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,
100 .set_acl = ceph_set_acl,
105 * We use a 'frag tree' to keep track of the MDS's directory fragments
106 * for a given inode (usually there is just a single fragment). We
107 * need to know when a child frag is delegated to a new MDS, or when
108 * it is flagged as replicated, so we can direct our requests
113 * find/create a frag in the tree
115 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
119 struct rb_node *parent = NULL;
120 struct ceph_inode_frag *frag;
123 p = &ci->i_fragtree.rb_node;
126 frag = rb_entry(parent, struct ceph_inode_frag, node);
127 c = ceph_frag_compare(f, frag->frag);
136 frag = kmalloc(sizeof(*frag), GFP_NOFS);
138 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
139 "frag %x\n", &ci->vfs_inode,
140 ceph_vinop(&ci->vfs_inode), f);
141 return ERR_PTR(-ENOMEM);
148 rb_link_node(&frag->node, parent, p);
149 rb_insert_color(&frag->node, &ci->i_fragtree);
151 dout("get_or_create_frag added %llx.%llx frag %x\n",
152 ceph_vinop(&ci->vfs_inode), f);
157 * find a specific frag @f
159 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
161 struct rb_node *n = ci->i_fragtree.rb_node;
164 struct ceph_inode_frag *frag =
165 rb_entry(n, struct ceph_inode_frag, node);
166 int c = ceph_frag_compare(f, frag->frag);
178 * Choose frag containing the given value @v. If @pfrag is
179 * specified, copy the frag delegation info to the caller if
182 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
183 struct ceph_inode_frag *pfrag,
186 u32 t = ceph_frag_make(0, 0);
187 struct ceph_inode_frag *frag;
194 mutex_lock(&ci->i_fragtree_mutex);
196 WARN_ON(!ceph_frag_contains_value(t, v));
197 frag = __ceph_find_frag(ci, t);
199 break; /* t is a leaf */
200 if (frag->split_by == 0) {
202 memcpy(pfrag, frag, sizeof(*pfrag));
209 nway = 1 << frag->split_by;
210 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
211 frag->split_by, nway);
212 for (i = 0; i < nway; i++) {
213 n = ceph_frag_make_child(t, frag->split_by, i);
214 if (ceph_frag_contains_value(n, v)) {
221 dout("choose_frag(%x) = %x\n", v, t);
223 mutex_unlock(&ci->i_fragtree_mutex);
228 * Process dirfrag (delegation) info from the mds. Include leaf
229 * fragment in tree ONLY if ndist > 0. Otherwise, only
230 * branches/splits are included in i_fragtree)
232 static int ceph_fill_dirfrag(struct inode *inode,
233 struct ceph_mds_reply_dirfrag *dirinfo)
235 struct ceph_inode_info *ci = ceph_inode(inode);
236 struct ceph_inode_frag *frag;
237 u32 id = le32_to_cpu(dirinfo->frag);
238 int mds = le32_to_cpu(dirinfo->auth);
239 int ndist = le32_to_cpu(dirinfo->ndist);
243 mutex_lock(&ci->i_fragtree_mutex);
245 /* no delegation info needed. */
246 frag = __ceph_find_frag(ci, id);
249 if (frag->split_by == 0) {
250 /* tree leaf, remove */
251 dout("fill_dirfrag removed %llx.%llx frag %x"
252 " (no ref)\n", ceph_vinop(inode), id);
253 rb_erase(&frag->node, &ci->i_fragtree);
256 /* tree branch, keep and clear */
257 dout("fill_dirfrag cleared %llx.%llx frag %x"
258 " referral\n", ceph_vinop(inode), id);
266 /* find/add this frag to store mds delegation info */
267 frag = __get_or_create_frag(ci, id);
269 /* this is not the end of the world; we can continue
270 with bad/inaccurate delegation info */
271 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
272 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
278 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
279 for (i = 0; i < frag->ndist; i++)
280 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
281 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
282 ceph_vinop(inode), frag->frag, frag->ndist);
285 mutex_unlock(&ci->i_fragtree_mutex);
291 * initialize a newly allocated inode.
293 struct inode *ceph_alloc_inode(struct super_block *sb)
295 struct ceph_inode_info *ci;
298 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
302 dout("alloc_inode %p\n", &ci->vfs_inode);
304 spin_lock_init(&ci->i_ceph_lock);
307 ci->i_time_warp_seq = 0;
308 ci->i_ceph_flags = 0;
309 atomic_set(&ci->i_release_count, 1);
310 atomic_set(&ci->i_complete_count, 0);
311 ci->i_symlink = NULL;
313 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
315 ci->i_fragtree = RB_ROOT;
316 mutex_init(&ci->i_fragtree_mutex);
318 ci->i_xattrs.blob = NULL;
319 ci->i_xattrs.prealloc_blob = NULL;
320 ci->i_xattrs.dirty = false;
321 ci->i_xattrs.index = RB_ROOT;
322 ci->i_xattrs.count = 0;
323 ci->i_xattrs.names_size = 0;
324 ci->i_xattrs.vals_size = 0;
325 ci->i_xattrs.version = 0;
326 ci->i_xattrs.index_version = 0;
328 ci->i_caps = RB_ROOT;
329 ci->i_auth_cap = NULL;
330 ci->i_dirty_caps = 0;
331 ci->i_flushing_caps = 0;
332 INIT_LIST_HEAD(&ci->i_dirty_item);
333 INIT_LIST_HEAD(&ci->i_flushing_item);
334 ci->i_cap_flush_seq = 0;
335 ci->i_cap_flush_last_tid = 0;
336 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
337 init_waitqueue_head(&ci->i_cap_wq);
338 ci->i_hold_caps_min = 0;
339 ci->i_hold_caps_max = 0;
340 INIT_LIST_HEAD(&ci->i_cap_delay_list);
341 INIT_LIST_HEAD(&ci->i_cap_snaps);
342 ci->i_head_snapc = NULL;
344 ci->i_cap_exporting_issued = 0;
346 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
347 ci->i_nr_by_mode[i] = 0;
349 mutex_init(&ci->i_truncate_mutex);
350 ci->i_truncate_seq = 0;
351 ci->i_truncate_size = 0;
352 ci->i_truncate_pending = 0;
355 ci->i_reported_size = 0;
356 ci->i_wanted_max_size = 0;
357 ci->i_requested_max_size = 0;
361 ci->i_rdcache_ref = 0;
364 ci->i_wrbuffer_ref = 0;
365 ci->i_wrbuffer_ref_head = 0;
366 ci->i_shared_gen = 0;
367 ci->i_rdcache_gen = 0;
368 ci->i_rdcache_revoking = 0;
370 INIT_LIST_HEAD(&ci->i_unsafe_writes);
371 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
372 spin_lock_init(&ci->i_unsafe_lock);
374 ci->i_snap_realm = NULL;
375 INIT_LIST_HEAD(&ci->i_snap_realm_item);
376 INIT_LIST_HEAD(&ci->i_snap_flush_item);
378 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
379 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
381 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
383 ceph_fscache_inode_init(ci);
385 return &ci->vfs_inode;
388 static void ceph_i_callback(struct rcu_head *head)
390 struct inode *inode = container_of(head, struct inode, i_rcu);
391 struct ceph_inode_info *ci = ceph_inode(inode);
393 kmem_cache_free(ceph_inode_cachep, ci);
396 void ceph_destroy_inode(struct inode *inode)
398 struct ceph_inode_info *ci = ceph_inode(inode);
399 struct ceph_inode_frag *frag;
402 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
404 ceph_fscache_unregister_inode_cookie(ci);
406 ceph_queue_caps_release(inode);
409 * we may still have a snap_realm reference if there are stray
410 * caps in i_cap_exporting_issued or i_snap_caps.
412 if (ci->i_snap_realm) {
413 struct ceph_mds_client *mdsc =
414 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
415 struct ceph_snap_realm *realm = ci->i_snap_realm;
417 dout(" dropping residual ref to snap realm %p\n", realm);
418 spin_lock(&realm->inodes_with_caps_lock);
419 list_del_init(&ci->i_snap_realm_item);
420 spin_unlock(&realm->inodes_with_caps_lock);
421 ceph_put_snap_realm(mdsc, realm);
424 kfree(ci->i_symlink);
425 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
426 frag = rb_entry(n, struct ceph_inode_frag, node);
427 rb_erase(n, &ci->i_fragtree);
431 __ceph_destroy_xattrs(ci);
432 if (ci->i_xattrs.blob)
433 ceph_buffer_put(ci->i_xattrs.blob);
434 if (ci->i_xattrs.prealloc_blob)
435 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
437 call_rcu(&inode->i_rcu, ceph_i_callback);
440 int ceph_drop_inode(struct inode *inode)
443 * Positve dentry and corresponding inode are always accompanied
444 * in MDS reply. So no need to keep inode in the cache after
445 * dropping all its aliases.
451 * Helpers to fill in size, ctime, mtime, and atime. We have to be
452 * careful because either the client or MDS may have more up to date
453 * info, depending on which capabilities are held, and whether
454 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
455 * and size are monotonically increasing, except when utimes() or
456 * truncate() increments the corresponding _seq values.)
458 int ceph_fill_file_size(struct inode *inode, int issued,
459 u32 truncate_seq, u64 truncate_size, u64 size)
461 struct ceph_inode_info *ci = ceph_inode(inode);
464 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
465 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
466 dout("size %lld -> %llu\n", inode->i_size, size);
467 inode->i_size = size;
468 inode->i_blocks = (size + (1<<9) - 1) >> 9;
469 ci->i_reported_size = size;
470 if (truncate_seq != ci->i_truncate_seq) {
471 dout("truncate_seq %u -> %u\n",
472 ci->i_truncate_seq, truncate_seq);
473 ci->i_truncate_seq = truncate_seq;
475 /* the MDS should have revoked these caps */
476 WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
479 CEPH_CAP_FILE_LAZYIO));
481 * If we hold relevant caps, or in the case where we're
482 * not the only client referencing this file and we
483 * don't hold those caps, then we need to check whether
484 * the file is either opened or mmaped
486 if ((issued & (CEPH_CAP_FILE_CACHE|
487 CEPH_CAP_FILE_BUFFER)) ||
488 mapping_mapped(inode->i_mapping) ||
489 __ceph_caps_file_wanted(ci)) {
490 ci->i_truncate_pending++;
495 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
496 ci->i_truncate_size != truncate_size) {
497 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
499 ci->i_truncate_size = truncate_size;
503 ceph_fscache_invalidate(inode);
508 void ceph_fill_file_time(struct inode *inode, int issued,
509 u64 time_warp_seq, struct timespec *ctime,
510 struct timespec *mtime, struct timespec *atime)
512 struct ceph_inode_info *ci = ceph_inode(inode);
515 if (issued & (CEPH_CAP_FILE_EXCL|
517 CEPH_CAP_FILE_BUFFER|
519 CEPH_CAP_XATTR_EXCL)) {
520 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
521 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
522 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
523 ctime->tv_sec, ctime->tv_nsec);
524 inode->i_ctime = *ctime;
526 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
527 /* the MDS did a utimes() */
528 dout("mtime %ld.%09ld -> %ld.%09ld "
530 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
531 mtime->tv_sec, mtime->tv_nsec,
532 ci->i_time_warp_seq, (int)time_warp_seq);
534 inode->i_mtime = *mtime;
535 inode->i_atime = *atime;
536 ci->i_time_warp_seq = time_warp_seq;
537 } else if (time_warp_seq == ci->i_time_warp_seq) {
538 /* nobody did utimes(); take the max */
539 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
540 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
541 inode->i_mtime.tv_sec,
542 inode->i_mtime.tv_nsec,
543 mtime->tv_sec, mtime->tv_nsec);
544 inode->i_mtime = *mtime;
546 if (timespec_compare(atime, &inode->i_atime) > 0) {
547 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
548 inode->i_atime.tv_sec,
549 inode->i_atime.tv_nsec,
550 atime->tv_sec, atime->tv_nsec);
551 inode->i_atime = *atime;
553 } else if (issued & CEPH_CAP_FILE_EXCL) {
554 /* we did a utimes(); ignore mds values */
559 /* we have no write|excl caps; whatever the MDS says is true */
560 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
561 inode->i_ctime = *ctime;
562 inode->i_mtime = *mtime;
563 inode->i_atime = *atime;
564 ci->i_time_warp_seq = time_warp_seq;
569 if (warn) /* time_warp_seq shouldn't go backwards */
570 dout("%p mds time_warp_seq %llu < %u\n",
571 inode, time_warp_seq, ci->i_time_warp_seq);
575 * Populate an inode based on info from mds. May be called on new or
578 static int fill_inode(struct inode *inode,
579 struct ceph_mds_reply_info_in *iinfo,
580 struct ceph_mds_reply_dirfrag *dirinfo,
581 struct ceph_mds_session *session,
582 unsigned long ttl_from, int cap_fmode,
583 struct ceph_cap_reservation *caps_reservation)
585 struct ceph_mds_reply_inode *info = iinfo->in;
586 struct ceph_inode_info *ci = ceph_inode(inode);
588 int issued = 0, implemented, new_issued;
589 struct timespec mtime, atime, ctime;
591 struct ceph_inode_frag *frag;
592 struct rb_node *rb_node;
593 struct ceph_buffer *xattr_blob = NULL;
595 bool queue_trunc = false;
596 bool new_version = false;
598 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
599 inode, ceph_vinop(inode), le64_to_cpu(info->version),
603 * prealloc xattr data, if it looks like we'll need it. only
604 * if len > 4 (meaning there are actually xattrs; the first 4
605 * bytes are the xattr count).
607 if (iinfo->xattr_len > 4) {
608 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
610 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
614 spin_lock(&ci->i_ceph_lock);
617 * provided version will be odd if inode value is projected,
618 * even if stable. skip the update if we have newer stable
619 * info (ours>=theirs, e.g. due to racing mds replies), unless
620 * we are getting projected (unstable) info (in which case the
621 * version is odd, and we want ours>theirs).
627 if (ci->i_version == 0 ||
628 ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
629 le64_to_cpu(info->version) > (ci->i_version & ~1)))
632 issued = __ceph_caps_issued(ci, &implemented);
633 issued |= implemented | __ceph_caps_dirty(ci);
634 new_issued = ~issued & le32_to_cpu(info->cap.caps);
637 ci->i_version = le64_to_cpu(info->version);
639 inode->i_rdev = le32_to_cpu(info->rdev);
640 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
642 if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
643 (issued & CEPH_CAP_AUTH_EXCL) == 0) {
644 inode->i_mode = le32_to_cpu(info->mode);
645 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
646 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
647 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
648 from_kuid(&init_user_ns, inode->i_uid),
649 from_kgid(&init_user_ns, inode->i_gid));
652 if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
653 (issued & CEPH_CAP_LINK_EXCL) == 0)
654 set_nlink(inode, le32_to_cpu(info->nlink));
656 if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
657 /* be careful with mtime, atime, size */
658 ceph_decode_timespec(&atime, &info->atime);
659 ceph_decode_timespec(&mtime, &info->mtime);
660 ceph_decode_timespec(&ctime, &info->ctime);
661 ceph_fill_file_time(inode, issued,
662 le32_to_cpu(info->time_warp_seq),
663 &ctime, &mtime, &atime);
667 (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
668 ci->i_layout = info->layout;
669 queue_trunc = ceph_fill_file_size(inode, issued,
670 le32_to_cpu(info->truncate_seq),
671 le64_to_cpu(info->truncate_size),
672 le64_to_cpu(info->size));
673 /* only update max_size on auth cap */
674 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
675 ci->i_max_size != le64_to_cpu(info->max_size)) {
676 dout("max_size %lld -> %llu\n", ci->i_max_size,
677 le64_to_cpu(info->max_size));
678 ci->i_max_size = le64_to_cpu(info->max_size);
683 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
684 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
685 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
686 if (ci->i_xattrs.blob)
687 ceph_buffer_put(ci->i_xattrs.blob);
688 ci->i_xattrs.blob = xattr_blob;
690 memcpy(ci->i_xattrs.blob->vec.iov_base,
691 iinfo->xattr_data, iinfo->xattr_len);
692 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
693 ceph_forget_all_cached_acls(inode);
697 inode->i_mapping->a_ops = &ceph_aops;
698 inode->i_mapping->backing_dev_info =
699 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
701 switch (inode->i_mode & S_IFMT) {
706 init_special_inode(inode, inode->i_mode, inode->i_rdev);
707 inode->i_op = &ceph_file_iops;
710 inode->i_op = &ceph_file_iops;
711 inode->i_fop = &ceph_file_fops;
714 inode->i_op = &ceph_symlink_iops;
715 if (!ci->i_symlink) {
716 u32 symlen = iinfo->symlink_len;
719 spin_unlock(&ci->i_ceph_lock);
722 if (WARN_ON(symlen != inode->i_size))
726 sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
730 spin_lock(&ci->i_ceph_lock);
734 kfree(sym); /* lost a race */
738 inode->i_op = &ceph_dir_iops;
739 inode->i_fop = &ceph_dir_fops;
741 ci->i_dir_layout = iinfo->dir_layout;
743 ci->i_files = le64_to_cpu(info->files);
744 ci->i_subdirs = le64_to_cpu(info->subdirs);
745 ci->i_rbytes = le64_to_cpu(info->rbytes);
746 ci->i_rfiles = le64_to_cpu(info->rfiles);
747 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
748 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
751 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
752 ceph_vinop(inode), inode->i_mode);
755 /* set dir completion flag? */
756 if (S_ISDIR(inode->i_mode) &&
757 ci->i_files == 0 && ci->i_subdirs == 0 &&
758 ceph_snap(inode) == CEPH_NOSNAP &&
759 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
760 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
761 !__ceph_dir_is_complete(ci)) {
762 dout(" marking %p complete (empty)\n", inode);
763 __ceph_dir_set_complete(ci, atomic_read(&ci->i_release_count));
765 spin_unlock(&ci->i_ceph_lock);
767 /* queue truncate if we saw i_size decrease */
769 ceph_queue_vmtruncate(inode);
771 /* populate frag tree */
772 /* FIXME: move me up, if/when version reflects fragtree changes */
773 nsplits = le32_to_cpu(info->fragtree.nsplits);
774 mutex_lock(&ci->i_fragtree_mutex);
775 rb_node = rb_first(&ci->i_fragtree);
776 for (i = 0; i < nsplits; i++) {
777 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
780 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
781 if (ceph_frag_compare(frag->frag, id) >= 0) {
782 if (frag->frag != id)
785 rb_node = rb_next(rb_node);
788 rb_node = rb_next(rb_node);
789 rb_erase(&frag->node, &ci->i_fragtree);
794 frag = __get_or_create_frag(ci, id);
798 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
799 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
802 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
803 rb_node = rb_next(rb_node);
804 rb_erase(&frag->node, &ci->i_fragtree);
807 mutex_unlock(&ci->i_fragtree_mutex);
809 /* were we issued a capability? */
810 if (info->cap.caps) {
811 if (ceph_snap(inode) == CEPH_NOSNAP) {
812 ceph_add_cap(inode, session,
813 le64_to_cpu(info->cap.cap_id),
815 le32_to_cpu(info->cap.caps),
816 le32_to_cpu(info->cap.wanted),
817 le32_to_cpu(info->cap.seq),
818 le32_to_cpu(info->cap.mseq),
819 le64_to_cpu(info->cap.realm),
823 spin_lock(&ci->i_ceph_lock);
824 dout(" %p got snap_caps %s\n", inode,
825 ceph_cap_string(le32_to_cpu(info->cap.caps)));
826 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
828 __ceph_get_fmode(ci, cap_fmode);
829 spin_unlock(&ci->i_ceph_lock);
831 } else if (cap_fmode >= 0) {
832 pr_warning("mds issued no caps on %llx.%llx\n",
834 __ceph_get_fmode(ci, cap_fmode);
837 /* update delegation info? */
839 ceph_fill_dirfrag(inode, dirinfo);
845 ceph_buffer_put(xattr_blob);
850 * caller should hold session s_mutex.
852 static void update_dentry_lease(struct dentry *dentry,
853 struct ceph_mds_reply_lease *lease,
854 struct ceph_mds_session *session,
855 unsigned long from_time)
857 struct ceph_dentry_info *di = ceph_dentry(dentry);
858 long unsigned duration = le32_to_cpu(lease->duration_ms);
859 long unsigned ttl = from_time + (duration * HZ) / 1000;
860 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
863 /* only track leases on regular dentries */
864 if (dentry->d_op != &ceph_dentry_ops)
867 spin_lock(&dentry->d_lock);
868 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
869 dentry, duration, ttl);
871 /* make lease_rdcache_gen match directory */
872 dir = dentry->d_parent->d_inode;
873 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
878 if (di->lease_gen == session->s_cap_gen &&
879 time_before(ttl, dentry->d_time))
880 goto out_unlock; /* we already have a newer lease. */
882 if (di->lease_session && di->lease_session != session)
885 ceph_dentry_lru_touch(dentry);
887 if (!di->lease_session)
888 di->lease_session = ceph_get_mds_session(session);
889 di->lease_gen = session->s_cap_gen;
890 di->lease_seq = le32_to_cpu(lease->seq);
891 di->lease_renew_after = half_ttl;
892 di->lease_renew_from = 0;
893 dentry->d_time = ttl;
895 spin_unlock(&dentry->d_lock);
900 * splice a dentry to an inode.
901 * caller must hold directory i_mutex for this to be safe.
903 * we will only rehash the resulting dentry if @prehash is
904 * true; @prehash will be set to false (for the benefit of
905 * the caller) if we fail.
907 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
910 struct dentry *realdn;
914 /* dn must be unhashed */
917 realdn = d_materialise_unique(dn, in);
918 if (IS_ERR(realdn)) {
919 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
920 PTR_ERR(realdn), dn, in, ceph_vinop(in));
922 *prehash = false; /* don't rehash on error */
923 dn = realdn; /* note realdn contains the error */
926 dout("dn %p (%d) spliced with %p (%d) "
927 "inode %p ino %llx.%llx\n",
929 realdn, d_count(realdn),
930 realdn->d_inode, ceph_vinop(realdn->d_inode));
934 BUG_ON(!ceph_dentry(dn));
935 dout("dn %p attached to %p ino %llx.%llx\n",
936 dn, dn->d_inode, ceph_vinop(dn->d_inode));
938 if ((!prehash || *prehash) && d_unhashed(dn))
945 * Incorporate results into the local cache. This is either just
946 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
949 * A reply may contain
950 * a directory inode along with a dentry.
951 * and/or a target inode
953 * Called with snap_rwsem (read).
955 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
956 struct ceph_mds_session *session)
958 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
959 struct inode *in = NULL;
960 struct ceph_vino vino;
961 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
964 dout("fill_trace %p is_dentry %d is_target %d\n", req,
965 rinfo->head->is_dentry, rinfo->head->is_target);
971 * If we resend completed ops to a recovering mds, we get no
972 * trace. Since that is very rare, pretend this is the case
973 * to ensure the 'no trace' handlers in the callers behave.
975 * Fill in inodes unconditionally to avoid breaking cap
978 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
979 pr_info("fill_trace faking empty trace on %lld %s\n",
980 req->r_tid, ceph_mds_op_name(rinfo->head->op));
981 if (rinfo->head->is_dentry) {
982 rinfo->head->is_dentry = 0;
983 err = fill_inode(req->r_locked_dir,
984 &rinfo->diri, rinfo->dirfrag,
985 session, req->r_request_started, -1);
987 if (rinfo->head->is_target) {
988 rinfo->head->is_target = 0;
989 ininfo = rinfo->targeti.in;
990 vino.ino = le64_to_cpu(ininfo->ino);
991 vino.snap = le64_to_cpu(ininfo->snapid);
992 in = ceph_get_inode(sb, vino);
993 err = fill_inode(in, &rinfo->targeti, NULL,
994 session, req->r_request_started,
1001 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1002 dout("fill_trace reply is empty!\n");
1003 if (rinfo->head->result == 0 && req->r_locked_dir)
1004 ceph_invalidate_dir_request(req);
1008 if (rinfo->head->is_dentry) {
1009 struct inode *dir = req->r_locked_dir;
1012 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
1013 session, req->r_request_started, -1,
1014 &req->r_caps_reservation);
1021 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1023 struct dentry *dn, *parent;
1025 BUG_ON(!rinfo->head->is_target);
1026 BUG_ON(req->r_dentry);
1028 parent = d_find_any_alias(dir);
1031 dname.name = rinfo->dname;
1032 dname.len = rinfo->dname_len;
1033 dname.hash = full_name_hash(dname.name, dname.len);
1034 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1035 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1037 dn = d_lookup(parent, &dname);
1038 dout("d_lookup on parent=%p name=%.*s got %p\n",
1039 parent, dname.len, dname.name, dn);
1042 dn = d_alloc(parent, &dname);
1043 dout("d_alloc %p '%.*s' = %p\n", parent,
1044 dname.len, dname.name, dn);
1050 err = ceph_init_dentry(dn);
1056 } else if (dn->d_inode &&
1057 (ceph_ino(dn->d_inode) != vino.ino ||
1058 ceph_snap(dn->d_inode) != vino.snap)) {
1059 dout(" dn %p points to wrong inode %p\n",
1071 if (rinfo->head->is_target) {
1072 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1073 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1075 in = ceph_get_inode(sb, vino);
1080 req->r_target_inode = in;
1082 err = fill_inode(in, &rinfo->targeti, NULL,
1083 session, req->r_request_started,
1084 (!req->r_aborted && rinfo->head->result == 0) ?
1086 &req->r_caps_reservation);
1088 pr_err("fill_inode badness %p %llx.%llx\n",
1089 in, ceph_vinop(in));
1095 * ignore null lease/binding on snapdir ENOENT, or else we
1096 * will have trouble splicing in the virtual snapdir later
1098 if (rinfo->head->is_dentry && !req->r_aborted &&
1099 req->r_locked_dir &&
1100 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1101 fsc->mount_options->snapdir_name,
1102 req->r_dentry->d_name.len))) {
1104 * lookup link rename : null -> possibly existing inode
1105 * mknod symlink mkdir : null -> new inode
1106 * unlink : linked -> null
1108 struct inode *dir = req->r_locked_dir;
1109 struct dentry *dn = req->r_dentry;
1110 bool have_dir_cap, have_lease;
1114 BUG_ON(dn->d_parent->d_inode != dir);
1115 BUG_ON(ceph_ino(dir) !=
1116 le64_to_cpu(rinfo->diri.in->ino));
1117 BUG_ON(ceph_snap(dir) !=
1118 le64_to_cpu(rinfo->diri.in->snapid));
1120 /* do we have a lease on the whole dir? */
1122 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1123 CEPH_CAP_FILE_SHARED);
1125 /* do we have a dn lease? */
1126 have_lease = have_dir_cap ||
1127 le32_to_cpu(rinfo->dlease->duration_ms);
1129 dout("fill_trace no dentry lease or dir cap\n");
1132 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1133 struct inode *olddir = req->r_old_dentry_dir;
1136 dout(" src %p '%.*s' dst %p '%.*s'\n",
1138 req->r_old_dentry->d_name.len,
1139 req->r_old_dentry->d_name.name,
1140 dn, dn->d_name.len, dn->d_name.name);
1141 dout("fill_trace doing d_move %p -> %p\n",
1142 req->r_old_dentry, dn);
1144 d_move(req->r_old_dentry, dn);
1145 dout(" src %p '%.*s' dst %p '%.*s'\n",
1147 req->r_old_dentry->d_name.len,
1148 req->r_old_dentry->d_name.name,
1149 dn, dn->d_name.len, dn->d_name.name);
1151 /* ensure target dentry is invalidated, despite
1152 rehashing bug in vfs_rename_dir */
1153 ceph_invalidate_dentry_lease(dn);
1155 /* d_move screws up sibling dentries' offsets */
1156 ceph_dir_clear_complete(dir);
1157 ceph_dir_clear_complete(olddir);
1159 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1160 ceph_dentry(req->r_old_dentry)->offset);
1162 dn = req->r_old_dentry; /* use old_dentry */
1166 if (!rinfo->head->is_target) {
1167 dout("fill_trace null dentry\n");
1169 dout("d_delete %p\n", dn);
1172 dout("d_instantiate %p NULL\n", dn);
1173 d_instantiate(dn, NULL);
1174 if (have_lease && d_unhashed(dn))
1176 update_dentry_lease(dn, rinfo->dlease,
1178 req->r_request_started);
1183 /* attach proper inode */
1185 ceph_dir_clear_complete(dir);
1187 dn = splice_dentry(dn, in, &have_lease);
1192 req->r_dentry = dn; /* may have spliced */
1193 } else if (dn->d_inode && dn->d_inode != in) {
1194 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1195 dn, dn->d_inode, ceph_vinop(dn->d_inode),
1201 update_dentry_lease(dn, rinfo->dlease, session,
1202 req->r_request_started);
1203 dout(" final dn %p\n", dn);
1204 } else if (!req->r_aborted &&
1205 (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1206 req->r_op == CEPH_MDS_OP_MKSNAP)) {
1207 struct dentry *dn = req->r_dentry;
1208 struct inode *dir = req->r_locked_dir;
1210 /* fill out a snapdir LOOKUPSNAP dentry */
1213 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1214 dout(" linking snapped dir %p to dn %p\n", in, dn);
1215 ceph_dir_clear_complete(dir);
1217 dn = splice_dentry(dn, in, NULL);
1222 req->r_dentry = dn; /* may have spliced */
1225 dout("fill_trace done err=%d\n", err);
1230 * Prepopulate our cache with readdir results, leases, etc.
1232 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1233 struct ceph_mds_session *session)
1235 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1238 for (i = 0; i < rinfo->dir_nr; i++) {
1239 struct ceph_vino vino;
1243 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1244 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1246 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1249 dout("new_inode badness got %d\n", err);
1252 rc = fill_inode(in, &rinfo->dir_in[i], NULL, session,
1253 req->r_request_started, -1,
1254 &req->r_caps_reservation);
1256 pr_err("fill_inode badness on %p got %d\n", in, rc);
1265 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1266 struct ceph_mds_session *session)
1268 struct dentry *parent = req->r_dentry;
1269 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1273 int err = 0, ret, i;
1274 struct inode *snapdir = NULL;
1275 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1276 struct ceph_dentry_info *di;
1277 u64 r_readdir_offset = req->r_readdir_offset;
1278 u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1280 if (rinfo->dir_dir &&
1281 le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1282 dout("readdir_prepopulate got new frag %x -> %x\n",
1283 frag, le32_to_cpu(rinfo->dir_dir->frag));
1284 frag = le32_to_cpu(rinfo->dir_dir->frag);
1285 if (ceph_frag_is_leftmost(frag))
1286 r_readdir_offset = 2;
1288 r_readdir_offset = 0;
1292 return readdir_prepopulate_inodes_only(req, session);
1294 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1295 snapdir = ceph_get_snapdir(parent->d_inode);
1296 parent = d_find_alias(snapdir);
1297 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1298 rinfo->dir_nr, parent);
1300 dout("readdir_prepopulate %d items under dn %p\n",
1301 rinfo->dir_nr, parent);
1303 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1306 /* FIXME: release caps/leases if error occurs */
1307 for (i = 0; i < rinfo->dir_nr; i++) {
1308 struct ceph_vino vino;
1310 dname.name = rinfo->dir_dname[i];
1311 dname.len = rinfo->dir_dname_len[i];
1312 dname.hash = full_name_hash(dname.name, dname.len);
1314 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1315 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1318 dn = d_lookup(parent, &dname);
1319 dout("d_lookup on parent=%p name=%.*s got %p\n",
1320 parent, dname.len, dname.name, dn);
1323 dn = d_alloc(parent, &dname);
1324 dout("d_alloc %p '%.*s' = %p\n", parent,
1325 dname.len, dname.name, dn);
1327 dout("d_alloc badness\n");
1331 ret = ceph_init_dentry(dn);
1337 } else if (dn->d_inode &&
1338 (ceph_ino(dn->d_inode) != vino.ino ||
1339 ceph_snap(dn->d_inode) != vino.snap)) {
1340 dout(" dn %p points to wrong inode %p\n",
1346 /* reorder parent's d_subdirs */
1347 spin_lock(&parent->d_lock);
1348 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1349 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1350 spin_unlock(&dn->d_lock);
1351 spin_unlock(&parent->d_lock);
1358 in = ceph_get_inode(parent->d_sb, vino);
1360 dout("new_inode badness\n");
1368 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1369 req->r_request_started, -1,
1370 &req->r_caps_reservation) < 0) {
1371 pr_err("fill_inode badness on %p\n", in);
1379 dn = splice_dentry(dn, in, NULL);
1388 di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1390 update_dentry_lease(dn, rinfo->dir_dlease[i],
1392 req->r_request_started);
1398 req->r_did_prepopulate = true;
1405 dout("readdir_prepopulate done\n");
1409 int ceph_inode_set_size(struct inode *inode, loff_t size)
1411 struct ceph_inode_info *ci = ceph_inode(inode);
1414 spin_lock(&ci->i_ceph_lock);
1415 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1416 inode->i_size = size;
1417 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1419 /* tell the MDS if we are approaching max_size */
1420 if ((size << 1) >= ci->i_max_size &&
1421 (ci->i_reported_size << 1) < ci->i_max_size)
1424 spin_unlock(&ci->i_ceph_lock);
1429 * Write back inode data in a worker thread. (This can't be done
1430 * in the message handler context.)
1432 void ceph_queue_writeback(struct inode *inode)
1435 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1436 &ceph_inode(inode)->i_wb_work)) {
1437 dout("ceph_queue_writeback %p\n", inode);
1439 dout("ceph_queue_writeback %p failed\n", inode);
1444 static void ceph_writeback_work(struct work_struct *work)
1446 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1448 struct inode *inode = &ci->vfs_inode;
1450 dout("writeback %p\n", inode);
1451 filemap_fdatawrite(&inode->i_data);
1456 * queue an async invalidation
1458 void ceph_queue_invalidate(struct inode *inode)
1461 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1462 &ceph_inode(inode)->i_pg_inv_work)) {
1463 dout("ceph_queue_invalidate %p\n", inode);
1465 dout("ceph_queue_invalidate %p failed\n", inode);
1471 * Invalidate inode pages in a worker thread. (This can't be done
1472 * in the message handler context.)
1474 static void ceph_invalidate_work(struct work_struct *work)
1476 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1478 struct inode *inode = &ci->vfs_inode;
1482 mutex_lock(&ci->i_truncate_mutex);
1483 spin_lock(&ci->i_ceph_lock);
1484 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1485 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1486 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1487 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1489 spin_unlock(&ci->i_ceph_lock);
1490 mutex_unlock(&ci->i_truncate_mutex);
1493 orig_gen = ci->i_rdcache_gen;
1494 spin_unlock(&ci->i_ceph_lock);
1496 truncate_inode_pages(inode->i_mapping, 0);
1498 spin_lock(&ci->i_ceph_lock);
1499 if (orig_gen == ci->i_rdcache_gen &&
1500 orig_gen == ci->i_rdcache_revoking) {
1501 dout("invalidate_pages %p gen %d successful\n", inode,
1503 ci->i_rdcache_revoking--;
1506 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1507 inode, orig_gen, ci->i_rdcache_gen,
1508 ci->i_rdcache_revoking);
1509 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1512 spin_unlock(&ci->i_ceph_lock);
1513 mutex_unlock(&ci->i_truncate_mutex);
1516 ceph_check_caps(ci, 0, NULL);
1522 * called by trunc_wq;
1524 * We also truncate in a separate thread as well.
1526 static void ceph_vmtruncate_work(struct work_struct *work)
1528 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1530 struct inode *inode = &ci->vfs_inode;
1532 dout("vmtruncate_work %p\n", inode);
1533 __ceph_do_pending_vmtruncate(inode);
1538 * Queue an async vmtruncate. If we fail to queue work, we will handle
1539 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1541 void ceph_queue_vmtruncate(struct inode *inode)
1543 struct ceph_inode_info *ci = ceph_inode(inode);
1547 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1548 &ci->i_vmtruncate_work)) {
1549 dout("ceph_queue_vmtruncate %p\n", inode);
1551 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1552 inode, ci->i_truncate_pending);
1558 * Make sure any pending truncation is applied before doing anything
1559 * that may depend on it.
1561 void __ceph_do_pending_vmtruncate(struct inode *inode)
1563 struct ceph_inode_info *ci = ceph_inode(inode);
1565 int wrbuffer_refs, finish = 0;
1567 mutex_lock(&ci->i_truncate_mutex);
1569 spin_lock(&ci->i_ceph_lock);
1570 if (ci->i_truncate_pending == 0) {
1571 dout("__do_pending_vmtruncate %p none pending\n", inode);
1572 spin_unlock(&ci->i_ceph_lock);
1573 mutex_unlock(&ci->i_truncate_mutex);
1578 * make sure any dirty snapped pages are flushed before we
1579 * possibly truncate them.. so write AND block!
1581 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1582 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1584 spin_unlock(&ci->i_ceph_lock);
1585 filemap_write_and_wait_range(&inode->i_data, 0,
1586 inode->i_sb->s_maxbytes);
1590 /* there should be no reader or writer */
1591 WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1593 to = ci->i_truncate_size;
1594 wrbuffer_refs = ci->i_wrbuffer_ref;
1595 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1596 ci->i_truncate_pending, to);
1597 spin_unlock(&ci->i_ceph_lock);
1599 truncate_inode_pages(inode->i_mapping, to);
1601 spin_lock(&ci->i_ceph_lock);
1602 if (to == ci->i_truncate_size) {
1603 ci->i_truncate_pending = 0;
1606 spin_unlock(&ci->i_ceph_lock);
1610 mutex_unlock(&ci->i_truncate_mutex);
1612 if (wrbuffer_refs == 0)
1613 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1615 wake_up_all(&ci->i_cap_wq);
1621 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1623 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1624 nd_set_link(nd, ci->i_symlink);
1628 static const struct inode_operations ceph_symlink_iops = {
1629 .readlink = generic_readlink,
1630 .follow_link = ceph_sym_follow_link,
1631 .setattr = ceph_setattr,
1632 .getattr = ceph_getattr,
1633 .setxattr = ceph_setxattr,
1634 .getxattr = ceph_getxattr,
1635 .listxattr = ceph_listxattr,
1636 .removexattr = ceph_removexattr,
1642 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1644 struct inode *inode = dentry->d_inode;
1645 struct ceph_inode_info *ci = ceph_inode(inode);
1646 const unsigned int ia_valid = attr->ia_valid;
1647 struct ceph_mds_request *req;
1648 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1650 int release = 0, dirtied = 0;
1653 int inode_dirty_flags = 0;
1655 if (ceph_snap(inode) != CEPH_NOSNAP)
1658 err = inode_change_ok(inode, attr);
1662 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1665 return PTR_ERR(req);
1667 spin_lock(&ci->i_ceph_lock);
1668 issued = __ceph_caps_issued(ci, NULL);
1669 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1671 if (ia_valid & ATTR_UID) {
1672 dout("setattr %p uid %d -> %d\n", inode,
1673 from_kuid(&init_user_ns, inode->i_uid),
1674 from_kuid(&init_user_ns, attr->ia_uid));
1675 if (issued & CEPH_CAP_AUTH_EXCL) {
1676 inode->i_uid = attr->ia_uid;
1677 dirtied |= CEPH_CAP_AUTH_EXCL;
1678 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1679 !uid_eq(attr->ia_uid, inode->i_uid)) {
1680 req->r_args.setattr.uid = cpu_to_le32(
1681 from_kuid(&init_user_ns, attr->ia_uid));
1682 mask |= CEPH_SETATTR_UID;
1683 release |= CEPH_CAP_AUTH_SHARED;
1686 if (ia_valid & ATTR_GID) {
1687 dout("setattr %p gid %d -> %d\n", inode,
1688 from_kgid(&init_user_ns, inode->i_gid),
1689 from_kgid(&init_user_ns, attr->ia_gid));
1690 if (issued & CEPH_CAP_AUTH_EXCL) {
1691 inode->i_gid = attr->ia_gid;
1692 dirtied |= CEPH_CAP_AUTH_EXCL;
1693 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1694 !gid_eq(attr->ia_gid, inode->i_gid)) {
1695 req->r_args.setattr.gid = cpu_to_le32(
1696 from_kgid(&init_user_ns, attr->ia_gid));
1697 mask |= CEPH_SETATTR_GID;
1698 release |= CEPH_CAP_AUTH_SHARED;
1701 if (ia_valid & ATTR_MODE) {
1702 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1704 if (issued & CEPH_CAP_AUTH_EXCL) {
1705 inode->i_mode = attr->ia_mode;
1706 dirtied |= CEPH_CAP_AUTH_EXCL;
1707 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1708 attr->ia_mode != inode->i_mode) {
1709 inode->i_mode = attr->ia_mode;
1710 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1711 mask |= CEPH_SETATTR_MODE;
1712 release |= CEPH_CAP_AUTH_SHARED;
1716 if (ia_valid & ATTR_ATIME) {
1717 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1718 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1719 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1720 if (issued & CEPH_CAP_FILE_EXCL) {
1721 ci->i_time_warp_seq++;
1722 inode->i_atime = attr->ia_atime;
1723 dirtied |= CEPH_CAP_FILE_EXCL;
1724 } else if ((issued & CEPH_CAP_FILE_WR) &&
1725 timespec_compare(&inode->i_atime,
1726 &attr->ia_atime) < 0) {
1727 inode->i_atime = attr->ia_atime;
1728 dirtied |= CEPH_CAP_FILE_WR;
1729 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1730 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1731 ceph_encode_timespec(&req->r_args.setattr.atime,
1733 mask |= CEPH_SETATTR_ATIME;
1734 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1738 if (ia_valid & ATTR_MTIME) {
1739 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1740 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1741 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1742 if (issued & CEPH_CAP_FILE_EXCL) {
1743 ci->i_time_warp_seq++;
1744 inode->i_mtime = attr->ia_mtime;
1745 dirtied |= CEPH_CAP_FILE_EXCL;
1746 } else if ((issued & CEPH_CAP_FILE_WR) &&
1747 timespec_compare(&inode->i_mtime,
1748 &attr->ia_mtime) < 0) {
1749 inode->i_mtime = attr->ia_mtime;
1750 dirtied |= CEPH_CAP_FILE_WR;
1751 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1752 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1753 ceph_encode_timespec(&req->r_args.setattr.mtime,
1755 mask |= CEPH_SETATTR_MTIME;
1756 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1760 if (ia_valid & ATTR_SIZE) {
1761 dout("setattr %p size %lld -> %lld\n", inode,
1762 inode->i_size, attr->ia_size);
1763 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1767 if ((issued & CEPH_CAP_FILE_EXCL) &&
1768 attr->ia_size > inode->i_size) {
1769 inode->i_size = attr->ia_size;
1771 (attr->ia_size + (1 << 9) - 1) >> 9;
1772 inode->i_ctime = attr->ia_ctime;
1773 ci->i_reported_size = attr->ia_size;
1774 dirtied |= CEPH_CAP_FILE_EXCL;
1775 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1776 attr->ia_size != inode->i_size) {
1777 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1778 req->r_args.setattr.old_size =
1779 cpu_to_le64(inode->i_size);
1780 mask |= CEPH_SETATTR_SIZE;
1781 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1786 /* these do nothing */
1787 if (ia_valid & ATTR_CTIME) {
1788 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1789 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1790 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1791 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1792 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1793 only ? "ctime only" : "ignored");
1794 inode->i_ctime = attr->ia_ctime;
1797 * if kernel wants to dirty ctime but nothing else,
1798 * we need to choose a cap to dirty under, or do
1799 * a almost-no-op setattr
1801 if (issued & CEPH_CAP_AUTH_EXCL)
1802 dirtied |= CEPH_CAP_AUTH_EXCL;
1803 else if (issued & CEPH_CAP_FILE_EXCL)
1804 dirtied |= CEPH_CAP_FILE_EXCL;
1805 else if (issued & CEPH_CAP_XATTR_EXCL)
1806 dirtied |= CEPH_CAP_XATTR_EXCL;
1808 mask |= CEPH_SETATTR_CTIME;
1811 if (ia_valid & ATTR_FILE)
1812 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1815 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1816 inode->i_ctime = CURRENT_TIME;
1820 spin_unlock(&ci->i_ceph_lock);
1822 if (inode_dirty_flags)
1823 __mark_inode_dirty(inode, inode_dirty_flags);
1825 if (ia_valid & ATTR_MODE) {
1826 err = posix_acl_chmod(inode, attr->ia_mode);
1832 req->r_inode = inode;
1834 req->r_inode_drop = release;
1835 req->r_args.setattr.mask = cpu_to_le32(mask);
1836 req->r_num_caps = 1;
1837 err = ceph_mdsc_do_request(mdsc, NULL, req);
1839 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1840 ceph_cap_string(dirtied), mask);
1842 ceph_mdsc_put_request(req);
1843 if (mask & CEPH_SETATTR_SIZE)
1844 __ceph_do_pending_vmtruncate(inode);
1847 spin_unlock(&ci->i_ceph_lock);
1849 ceph_mdsc_put_request(req);
1854 * Verify that we have a lease on the given mask. If not,
1855 * do a getattr against an mds.
1857 int ceph_do_getattr(struct inode *inode, int mask)
1859 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1860 struct ceph_mds_client *mdsc = fsc->mdsc;
1861 struct ceph_mds_request *req;
1864 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1865 dout("do_getattr inode %p SNAPDIR\n", inode);
1869 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1870 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1873 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1875 return PTR_ERR(req);
1876 req->r_inode = inode;
1878 req->r_num_caps = 1;
1879 req->r_args.getattr.mask = cpu_to_le32(mask);
1880 err = ceph_mdsc_do_request(mdsc, NULL, req);
1881 ceph_mdsc_put_request(req);
1882 dout("do_getattr result=%d\n", err);
1888 * Check inode permissions. We verify we have a valid value for
1889 * the AUTH cap, then call the generic handler.
1891 int ceph_permission(struct inode *inode, int mask)
1895 if (mask & MAY_NOT_BLOCK)
1898 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1901 err = generic_permission(inode, mask);
1906 * Get all attributes. Hopefully somedata we'll have a statlite()
1907 * and can limit the fields we require to be accurate.
1909 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1912 struct inode *inode = dentry->d_inode;
1913 struct ceph_inode_info *ci = ceph_inode(inode);
1916 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1918 generic_fillattr(inode, stat);
1919 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1920 if (ceph_snap(inode) != CEPH_NOSNAP)
1921 stat->dev = ceph_snap(inode);
1924 if (S_ISDIR(inode->i_mode)) {
1925 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1927 stat->size = ci->i_rbytes;
1929 stat->size = ci->i_files + ci->i_subdirs;
1931 stat->blksize = 65536;