) {
606 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
608 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
612 spin_lock(&ci->i_ceph_lock);
615 * provided version will be odd if inode value is projected,
616 * even if stable. skip the update if we have newer stable
617 * info (ours>=theirs, e.g. due to racing mds replies), unless
618 * we are getting projected (unstable) info (in which case the
619 * version is odd, and we want ours>theirs).
625 if (le64_to_cpu(info->version) > 0 &&
626 (ci->i_version & ~1) >= le64_to_cpu(info->version))
629 issued = __ceph_caps_issued(ci, &implemented);
630 issued |= implemented | __ceph_caps_dirty(ci);
633 ci->i_version = le64_to_cpu(info->version);
635 inode->i_rdev = le32_to_cpu(info->rdev);
637 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
638 inode->i_mode = le32_to_cpu(info->mode);
639 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
640 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
641 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
642 from_kuid(&init_user_ns, inode->i_uid),
643 from_kgid(&init_user_ns, inode->i_gid));
646 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
647 set_nlink(inode, le32_to_cpu(info->nlink));
649 /* be careful with mtime, atime, size */
650 ceph_decode_timespec(&atime, &info->atime);
651 ceph_decode_timespec(&mtime, &info->mtime);
652 ceph_decode_timespec(&ctime, &info->ctime);
653 queue_trunc = ceph_fill_file_size(inode, issued,
654 le32_to_cpu(info->truncate_seq),
655 le64_to_cpu(info->truncate_size),
656 le64_to_cpu(info->size));
657 ceph_fill_file_time(inode, issued,
658 le32_to_cpu(info->time_warp_seq),
659 &ctime, &mtime, &atime);
661 /* only update max_size on auth cap */
662 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
663 ci->i_max_size != le64_to_cpu(info->max_size)) {
664 dout("max_size %lld -> %llu\n", ci->i_max_size,
665 le64_to_cpu(info->max_size));
666 ci->i_max_size = le64_to_cpu(info->max_size);
669 ci->i_layout = info->layout;
670 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
673 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
674 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
675 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
676 if (ci->i_xattrs.blob)
677 ceph_buffer_put(ci->i_xattrs.blob);
678 ci->i_xattrs.blob = xattr_blob;
680 memcpy(ci->i_xattrs.blob->vec.iov_base,
681 iinfo->xattr_data, iinfo->xattr_len);
682 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
683 ceph_forget_all_cached_acls(inode);
687 inode->i_mapping->a_ops = &ceph_aops;
688 inode->i_mapping->backing_dev_info =
689 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
691 switch (inode->i_mode & S_IFMT) {
696 init_special_inode(inode, inode->i_mode, inode->i_rdev);
697 inode->i_op = &ceph_file_iops;
700 inode->i_op = &ceph_file_iops;
701 inode->i_fop = &ceph_file_fops;
704 inode->i_op = &ceph_symlink_iops;
705 if (!ci->i_symlink) {
706 u32 symlen = iinfo->symlink_len;
709 spin_unlock(&ci->i_ceph_lock);
712 if (WARN_ON(symlen != inode->i_size))
716 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
720 spin_lock(&ci->i_ceph_lock);
724 kfree(sym); /* lost a race */
728 inode->i_op = &ceph_dir_iops;
729 inode->i_fop = &ceph_dir_fops;
731 ci->i_dir_layout = iinfo->dir_layout;
733 ci->i_files = le64_to_cpu(info->files);
734 ci->i_subdirs = le64_to_cpu(info->subdirs);
735 ci->i_rbytes = le64_to_cpu(info->rbytes);
736 ci->i_rfiles = le64_to_cpu(info->rfiles);
737 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
738 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
741 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
742 ceph_vinop(inode), inode->i_mode);
745 /* set dir completion flag? */
746 if (S_ISDIR(inode->i_mode) &&
747 ci->i_files == 0 && ci->i_subdirs == 0 &&
748 ceph_snap(inode) == CEPH_NOSNAP &&
749 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
750 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
751 !__ceph_dir_is_complete(ci)) {
752 dout(" marking %p complete (empty)\n", inode);
753 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count));
754 ci->i_max_offset = 2;
757 spin_unlock(&ci->i_ceph_lock);
759 /* queue truncate if we saw i_size decrease */
761 ceph_queue_vmtruncate(inode);
763 /* populate frag tree */
764 /* FIXME: move me up, if/when version reflects fragtree changes */
765 nsplits = le32_to_cpu(info->fragtree.nsplits);
766 mutex_lock(&ci->i_fragtree_mutex);
767 rb_node = rb_first(&ci->i_fragtree);
768 for (i = 0; i < nsplits; i++) {
769 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
772 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
773 if (ceph_frag_compare(frag->frag, id) >= 0) {
774 if (frag->frag != id)
777 rb_node = rb_next(rb_node);
780 rb_node = rb_next(rb_node);
781 rb_erase(&frag->node, &ci->i_fragtree);
786 frag = __get_or_create_frag(ci, id);
790 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
791 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
794 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
795 rb_node = rb_next(rb_node);
796 rb_erase(&frag->node, &ci->i_fragtree);
799 mutex_unlock(&ci->i_fragtree_mutex);
801 /* were we issued a capability? */
802 if (info->cap.caps) {
803 if (ceph_snap(inode) == CEPH_NOSNAP) {
804 ceph_add_cap(inode, session,
805 le64_to_cpu(info->cap.cap_id),
807 le32_to_cpu(info->cap.caps),
808 le32_to_cpu(info->cap.wanted),
809 le32_to_cpu(info->cap.seq),
810 le32_to_cpu(info->cap.mseq),
811 le64_to_cpu(info->cap.realm),
815 spin_lock(&ci->i_ceph_lock);
816 dout(" %p got snap_caps %s\n", inode,
817 ceph_cap_string(le32_to_cpu(info->cap.caps)));
818 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
820 __ceph_get_fmode(ci, cap_fmode);
821 spin_unlock(&ci->i_ceph_lock);
823 } else if (cap_fmode >= 0) {
824 pr_warning("mds issued no caps on %llx.%llx\n",
826 __ceph_get_fmode(ci, cap_fmode);
829 /* update delegation info? */
831 ceph_fill_dirfrag(inode, dirinfo);
837 ceph_buffer_put(xattr_blob);
842 * caller should hold session s_mutex.
844 static void update_dentry_lease(struct dentry *dentry,
845 struct ceph_mds_reply_lease *lease,
846 struct ceph_mds_session *session,
847 unsigned long from_time)
849 struct ceph_dentry_info *di = ceph_dentry(dentry);
850 long unsigned duration = le32_to_cpu(lease->duration_ms);
851 long unsigned ttl = from_time + (duration * HZ) / 1000;
852 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
855 /* only track leases on regular dentries */
856 if (dentry->d_op != &ceph_dentry_ops)
859 spin_lock(&dentry->d_lock);
860 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
861 dentry, duration, ttl);
863 /* make lease_rdcache_gen match directory */
864 dir = dentry->d_parent->d_inode;
865 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
870 if (di->lease_gen == session->s_cap_gen &&
871 time_before(ttl, dentry->d_time))
872 goto out_unlock; /* we already have a newer lease. */
874 if (di->lease_session && di->lease_session != session)
877 ceph_dentry_lru_touch(dentry);
879 if (!di->lease_session)
880 di->lease_session = ceph_get_mds_session(session);
881 di->lease_gen = session->s_cap_gen;
882 di->lease_seq = le32_to_cpu(lease->seq);
883 di->lease_renew_after = half_ttl;
884 di->lease_renew_from = 0;
885 dentry->d_time = ttl;
887 spin_unlock(&dentry->d_lock);
892 * Set dentry's directory position based on the current dir's max, and
893 * order it in d_subdirs, so that dcache_readdir behaves.
895 * Always called under directory's i_mutex.
897 static void ceph_set_dentry_offset(struct dentry *dn)
899 struct dentry *dir = dn->d_parent;
900 struct inode *inode = dir->d_inode;
901 struct ceph_inode_info *ci;
902 struct ceph_dentry_info *di;
906 ci = ceph_inode(inode);
907 di = ceph_dentry(dn);
909 spin_lock(&ci->i_ceph_lock);
910 if (!__ceph_dir_is_complete(ci)) {
911 spin_unlock(&ci->i_ceph_lock);
914 di->offset = ceph_inode(inode)->i_max_offset++;
915 spin_unlock(&ci->i_ceph_lock);
917 spin_lock(&dir->d_lock);
918 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
919 list_move(&dn->d_u.d_child, &dir->d_subdirs);
920 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
921 dn->d_u.d_child.prev, dn->d_u.d_child.next);
922 spin_unlock(&dn->d_lock);
923 spin_unlock(&dir->d_lock);
927 * splice a dentry to an inode.
928 * caller must hold directory i_mutex for this to be safe.
930 * we will only rehash the resulting dentry if @prehash is
931 * true; @prehash will be set to false (for the benefit of
932 * the caller) if we fail.
934 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
935 bool *prehash, bool set_offset)
937 struct dentry *realdn;
941 /* dn must be unhashed */
944 realdn = d_materialise_unique(dn, in);
945 if (IS_ERR(realdn)) {
946 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
947 PTR_ERR(realdn), dn, in, ceph_vinop(in));
949 *prehash = false; /* don't rehash on error */
950 dn = realdn; /* note realdn contains the error */
953 dout("dn %p (%d) spliced with %p (%d) "
954 "inode %p ino %llx.%llx\n",
956 realdn, d_count(realdn),
957 realdn->d_inode, ceph_vinop(realdn->d_inode));
961 BUG_ON(!ceph_dentry(dn));
962 dout("dn %p attached to %p ino %llx.%llx\n",
963 dn, dn->d_inode, ceph_vinop(dn->d_inode));
965 if ((!prehash || *prehash) && d_unhashed(dn))
968 ceph_set_dentry_offset(dn);
974 * Incorporate results into the local cache. This is either just
975 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
978 * A reply may contain
979 * a directory inode along with a dentry.
980 * and/or a target inode
982 * Called with snap_rwsem (read).
984 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
985 struct ceph_mds_session *session)
987 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
988 struct inode *in = NULL;
989 struct ceph_mds_reply_inode *ininfo;
990 struct ceph_vino vino;
991 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
994 dout("fill_trace %p is_dentry %d is_target %d\n", req,
995 rinfo->head->is_dentry, rinfo->head->is_target);
1001 * If we resend completed ops to a recovering mds, we get no
1002 * trace. Since that is very rare, pretend this is the case
1003 * to ensure the 'no trace' handlers in the callers behave.
1005 * Fill in inodes unconditionally to avoid breaking cap
1008 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1009 pr_info("fill_trace faking empty trace on %lld %s\n",
1010 req->r_tid, ceph_mds_op_name(rinfo->head->op));
1011 if (rinfo->head->is_dentry) {
1012 rinfo->head->is_dentry = 0;
1013 err = fill_inode(req->r_locked_dir,
1014 &rinfo->diri, rinfo->dirfrag,
1015 session, req->r_request_started, -1);
1017 if (rinfo->head->is_target) {
1018 rinfo->head->is_target = 0;
1019 ininfo = rinfo->targeti.in;
1020 vino.ino = le64_to_cpu(ininfo->ino);
1021 vino.snap = le64_to_cpu(ininfo->snapid);
1022 in = ceph_get_inode(sb, vino);
1023 err = fill_inode(in, &rinfo->targeti, NULL,
1024 session, req->r_request_started,
1031 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1032 dout("fill_trace reply is empty!\n");
1033 if (rinfo->head->result == 0 && req->r_locked_dir)
1034 ceph_invalidate_dir_request(req);
1038 if (rinfo->head->is_dentry) {
1039 struct inode *dir = req->r_locked_dir;
1042 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
1043 session, req->r_request_started, -1,
1044 &req->r_caps_reservation);
1052 if (rinfo->head->is_target) {
1053 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1054 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1056 in = ceph_get_inode(sb, vino);
1061 req->r_target_inode = in;
1063 err = fill_inode(in, &rinfo->targeti, NULL,
1064 session, req->r_request_started,
1065 (le32_to_cpu(rinfo->head->result) == 0) ?
1067 &req->r_caps_reservation);
1069 pr_err("fill_inode badness %p %llx.%llx\n",
1070 in, ceph_vinop(in));
1076 * ignore null lease/binding on snapdir ENOENT, or else we
1077 * will have trouble splicing in the virtual snapdir later
1079 if (rinfo->head->is_dentry && !req->r_aborted &&
1080 req->r_locked_dir &&
1081 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1082 fsc->mount_options->snapdir_name,
1083 req->r_dentry->d_name.len))) {
1085 * lookup link rename : null -> possibly existing inode
1086 * mknod symlink mkdir : null -> new inode
1087 * unlink : linked -> null
1089 struct inode *dir = req->r_locked_dir;
1090 struct dentry *dn = req->r_dentry;
1091 bool have_dir_cap, have_lease;
1095 BUG_ON(dn->d_parent->d_inode != dir);
1096 BUG_ON(ceph_ino(dir) !=
1097 le64_to_cpu(rinfo->diri.in->ino));
1098 BUG_ON(ceph_snap(dir) !=
1099 le64_to_cpu(rinfo->diri.in->snapid));
1101 /* do we have a lease on the whole dir? */
1103 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1104 CEPH_CAP_FILE_SHARED);
1106 /* do we have a dn lease? */
1107 have_lease = have_dir_cap ||
1108 le32_to_cpu(rinfo->dlease->duration_ms);
1110 dout("fill_trace no dentry lease or dir cap\n");
1113 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1114 dout(" src %p '%.*s' dst %p '%.*s'\n",
1116 req->r_old_dentry->d_name.len,
1117 req->r_old_dentry->d_name.name,
1118 dn, dn->d_name.len, dn->d_name.name);
1119 dout("fill_trace doing d_move %p -> %p\n",
1120 req->r_old_dentry, dn);
1122 d_move(req->r_old_dentry, dn);
1123 dout(" src %p '%.*s' dst %p '%.*s'\n",
1125 req->r_old_dentry->d_name.len,
1126 req->r_old_dentry->d_name.name,
1127 dn, dn->d_name.len, dn->d_name.name);
1129 /* ensure target dentry is invalidated, despite
1130 rehashing bug in vfs_rename_dir */
1131 ceph_invalidate_dentry_lease(dn);
1134 * d_move() puts the renamed dentry at the end of
1135 * d_subdirs. We need to assign it an appropriate
1136 * directory offset so we can behave when dir is
1139 ceph_set_dentry_offset(req->r_old_dentry);
1140 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1141 ceph_dentry(req->r_old_dentry)->offset);
1143 dn = req->r_old_dentry; /* use old_dentry */
1147 if (!rinfo->head->is_target) {
1148 dout("fill_trace null dentry\n");
1150 dout("d_delete %p\n", dn);
1153 dout("d_instantiate %p NULL\n", dn);
1154 d_instantiate(dn, NULL);
1155 if (have_lease && d_unhashed(dn))
1157 update_dentry_lease(dn, rinfo->dlease,
1159 req->r_request_started);
1164 /* attach proper inode */
1167 dn = splice_dentry(dn, in, &have_lease, true);
1172 req->r_dentry = dn; /* may have spliced */
1173 } else if (dn->d_inode && dn->d_inode != in) {
1174 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1175 dn, dn->d_inode, ceph_vinop(dn->d_inode),
1181 update_dentry_lease(dn, rinfo->dlease, session,
1182 req->r_request_started);
1183 dout(" final dn %p\n", dn);
1184 } else if (!req->r_aborted &&
1185 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1186 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1187 struct dentry *dn = req->r_dentry;
1189 /* fill out a snapdir LOOKUPSNAP dentry */
1191 BUG_ON(!req->r_locked_dir);
1192 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1193 ininfo = rinfo->targeti.in;
1194 vino.ino = le64_to_cpu(ininfo->ino);
1195 vino.snap = le64_to_cpu(ininfo->snapid);
1196 dout(" linking snapped dir %p to dn %p\n", in, dn);
1198 dn = splice_dentry(dn, in, NULL, true);
1203 req->r_dentry = dn; /* may have spliced */
1206 dout("fill_trace done err=%d\n", err);
1211 * Prepopulate our cache with readdir results, leases, etc.
1213 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1214 struct ceph_mds_session *session)
1216 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1219 for (i = 0; i < rinfo->dir_nr; i++) {
1220 struct ceph_vino vino;
1224 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1225 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1227 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1230 dout("new_inode badness got %d\n", err);
1233 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1234 req->r_request_started, -1,
1235 &req->r_caps_reservation);
1237 pr_err("fill_inode badness on %p got %d\n", in, rc);
1246 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1247 struct ceph_mds_session *session)
1249 struct dentry *parent = req->r_dentry;
1250 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1254 int err = 0, ret, i;
1255 struct inode *snapdir = NULL;
1256 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1257 struct ceph_dentry_info *di;
1258 u64 r_readdir_offset = req->r_readdir_offset;
1259 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1261 if (rinfo->dir_dir &&
1262 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1263 dout("readdir_prepopulate got new frag %x -> %x\n",
1264 frag, le32_to_cpu(rinfo->dir_dir->frag));
1265 frag = le32_to_cpu(rinfo->dir_dir->frag);
1266 if (ceph_frag_is_leftmost(frag))
1267 r_readdir_offset = 2;
1269 r_readdir_offset = 0;
1273 return readdir_prepopulate_inodes_only(req, session);
1275 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1276 snapdir = ceph_get_snapdir(parent->d_inode);
1277 parent = d_find_alias(snapdir);
1278 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1279 rinfo->dir_nr, parent);
1281 dout("readdir_prepopulate %d items under dn %p\n",
1282 rinfo->dir_nr, parent);
1284 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1287 /* FIXME: release caps/leases if error occurs */
1288 for (i = 0; i < rinfo->dir_nr; i++) {
1289 struct ceph_vino vino;
1291 dname.name = rinfo->dir_dname[i];
1292 dname.len = rinfo->dir_dname_len[i];
1293 dname.hash = full_name_hash(dname.name, dname.len);
1295 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1296 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1299 dn = d_lookup(parent, &dname);
1300 dout("d_lookup on parent=%p name=%.*s got %p\n",
1301 parent, dname.len, dname.name, dn);
1304 dn = d_alloc(parent, &dname);
1305 dout("d_alloc %p '%.*s' = %p\n", parent,
1306 dname.len, dname.name, dn);
1308 dout("d_alloc badness\n");
1312 ret = ceph_init_dentry(dn);
1318 } else if (dn->d_inode &&
1319 (ceph_ino(dn->d_inode) != vino.ino ||
1320 ceph_snap(dn->d_inode) != vino.snap)) {
1321 dout(" dn %p points to wrong inode %p\n",
1327 /* reorder parent's d_subdirs */
1328 spin_lock(&parent->d_lock);
1329 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1330 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1331 spin_unlock(&dn->d_lock);
1332 spin_unlock(&parent->d_lock);
1339 in = ceph_get_inode(parent->d_sb, vino);
1341 dout("new_inode badness\n");
1349 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1350 req->r_request_started, -1,
1351 &req->r_caps_reservation) < 0) {
1352 pr_err("fill_inode badness on %p\n", in);
1360 dn = splice_dentry(dn, in, NULL, false);
1369 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1371 update_dentry_lease(dn, rinfo->dir_dlease[i],
1373 req->r_request_started);
1379 req->r_did_prepopulate = true;
1386 dout("readdir_prepopulate done\n");
1390 int ceph_inode_set_size(struct inode *inode, loff_t size)
1392 struct ceph_inode_info *ci = ceph_inode(inode);
1395 spin_lock(&ci->i_ceph_lock);
1396 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1397 inode->i_size = size;
1398 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1400 /* tell the MDS if we are approaching max_size */
1401 if ((size << 1) >= ci->i_max_size &&
1402 (ci->i_reported_size << 1) < ci->i_max_size)
1405 spin_unlock(&ci->i_ceph_lock);
1410 * Write back inode data in a worker thread. (This can't be done
1411 * in the message handler context.)
1413 void ceph_queue_writeback(struct inode *inode)
1416 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1417 &ceph_inode(inode)->i_wb_work)) {
1418 dout("ceph_queue_writeback %p\n", inode);
1420 dout("ceph_queue_writeback %p failed\n", inode);
1425 static void ceph_writeback_work(struct work_struct *work)
1427 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1429 struct inode *inode = &ci->vfs_inode;
1431 dout("writeback %p\n", inode);
1432 filemap_fdatawrite(&inode->i_data);
1437 * queue an async invalidation
1439 void ceph_queue_invalidate(struct inode *inode)
1442 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1443 &ceph_inode(inode)->i_pg_inv_work)) {
1444 dout("ceph_queue_invalidate %p\n", inode);
1446 dout("ceph_queue_invalidate %p failed\n", inode);
1452 * Invalidate inode pages in a worker thread. (This can't be done
1453 * in the message handler context.)
1455 static void ceph_invalidate_work(struct work_struct *work)
1457 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1459 struct inode *inode = &ci->vfs_inode;
1463 mutex_lock(&ci->i_truncate_mutex);
1464 spin_lock(&ci->i_ceph_lock);
1465 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1466 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1467 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1468 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1470 spin_unlock(&ci->i_ceph_lock);
1471 mutex_unlock(&ci->i_truncate_mutex);
1474 orig_gen = ci->i_rdcache_gen;
1475 spin_unlock(&ci->i_ceph_lock);
1477 truncate_inode_pages(inode->i_mapping, 0);
1479 spin_lock(&ci->i_ceph_lock);
1480 if (orig_gen == ci->i_rdcache_gen &&
1481 orig_gen == ci->i_rdcache_revoking) {
1482 dout("invalidate_pages %p gen %d successful\n", inode,
1484 ci->i_rdcache_revoking--;
1487 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1488 inode, orig_gen, ci->i_rdcache_gen,
1489 ci->i_rdcache_revoking);
1490 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1493 spin_unlock(&ci->i_ceph_lock);
1494 mutex_unlock(&ci->i_truncate_mutex);
1497 ceph_check_caps(ci, 0, NULL);
1503 * called by trunc_wq;
1505 * We also truncate in a separate thread as well.
1507 static void ceph_vmtruncate_work(struct work_struct *work)
1509 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1511 struct inode *inode = &ci->vfs_inode;
1513 dout("vmtruncate_work %p\n", inode);
1514 __ceph_do_pending_vmtruncate(inode);
1519 * Queue an async vmtruncate. If we fail to queue work, we will handle
1520 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1522 void ceph_queue_vmtruncate(struct inode *inode)
1524 struct ceph_inode_info *ci = ceph_inode(inode);
1528 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1529 &ci->i_vmtruncate_work)) {
1530 dout("ceph_queue_vmtruncate %p\n", inode);
1532 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1533 inode, ci->i_truncate_pending);
1539 * Make sure any pending truncation is applied before doing anything
1540 * that may depend on it.
1542 void __ceph_do_pending_vmtruncate(struct inode *inode)
1544 struct ceph_inode_info *ci = ceph_inode(inode);
1546 int wrbuffer_refs, finish = 0;
1548 mutex_lock(&ci->i_truncate_mutex);
1550 spin_lock(&ci->i_ceph_lock);
1551 if (ci->i_truncate_pending == 0) {
1552 dout("__do_pending_vmtruncate %p none pending\n", inode);
1553 spin_unlock(&ci->i_ceph_lock);
1554 mutex_unlock(&ci->i_truncate_mutex);
1559 * make sure any dirty snapped pages are flushed before we
1560 * possibly truncate them.. so write AND block!
1562 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1563 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1565 spin_unlock(&ci->i_ceph_lock);
1566 filemap_write_and_wait_range(&inode->i_data, 0,
1567 inode->i_sb->s_maxbytes);
1571 /* there should be no reader or writer */
1572 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1574 to = ci->i_truncate_size;
1575 wrbuffer_refs = ci->i_wrbuffer_ref;
1576 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1577 ci->i_truncate_pending, to);
1578 spin_unlock(&ci->i_ceph_lock);
1580 truncate_inode_pages(inode->i_mapping, to);
1582 spin_lock(&ci->i_ceph_lock);
1583 if (to == ci->i_truncate_size) {
1584 ci->i_truncate_pending = 0;
1587 spin_unlock(&ci->i_ceph_lock);
1591 mutex_unlock(&ci->i_truncate_mutex);
1593 if (wrbuffer_refs == 0)
1594 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1596 wake_up_all(&ci->i_cap_wq);
1602 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1604 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1605 nd_set_link(nd, ci->i_symlink);
1609 static const struct inode_operations ceph_symlink_iops = {
1610 .readlink = generic_readlink,
1611 .follow_link = ceph_sym_follow_link,
1612 .setattr = ceph_setattr,
1613 .getattr = ceph_getattr,
1614 .setxattr = ceph_setxattr,
1615 .getxattr = ceph_getxattr,
1616 .listxattr = ceph_listxattr,
1617 .removexattr = ceph_removexattr,
1618 .get_acl = ceph_get_acl,
1624 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1626 struct inode *inode = dentry->d_inode;
1627 struct ceph_inode_info *ci = ceph_inode(inode);
1628 struct inode *parent_inode;
1629 const unsigned int ia_valid = attr->ia_valid;
1630 struct ceph_mds_request *req;
1631 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1633 int release = 0, dirtied = 0;
1636 int inode_dirty_flags = 0;
1638 if (ceph_snap(inode) != CEPH_NOSNAP)
1641 err = inode_change_ok(inode, attr);
1645 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1648 return PTR_ERR(req);
1650 spin_lock(&ci->i_ceph_lock);
1651 issued = __ceph_caps_issued(ci, NULL);
1652 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1654 if (ia_valid & ATTR_UID) {
1655 dout("setattr %p uid %d -> %d\n", inode,
1656 from_kuid(&init_user_ns, inode->i_uid),
1657 from_kuid(&init_user_ns, attr->ia_uid));
1658 if (issued & CEPH_CAP_AUTH_EXCL) {
1659 inode->i_uid = attr->ia_uid;
1660 dirtied |= CEPH_CAP_AUTH_EXCL;
1661 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1662 !uid_eq(attr->ia_uid, inode->i_uid)) {
1663 req->r_args.setattr.uid = cpu_to_le32(
1664 from_kuid(&init_user_ns, attr->ia_uid));
1665 mask |= CEPH_SETATTR_UID;
1666 release |= CEPH_CAP_AUTH_SHARED;
1669 if (ia_valid & ATTR_GID) {
1670 dout("setattr %p gid %d -> %d\n", inode,
1671 from_kgid(&init_user_ns, inode->i_gid),
1672 from_kgid(&init_user_ns, attr->ia_gid));
1673 if (issued & CEPH_CAP_AUTH_EXCL) {
1674 inode->i_gid = attr->ia_gid;
1675 dirtied |= CEPH_CAP_AUTH_EXCL;
1676 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1677 !gid_eq(attr->ia_gid, inode->i_gid)) {
1678 req->r_args.setattr.gid = cpu_to_le32(
1679 from_kgid(&init_user_ns, attr->ia_gid));
1680 mask |= CEPH_SETATTR_GID;
1681 release |= CEPH_CAP_AUTH_SHARED;
1684 if (ia_valid & ATTR_MODE) {
1685 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1687 if (issued & CEPH_CAP_AUTH_EXCL) {
1688 inode->i_mode = attr->ia_mode;
1689 dirtied |= CEPH_CAP_AUTH_EXCL;
1690 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1691 attr->ia_mode != inode->i_mode) {
1692 inode->i_mode = attr->ia_mode;
1693 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1694 mask |= CEPH_SETATTR_MODE;
1695 release |= CEPH_CAP_AUTH_SHARED;
1699 if (ia_valid & ATTR_ATIME) {
1700 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1701 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1702 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1703 if (issued & CEPH_CAP_FILE_EXCL) {
1704 ci->i_time_warp_seq++;
1705 inode->i_atime = attr->ia_atime;
1706 dirtied |= CEPH_CAP_FILE_EXCL;
1707 } else if ((issued & CEPH_CAP_FILE_WR) &&
1708 timespec_compare(&inode->i_atime,
1709 &attr->ia_atime) < 0) {
1710 inode->i_atime = attr->ia_atime;
1711 dirtied |= CEPH_CAP_FILE_WR;
1712 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1713 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1714 ceph_encode_timespec(&req->r_args.setattr.atime,
1716 mask |= CEPH_SETATTR_ATIME;
1717 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1721 if (ia_valid & ATTR_MTIME) {
1722 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1723 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1724 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1725 if (issued & CEPH_CAP_FILE_EXCL) {
1726 ci->i_time_warp_seq++;
1727 inode->i_mtime = attr->ia_mtime;
1728 dirtied |= CEPH_CAP_FILE_EXCL;
1729 } else if ((issued & CEPH_CAP_FILE_WR) &&
1730 timespec_compare(&inode->i_mtime,
1731 &attr->ia_mtime) < 0) {
1732 inode->i_mtime = attr->ia_mtime;
1733 dirtied |= CEPH_CAP_FILE_WR;
1734 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1735 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1736 ceph_encode_timespec(&req->r_args.setattr.mtime,
1738 mask |= CEPH_SETATTR_MTIME;
1739 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1743 if (ia_valid & ATTR_SIZE) {
1744 dout("setattr %p size %lld -> %lld\n", inode,
1745 inode->i_size, attr->ia_size);
1746 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1750 if ((issued & CEPH_CAP_FILE_EXCL) &&
1751 attr->ia_size > inode->i_size) {
1752 inode->i_size = attr->ia_size;
1754 (attr->ia_size + (1 << 9) - 1) >> 9;
1755 inode->i_ctime = attr->ia_ctime;
1756 ci->i_reported_size = attr->ia_size;
1757 dirtied |= CEPH_CAP_FILE_EXCL;
1758 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1759 attr->ia_size != inode->i_size) {
1760 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1761 req->r_args.setattr.old_size =
1762 cpu_to_le64(inode->i_size);
1763 mask |= CEPH_SETATTR_SIZE;
1764 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1769 /* these do nothing */
1770 if (ia_valid & ATTR_CTIME) {
1771 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1772 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1773 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1774 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1775 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1776 only ? "ctime only" : "ignored");
1777 inode->i_ctime = attr->ia_ctime;
1780 * if kernel wants to dirty ctime but nothing else,
1781 * we need to choose a cap to dirty under, or do
1782 * a almost-no-op setattr
1784 if (issued & CEPH_CAP_AUTH_EXCL)
1785 dirtied |= CEPH_CAP_AUTH_EXCL;
1786 else if (issued & CEPH_CAP_FILE_EXCL)
1787 dirtied |= CEPH_CAP_FILE_EXCL;
1788 else if (issued & CEPH_CAP_XATTR_EXCL)
1789 dirtied |= CEPH_CAP_XATTR_EXCL;
1791 mask |= CEPH_SETATTR_CTIME;
1794 if (ia_valid & ATTR_FILE)
1795 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1798 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1799 inode->i_ctime = CURRENT_TIME;
1803 spin_unlock(&ci->i_ceph_lock);
1805 if (inode_dirty_flags)
1806 __mark_inode_dirty(inode, inode_dirty_flags);
1808 if (ia_valid & ATTR_MODE) {
1809 err = posix_acl_chmod(inode, attr->ia_mode);
1815 req->r_inode = inode;
1817 req->r_inode_drop = release;
1818 req->r_args.setattr.mask = cpu_to_le32(mask);
1819 req->r_num_caps = 1;
1820 parent_inode = ceph_get_dentry_parent_inode(dentry);
1821 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1824 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1825 ceph_cap_string(dirtied), mask);
1827 ceph_mdsc_put_request(req);
1828 if (mask & CEPH_SETATTR_SIZE)
1829 __ceph_do_pending_vmtruncate(inode);
1832 spin_unlock(&ci->i_ceph_lock);
1834 ceph_mdsc_put_request(req);
1839 * Verify that we have a lease on the given mask. If not,
1840 * do a getattr against an mds.
1842 int ceph_do_getattr(struct inode *inode, int mask)
1844 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1845 struct ceph_mds_client *mdsc = fsc->mdsc;
1846 struct ceph_mds_request *req;
1849 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1850 dout("do_getattr inode %p SNAPDIR\n", inode);
1854 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1855 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1858 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1860 return PTR_ERR(req);
1861 req->r_inode = inode;
1863 req->r_num_caps = 1;
1864 req->r_args.getattr.mask = cpu_to_le32(mask);
1865 err = ceph_mdsc_do_request(mdsc, NULL, req);
1866 ceph_mdsc_put_request(req);
1867 dout("do_getattr result=%d\n", err);
1873 * Check inode permissions. We verify we have a valid value for
1874 * the AUTH cap, then call the generic handler.
1876 int ceph_permission(struct inode *inode, int mask)
1880 if (mask & MAY_NOT_BLOCK)
1883 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1886 err = generic_permission(inode, mask);
1891 * Get all attributes. Hopefully somedata we'll have a statlite()
1892 * and can limit the fields we require to be accurate.
1894 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1897 struct inode *inode = dentry->d_inode;
1898 struct ceph_inode_info *ci = ceph_inode(inode);
1901 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1903 generic_fillattr(inode, stat);
1904 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1905 if (ceph_snap(inode) != CEPH_NOSNAP)
1906 stat->dev = ceph_snap(inode);
1909 if (S_ISDIR(inode->i_mode)) {
1910 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1912 stat->size = ci->i_rbytes;
1914 stat->size = ci->i_files + ci->i_subdirs;
1916 stat->blksize = 65536;