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>
14 #include "mds_client.h"
15 #include <linux/ceph/decode.h>
18 * Ceph inode operations
20 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
21 * setattr, etc.), xattr helpers, and helpers for assimilating
22 * metadata returned by the MDS into our cache.
24 * Also define helpers for doing asynchronous writeback, invalidation,
25 * and truncation for the benefit of those who can't afford to block
26 * (typically because they are in the message handler path).
29 static const struct inode_operations ceph_symlink_iops;
31 static void ceph_invalidate_work(struct work_struct *work);
32 static void ceph_writeback_work(struct work_struct *work);
33 static void ceph_vmtruncate_work(struct work_struct *work);
36 * find or create an inode, given the ceph ino number
38 static int ceph_set_ino_cb(struct inode *inode, void *data)
40 ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
41 inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
45 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
48 ino_t t = ceph_vino_to_ino(vino);
50 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
52 return ERR_PTR(-ENOMEM);
53 if (inode->i_state & I_NEW) {
54 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
55 inode, ceph_vinop(inode), (u64)inode->i_ino);
56 unlock_new_inode(inode);
59 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
65 * get/constuct snapdir inode for a given directory
67 struct inode *ceph_get_snapdir(struct inode *parent)
69 struct ceph_vino vino = {
70 .ino = ceph_ino(parent),
73 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
74 struct ceph_inode_info *ci = ceph_inode(inode);
76 BUG_ON(!S_ISDIR(parent->i_mode));
79 inode->i_mode = parent->i_mode;
80 inode->i_uid = parent->i_uid;
81 inode->i_gid = parent->i_gid;
82 inode->i_op = &ceph_dir_iops;
83 inode->i_fop = &ceph_dir_fops;
84 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
89 const struct inode_operations ceph_file_iops = {
90 .permission = ceph_permission,
91 .setattr = ceph_setattr,
92 .getattr = ceph_getattr,
93 .setxattr = ceph_setxattr,
94 .getxattr = ceph_getxattr,
95 .listxattr = ceph_listxattr,
96 .removexattr = ceph_removexattr,
101 * We use a 'frag tree' to keep track of the MDS's directory fragments
102 * for a given inode (usually there is just a single fragment). We
103 * need to know when a child frag is delegated to a new MDS, or when
104 * it is flagged as replicated, so we can direct our requests
109 * find/create a frag in the tree
111 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
115 struct rb_node *parent = NULL;
116 struct ceph_inode_frag *frag;
119 p = &ci->i_fragtree.rb_node;
122 frag = rb_entry(parent, struct ceph_inode_frag, node);
123 c = ceph_frag_compare(f, frag->frag);
132 frag = kmalloc(sizeof(*frag), GFP_NOFS);
134 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
135 "frag %x\n", &ci->vfs_inode,
136 ceph_vinop(&ci->vfs_inode), f);
137 return ERR_PTR(-ENOMEM);
144 rb_link_node(&frag->node, parent, p);
145 rb_insert_color(&frag->node, &ci->i_fragtree);
147 dout("get_or_create_frag added %llx.%llx frag %x\n",
148 ceph_vinop(&ci->vfs_inode), f);
153 * find a specific frag @f
155 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
157 struct rb_node *n = ci->i_fragtree.rb_node;
160 struct ceph_inode_frag *frag =
161 rb_entry(n, struct ceph_inode_frag, node);
162 int c = ceph_frag_compare(f, frag->frag);
174 * Choose frag containing the given value @v. If @pfrag is
175 * specified, copy the frag delegation info to the caller if
178 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
179 struct ceph_inode_frag *pfrag,
182 u32 t = ceph_frag_make(0, 0);
183 struct ceph_inode_frag *frag;
190 mutex_lock(&ci->i_fragtree_mutex);
192 WARN_ON(!ceph_frag_contains_value(t, v));
193 frag = __ceph_find_frag(ci, t);
195 break; /* t is a leaf */
196 if (frag->split_by == 0) {
198 memcpy(pfrag, frag, sizeof(*pfrag));
205 nway = 1 << frag->split_by;
206 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
207 frag->split_by, nway);
208 for (i = 0; i < nway; i++) {
209 n = ceph_frag_make_child(t, frag->split_by, i);
210 if (ceph_frag_contains_value(n, v)) {
217 dout("choose_frag(%x) = %x\n", v, t);
219 mutex_unlock(&ci->i_fragtree_mutex);
224 * Process dirfrag (delegation) info from the mds. Include leaf
225 * fragment in tree ONLY if ndist > 0. Otherwise, only
226 * branches/splits are included in i_fragtree)
228 static int ceph_fill_dirfrag(struct inode *inode,
229 struct ceph_mds_reply_dirfrag *dirinfo)
231 struct ceph_inode_info *ci = ceph_inode(inode);
232 struct ceph_inode_frag *frag;
233 u32 id = le32_to_cpu(dirinfo->frag);
234 int mds = le32_to_cpu(dirinfo->auth);
235 int ndist = le32_to_cpu(dirinfo->ndist);
239 mutex_lock(&ci->i_fragtree_mutex);
241 /* no delegation info needed. */
242 frag = __ceph_find_frag(ci, id);
245 if (frag->split_by == 0) {
246 /* tree leaf, remove */
247 dout("fill_dirfrag removed %llx.%llx frag %x"
248 " (no ref)\n", ceph_vinop(inode), id);
249 rb_erase(&frag->node, &ci->i_fragtree);
252 /* tree branch, keep and clear */
253 dout("fill_dirfrag cleared %llx.%llx frag %x"
254 " referral\n", ceph_vinop(inode), id);
262 /* find/add this frag to store mds delegation info */
263 frag = __get_or_create_frag(ci, id);
265 /* this is not the end of the world; we can continue
266 with bad/inaccurate delegation info */
267 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
268 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
274 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
275 for (i = 0; i < frag->ndist; i++)
276 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
277 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
278 ceph_vinop(inode), frag->frag, frag->ndist);
281 mutex_unlock(&ci->i_fragtree_mutex);
287 * initialize a newly allocated inode.
289 struct inode *ceph_alloc_inode(struct super_block *sb)
291 struct ceph_inode_info *ci;
294 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
298 dout("alloc_inode %p\n", &ci->vfs_inode);
301 ci->i_time_warp_seq = 0;
302 ci->i_ceph_flags = 0;
303 ci->i_release_count = 0;
304 ci->i_symlink = NULL;
306 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
308 ci->i_fragtree = RB_ROOT;
309 mutex_init(&ci->i_fragtree_mutex);
311 ci->i_xattrs.blob = NULL;
312 ci->i_xattrs.prealloc_blob = NULL;
313 ci->i_xattrs.dirty = false;
314 ci->i_xattrs.index = RB_ROOT;
315 ci->i_xattrs.count = 0;
316 ci->i_xattrs.names_size = 0;
317 ci->i_xattrs.vals_size = 0;
318 ci->i_xattrs.version = 0;
319 ci->i_xattrs.index_version = 0;
321 ci->i_caps = RB_ROOT;
322 ci->i_auth_cap = NULL;
323 ci->i_dirty_caps = 0;
324 ci->i_flushing_caps = 0;
325 INIT_LIST_HEAD(&ci->i_dirty_item);
326 INIT_LIST_HEAD(&ci->i_flushing_item);
327 ci->i_cap_flush_seq = 0;
328 ci->i_cap_flush_last_tid = 0;
329 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
330 init_waitqueue_head(&ci->i_cap_wq);
331 ci->i_hold_caps_min = 0;
332 ci->i_hold_caps_max = 0;
333 INIT_LIST_HEAD(&ci->i_cap_delay_list);
334 ci->i_cap_exporting_mds = 0;
335 ci->i_cap_exporting_mseq = 0;
336 ci->i_cap_exporting_issued = 0;
337 INIT_LIST_HEAD(&ci->i_cap_snaps);
338 ci->i_head_snapc = NULL;
341 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
342 ci->i_nr_by_mode[i] = 0;
344 ci->i_truncate_seq = 0;
345 ci->i_truncate_size = 0;
346 ci->i_truncate_pending = 0;
349 ci->i_reported_size = 0;
350 ci->i_wanted_max_size = 0;
351 ci->i_requested_max_size = 0;
355 ci->i_rdcache_ref = 0;
358 ci->i_wrbuffer_ref = 0;
359 ci->i_wrbuffer_ref_head = 0;
360 ci->i_shared_gen = 0;
361 ci->i_rdcache_gen = 0;
362 ci->i_rdcache_revoking = 0;
364 INIT_LIST_HEAD(&ci->i_unsafe_writes);
365 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
366 spin_lock_init(&ci->i_unsafe_lock);
368 ci->i_snap_realm = NULL;
369 INIT_LIST_HEAD(&ci->i_snap_realm_item);
370 INIT_LIST_HEAD(&ci->i_snap_flush_item);
372 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
373 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
375 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
377 return &ci->vfs_inode;
380 static void ceph_i_callback(struct rcu_head *head)
382 struct inode *inode = container_of(head, struct inode, i_rcu);
383 struct ceph_inode_info *ci = ceph_inode(inode);
385 INIT_LIST_HEAD(&inode->i_dentry);
386 kmem_cache_free(ceph_inode_cachep, ci);
389 void ceph_destroy_inode(struct inode *inode)
391 struct ceph_inode_info *ci = ceph_inode(inode);
392 struct ceph_inode_frag *frag;
395 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
397 ceph_queue_caps_release(inode);
400 * we may still have a snap_realm reference if there are stray
401 * caps in i_cap_exporting_issued or i_snap_caps.
403 if (ci->i_snap_realm) {
404 struct ceph_mds_client *mdsc =
405 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
406 struct ceph_snap_realm *realm = ci->i_snap_realm;
408 dout(" dropping residual ref to snap realm %p\n", realm);
409 spin_lock(&realm->inodes_with_caps_lock);
410 list_del_init(&ci->i_snap_realm_item);
411 spin_unlock(&realm->inodes_with_caps_lock);
412 ceph_put_snap_realm(mdsc, realm);
415 kfree(ci->i_symlink);
416 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
417 frag = rb_entry(n, struct ceph_inode_frag, node);
418 rb_erase(n, &ci->i_fragtree);
422 __ceph_destroy_xattrs(ci);
423 if (ci->i_xattrs.blob)
424 ceph_buffer_put(ci->i_xattrs.blob);
425 if (ci->i_xattrs.prealloc_blob)
426 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
428 call_rcu(&inode->i_rcu, ceph_i_callback);
433 * Helpers to fill in size, ctime, mtime, and atime. We have to be
434 * careful because either the client or MDS may have more up to date
435 * info, depending on which capabilities are held, and whether
436 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
437 * and size are monotonically increasing, except when utimes() or
438 * truncate() increments the corresponding _seq values.)
440 int ceph_fill_file_size(struct inode *inode, int issued,
441 u32 truncate_seq, u64 truncate_size, u64 size)
443 struct ceph_inode_info *ci = ceph_inode(inode);
446 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
447 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
448 dout("size %lld -> %llu\n", inode->i_size, size);
449 inode->i_size = size;
450 inode->i_blocks = (size + (1<<9) - 1) >> 9;
451 ci->i_reported_size = size;
452 if (truncate_seq != ci->i_truncate_seq) {
453 dout("truncate_seq %u -> %u\n",
454 ci->i_truncate_seq, truncate_seq);
455 ci->i_truncate_seq = truncate_seq;
457 * If we hold relevant caps, or in the case where we're
458 * not the only client referencing this file and we
459 * don't hold those caps, then we need to check whether
460 * the file is either opened or mmaped
462 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
463 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
465 CEPH_CAP_FILE_LAZYIO)) ||
466 mapping_mapped(inode->i_mapping) ||
467 __ceph_caps_file_wanted(ci)) {
468 ci->i_truncate_pending++;
473 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
474 ci->i_truncate_size != truncate_size) {
475 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
477 ci->i_truncate_size = truncate_size;
482 void ceph_fill_file_time(struct inode *inode, int issued,
483 u64 time_warp_seq, struct timespec *ctime,
484 struct timespec *mtime, struct timespec *atime)
486 struct ceph_inode_info *ci = ceph_inode(inode);
489 if (issued & (CEPH_CAP_FILE_EXCL|
491 CEPH_CAP_FILE_BUFFER|
493 CEPH_CAP_XATTR_EXCL)) {
494 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
495 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
496 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
497 ctime->tv_sec, ctime->tv_nsec);
498 inode->i_ctime = *ctime;
500 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
501 /* the MDS did a utimes() */
502 dout("mtime %ld.%09ld -> %ld.%09ld "
504 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
505 mtime->tv_sec, mtime->tv_nsec,
506 ci->i_time_warp_seq, (int)time_warp_seq);
508 inode->i_mtime = *mtime;
509 inode->i_atime = *atime;
510 ci->i_time_warp_seq = time_warp_seq;
511 } else if (time_warp_seq == ci->i_time_warp_seq) {
512 /* nobody did utimes(); take the max */
513 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
514 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
515 inode->i_mtime.tv_sec,
516 inode->i_mtime.tv_nsec,
517 mtime->tv_sec, mtime->tv_nsec);
518 inode->i_mtime = *mtime;
520 if (timespec_compare(atime, &inode->i_atime) > 0) {
521 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
522 inode->i_atime.tv_sec,
523 inode->i_atime.tv_nsec,
524 atime->tv_sec, atime->tv_nsec);
525 inode->i_atime = *atime;
527 } else if (issued & CEPH_CAP_FILE_EXCL) {
528 /* we did a utimes(); ignore mds values */
533 /* we have no write|excl caps; whatever the MDS says is true */
534 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
535 inode->i_ctime = *ctime;
536 inode->i_mtime = *mtime;
537 inode->i_atime = *atime;
538 ci->i_time_warp_seq = time_warp_seq;
543 if (warn) /* time_warp_seq shouldn't go backwards */
544 dout("%p mds time_warp_seq %llu < %u\n",
545 inode, time_warp_seq, ci->i_time_warp_seq);
549 * Populate an inode based on info from mds. May be called on new or
552 static int fill_inode(struct inode *inode,
553 struct ceph_mds_reply_info_in *iinfo,
554 struct ceph_mds_reply_dirfrag *dirinfo,
555 struct ceph_mds_session *session,
556 unsigned long ttl_from, int cap_fmode,
557 struct ceph_cap_reservation *caps_reservation)
559 struct ceph_mds_reply_inode *info = iinfo->in;
560 struct ceph_inode_info *ci = ceph_inode(inode);
562 int issued = 0, implemented;
563 int updating_inode = 0;
564 struct timespec mtime, atime, ctime;
566 struct ceph_buffer *xattr_blob = NULL;
570 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
571 inode, ceph_vinop(inode), le64_to_cpu(info->version),
575 * prealloc xattr data, if it looks like we'll need it. only
576 * if len > 4 (meaning there are actually xattrs; the first 4
577 * bytes are the xattr count).
579 if (iinfo->xattr_len > 4) {
580 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
582 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
586 spin_lock(&inode->i_lock);
589 * provided version will be odd if inode value is projected,
590 * even if stable. skip the update if we have newer stable
591 * info (ours>=theirs, e.g. due to racing mds replies), unless
592 * we are getting projected (unstable) info (in which case the
593 * version is odd, and we want ours>theirs).
599 if (le64_to_cpu(info->version) > 0 &&
600 (ci->i_version & ~1) >= le64_to_cpu(info->version))
604 issued = __ceph_caps_issued(ci, &implemented);
605 issued |= implemented | __ceph_caps_dirty(ci);
608 ci->i_version = le64_to_cpu(info->version);
610 inode->i_rdev = le32_to_cpu(info->rdev);
612 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
613 inode->i_mode = le32_to_cpu(info->mode);
614 inode->i_uid = le32_to_cpu(info->uid);
615 inode->i_gid = le32_to_cpu(info->gid);
616 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
617 inode->i_uid, inode->i_gid);
620 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
621 set_nlink(inode, le32_to_cpu(info->nlink));
623 /* be careful with mtime, atime, size */
624 ceph_decode_timespec(&atime, &info->atime);
625 ceph_decode_timespec(&mtime, &info->mtime);
626 ceph_decode_timespec(&ctime, &info->ctime);
627 queue_trunc = ceph_fill_file_size(inode, issued,
628 le32_to_cpu(info->truncate_seq),
629 le64_to_cpu(info->truncate_size),
630 le64_to_cpu(info->size));
631 ceph_fill_file_time(inode, issued,
632 le32_to_cpu(info->time_warp_seq),
633 &ctime, &mtime, &atime);
635 /* only update max_size on auth cap */
636 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
637 ci->i_max_size != le64_to_cpu(info->max_size)) {
638 dout("max_size %lld -> %llu\n", ci->i_max_size,
639 le64_to_cpu(info->max_size));
640 ci->i_max_size = le64_to_cpu(info->max_size);
643 ci->i_layout = info->layout;
644 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
647 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
648 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
649 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
650 if (ci->i_xattrs.blob)
651 ceph_buffer_put(ci->i_xattrs.blob);
652 ci->i_xattrs.blob = xattr_blob;
654 memcpy(ci->i_xattrs.blob->vec.iov_base,
655 iinfo->xattr_data, iinfo->xattr_len);
656 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
660 inode->i_mapping->a_ops = &ceph_aops;
661 inode->i_mapping->backing_dev_info =
662 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
664 switch (inode->i_mode & S_IFMT) {
669 init_special_inode(inode, inode->i_mode, inode->i_rdev);
670 inode->i_op = &ceph_file_iops;
673 inode->i_op = &ceph_file_iops;
674 inode->i_fop = &ceph_file_fops;
677 inode->i_op = &ceph_symlink_iops;
678 if (!ci->i_symlink) {
679 int symlen = iinfo->symlink_len;
682 BUG_ON(symlen != inode->i_size);
683 spin_unlock(&inode->i_lock);
686 sym = kmalloc(symlen+1, GFP_NOFS);
689 memcpy(sym, iinfo->symlink, symlen);
692 spin_lock(&inode->i_lock);
696 kfree(sym); /* lost a race */
700 inode->i_op = &ceph_dir_iops;
701 inode->i_fop = &ceph_dir_fops;
703 ci->i_dir_layout = iinfo->dir_layout;
705 ci->i_files = le64_to_cpu(info->files);
706 ci->i_subdirs = le64_to_cpu(info->subdirs);
707 ci->i_rbytes = le64_to_cpu(info->rbytes);
708 ci->i_rfiles = le64_to_cpu(info->rfiles);
709 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
710 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
713 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
714 ceph_vinop(inode), inode->i_mode);
718 spin_unlock(&inode->i_lock);
720 /* queue truncate if we saw i_size decrease */
722 ceph_queue_vmtruncate(inode);
724 /* populate frag tree */
725 /* FIXME: move me up, if/when version reflects fragtree changes */
726 nsplits = le32_to_cpu(info->fragtree.nsplits);
727 mutex_lock(&ci->i_fragtree_mutex);
728 for (i = 0; i < nsplits; i++) {
729 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
730 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
734 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
735 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
737 mutex_unlock(&ci->i_fragtree_mutex);
739 /* were we issued a capability? */
740 if (info->cap.caps) {
741 if (ceph_snap(inode) == CEPH_NOSNAP) {
742 ceph_add_cap(inode, session,
743 le64_to_cpu(info->cap.cap_id),
745 le32_to_cpu(info->cap.caps),
746 le32_to_cpu(info->cap.wanted),
747 le32_to_cpu(info->cap.seq),
748 le32_to_cpu(info->cap.mseq),
749 le64_to_cpu(info->cap.realm),
753 spin_lock(&inode->i_lock);
754 dout(" %p got snap_caps %s\n", inode,
755 ceph_cap_string(le32_to_cpu(info->cap.caps)));
756 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
758 __ceph_get_fmode(ci, cap_fmode);
759 spin_unlock(&inode->i_lock);
761 } else if (cap_fmode >= 0) {
762 pr_warning("mds issued no caps on %llx.%llx\n",
764 __ceph_get_fmode(ci, cap_fmode);
767 /* set dir completion flag? */
768 if (S_ISDIR(inode->i_mode) &&
769 updating_inode && /* didn't jump to no_change */
770 ci->i_files == 0 && ci->i_subdirs == 0 &&
771 ceph_snap(inode) == CEPH_NOSNAP &&
772 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
773 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
774 !ceph_dir_test_complete(inode)) {
775 dout(" marking %p complete (empty)\n", inode);
776 ceph_dir_set_complete(inode);
777 ci->i_max_offset = 2;
780 /* update delegation info? */
782 ceph_fill_dirfrag(inode, dirinfo);
788 ceph_buffer_put(xattr_blob);
793 * caller should hold session s_mutex.
795 static void update_dentry_lease(struct dentry *dentry,
796 struct ceph_mds_reply_lease *lease,
797 struct ceph_mds_session *session,
798 unsigned long from_time)
800 struct ceph_dentry_info *di = ceph_dentry(dentry);
801 long unsigned duration = le32_to_cpu(lease->duration_ms);
802 long unsigned ttl = from_time + (duration * HZ) / 1000;
803 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
806 /* only track leases on regular dentries */
807 if (dentry->d_op != &ceph_dentry_ops)
810 spin_lock(&dentry->d_lock);
811 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
812 dentry, duration, ttl);
814 /* make lease_rdcache_gen match directory */
815 dir = dentry->d_parent->d_inode;
816 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
821 if (di->lease_gen == session->s_cap_gen &&
822 time_before(ttl, dentry->d_time))
823 goto out_unlock; /* we already have a newer lease. */
825 if (di->lease_session && di->lease_session != session)
828 ceph_dentry_lru_touch(dentry);
830 if (!di->lease_session)
831 di->lease_session = ceph_get_mds_session(session);
832 di->lease_gen = session->s_cap_gen;
833 di->lease_seq = le32_to_cpu(lease->seq);
834 di->lease_renew_after = half_ttl;
835 di->lease_renew_from = 0;
836 dentry->d_time = ttl;
838 spin_unlock(&dentry->d_lock);
843 * Set dentry's directory position based on the current dir's max, and
844 * order it in d_subdirs, so that dcache_readdir behaves.
846 * Always called under directory's i_mutex.
848 static void ceph_set_dentry_offset(struct dentry *dn)
850 struct dentry *dir = dn->d_parent;
851 struct inode *inode = dir->d_inode;
852 struct ceph_dentry_info *di;
856 di = ceph_dentry(dn);
858 spin_lock(&inode->i_lock);
859 if (!ceph_dir_test_complete(inode)) {
860 spin_unlock(&inode->i_lock);
863 di->offset = ceph_inode(inode)->i_max_offset++;
864 spin_unlock(&inode->i_lock);
866 spin_lock(&dir->d_lock);
867 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
868 list_move(&dn->d_u.d_child, &dir->d_subdirs);
869 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
870 dn->d_u.d_child.prev, dn->d_u.d_child.next);
871 spin_unlock(&dn->d_lock);
872 spin_unlock(&dir->d_lock);
876 * splice a dentry to an inode.
877 * caller must hold directory i_mutex for this to be safe.
879 * we will only rehash the resulting dentry if @prehash is
880 * true; @prehash will be set to false (for the benefit of
881 * the caller) if we fail.
883 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
884 bool *prehash, bool set_offset)
886 struct dentry *realdn;
890 /* dn must be unhashed */
893 realdn = d_materialise_unique(dn, in);
894 if (IS_ERR(realdn)) {
895 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
896 PTR_ERR(realdn), dn, in, ceph_vinop(in));
898 *prehash = false; /* don't rehash on error */
899 dn = realdn; /* note realdn contains the error */
902 dout("dn %p (%d) spliced with %p (%d) "
903 "inode %p ino %llx.%llx\n",
905 realdn, realdn->d_count,
906 realdn->d_inode, ceph_vinop(realdn->d_inode));
910 BUG_ON(!ceph_dentry(dn));
911 dout("dn %p attached to %p ino %llx.%llx\n",
912 dn, dn->d_inode, ceph_vinop(dn->d_inode));
914 if ((!prehash || *prehash) && d_unhashed(dn))
917 ceph_set_dentry_offset(dn);
923 * Incorporate results into the local cache. This is either just
924 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
927 * A reply may contain
928 * a directory inode along with a dentry.
929 * and/or a target inode
931 * Called with snap_rwsem (read).
933 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
934 struct ceph_mds_session *session)
936 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
937 struct inode *in = NULL;
938 struct ceph_mds_reply_inode *ininfo;
939 struct ceph_vino vino;
940 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
944 dout("fill_trace %p is_dentry %d is_target %d\n", req,
945 rinfo->head->is_dentry, rinfo->head->is_target);
951 * If we resend completed ops to a recovering mds, we get no
952 * trace. Since that is very rare, pretend this is the case
953 * to ensure the 'no trace' handlers in the callers behave.
955 * Fill in inodes unconditionally to avoid breaking cap
958 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
959 pr_info("fill_trace faking empty trace on %lld %s\n",
960 req->r_tid, ceph_mds_op_name(rinfo->head->op));
961 if (rinfo->head->is_dentry) {
962 rinfo->head->is_dentry = 0;
963 err = fill_inode(req->r_locked_dir,
964 &rinfo->diri, rinfo->dirfrag,
965 session, req->r_request_started, -1);
967 if (rinfo->head->is_target) {
968 rinfo->head->is_target = 0;
969 ininfo = rinfo->targeti.in;
970 vino.ino = le64_to_cpu(ininfo->ino);
971 vino.snap = le64_to_cpu(ininfo->snapid);
972 in = ceph_get_inode(sb, vino);
973 err = fill_inode(in, &rinfo->targeti, NULL,
974 session, req->r_request_started,
981 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
982 dout("fill_trace reply is empty!\n");
983 if (rinfo->head->result == 0 && req->r_locked_dir)
984 ceph_invalidate_dir_request(req);
988 if (rinfo->head->is_dentry) {
989 struct inode *dir = req->r_locked_dir;
991 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
992 session, req->r_request_started, -1,
993 &req->r_caps_reservation);
999 * ignore null lease/binding on snapdir ENOENT, or else we
1000 * will have trouble splicing in the virtual snapdir later
1002 if (rinfo->head->is_dentry && !req->r_aborted &&
1003 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1004 fsc->mount_options->snapdir_name,
1005 req->r_dentry->d_name.len))) {
1007 * lookup link rename : null -> possibly existing inode
1008 * mknod symlink mkdir : null -> new inode
1009 * unlink : linked -> null
1011 struct inode *dir = req->r_locked_dir;
1012 struct dentry *dn = req->r_dentry;
1013 bool have_dir_cap, have_lease;
1017 BUG_ON(dn->d_parent->d_inode != dir);
1018 BUG_ON(ceph_ino(dir) !=
1019 le64_to_cpu(rinfo->diri.in->ino));
1020 BUG_ON(ceph_snap(dir) !=
1021 le64_to_cpu(rinfo->diri.in->snapid));
1023 /* do we have a lease on the whole dir? */
1025 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1026 CEPH_CAP_FILE_SHARED);
1028 /* do we have a dn lease? */
1029 have_lease = have_dir_cap ||
1030 le32_to_cpu(rinfo->dlease->duration_ms);
1032 dout("fill_trace no dentry lease or dir cap\n");
1035 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1036 dout(" src %p '%.*s' dst %p '%.*s'\n",
1038 req->r_old_dentry->d_name.len,
1039 req->r_old_dentry->d_name.name,
1040 dn, dn->d_name.len, dn->d_name.name);
1041 dout("fill_trace doing d_move %p -> %p\n",
1042 req->r_old_dentry, dn);
1044 d_move(req->r_old_dentry, dn);
1045 dout(" src %p '%.*s' dst %p '%.*s'\n",
1047 req->r_old_dentry->d_name.len,
1048 req->r_old_dentry->d_name.name,
1049 dn, dn->d_name.len, dn->d_name.name);
1051 /* ensure target dentry is invalidated, despite
1052 rehashing bug in vfs_rename_dir */
1053 ceph_invalidate_dentry_lease(dn);
1056 * d_move() puts the renamed dentry at the end of
1057 * d_subdirs. We need to assign it an appropriate
1058 * directory offset so we can behave when holding
1061 ceph_set_dentry_offset(req->r_old_dentry);
1062 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1063 ceph_dentry(req->r_old_dentry)->offset);
1065 dn = req->r_old_dentry; /* use old_dentry */
1070 if (!rinfo->head->is_target) {
1071 dout("fill_trace null dentry\n");
1073 dout("d_delete %p\n", dn);
1076 dout("d_instantiate %p NULL\n", dn);
1077 d_instantiate(dn, NULL);
1078 if (have_lease && d_unhashed(dn))
1080 update_dentry_lease(dn, rinfo->dlease,
1082 req->r_request_started);
1087 /* attach proper inode */
1088 ininfo = rinfo->targeti.in;
1089 vino.ino = le64_to_cpu(ininfo->ino);
1090 vino.snap = le64_to_cpu(ininfo->snapid);
1093 in = ceph_get_inode(sb, vino);
1095 pr_err("fill_trace bad get_inode "
1096 "%llx.%llx\n", vino.ino, vino.snap);
1101 dn = splice_dentry(dn, in, &have_lease, true);
1106 req->r_dentry = dn; /* may have spliced */
1108 } else if (ceph_ino(in) == vino.ino &&
1109 ceph_snap(in) == vino.snap) {
1112 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1113 dn, in, ceph_ino(in), ceph_snap(in),
1114 vino.ino, vino.snap);
1120 update_dentry_lease(dn, rinfo->dlease, session,
1121 req->r_request_started);
1122 dout(" final dn %p\n", dn);
1124 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1125 req->r_op == CEPH_MDS_OP_MKSNAP) {
1126 struct dentry *dn = req->r_dentry;
1128 /* fill out a snapdir LOOKUPSNAP dentry */
1130 BUG_ON(!req->r_locked_dir);
1131 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1132 ininfo = rinfo->targeti.in;
1133 vino.ino = le64_to_cpu(ininfo->ino);
1134 vino.snap = le64_to_cpu(ininfo->snapid);
1135 in = ceph_get_inode(sb, vino);
1137 pr_err("fill_inode get_inode badness %llx.%llx\n",
1138 vino.ino, vino.snap);
1143 dout(" linking snapped dir %p to dn %p\n", in, dn);
1144 dn = splice_dentry(dn, in, NULL, true);
1149 req->r_dentry = dn; /* may have spliced */
1151 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1154 if (rinfo->head->is_target) {
1155 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1156 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1158 if (in == NULL || ceph_ino(in) != vino.ino ||
1159 ceph_snap(in) != vino.snap) {
1160 in = ceph_get_inode(sb, vino);
1166 req->r_target_inode = in;
1168 err = fill_inode(in,
1169 &rinfo->targeti, NULL,
1170 session, req->r_request_started,
1171 (le32_to_cpu(rinfo->head->result) == 0) ?
1173 &req->r_caps_reservation);
1175 pr_err("fill_inode badness %p %llx.%llx\n",
1176 in, ceph_vinop(in));
1182 dout("fill_trace done err=%d\n", err);
1187 * Prepopulate our cache with readdir results, leases, etc.
1189 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1190 struct ceph_mds_session *session)
1192 struct dentry *parent = req->r_dentry;
1193 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1198 struct inode *snapdir = NULL;
1199 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1200 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1201 struct ceph_dentry_info *di;
1203 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1204 snapdir = ceph_get_snapdir(parent->d_inode);
1205 parent = d_find_alias(snapdir);
1206 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1207 rinfo->dir_nr, parent);
1209 dout("readdir_prepopulate %d items under dn %p\n",
1210 rinfo->dir_nr, parent);
1212 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1215 for (i = 0; i < rinfo->dir_nr; i++) {
1216 struct ceph_vino vino;
1218 dname.name = rinfo->dir_dname[i];
1219 dname.len = rinfo->dir_dname_len[i];
1220 dname.hash = full_name_hash(dname.name, dname.len);
1222 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1223 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1226 dn = d_lookup(parent, &dname);
1227 dout("d_lookup on parent=%p name=%.*s got %p\n",
1228 parent, dname.len, dname.name, dn);
1231 dn = d_alloc(parent, &dname);
1232 dout("d_alloc %p '%.*s' = %p\n", parent,
1233 dname.len, dname.name, dn);
1235 dout("d_alloc badness\n");
1239 err = ceph_init_dentry(dn);
1244 } else if (dn->d_inode &&
1245 (ceph_ino(dn->d_inode) != vino.ino ||
1246 ceph_snap(dn->d_inode) != vino.snap)) {
1247 dout(" dn %p points to wrong inode %p\n",
1253 /* reorder parent's d_subdirs */
1254 spin_lock(&parent->d_lock);
1255 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1256 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1257 spin_unlock(&dn->d_lock);
1258 spin_unlock(&parent->d_lock);
1262 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1268 in = ceph_get_inode(parent->d_sb, vino);
1270 dout("new_inode badness\n");
1276 dn = splice_dentry(dn, in, NULL, false);
1281 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1282 req->r_request_started, -1,
1283 &req->r_caps_reservation) < 0) {
1284 pr_err("fill_inode badness on %p\n", in);
1288 update_dentry_lease(dn, rinfo->dir_dlease[i],
1290 req->r_request_started);
1295 req->r_did_prepopulate = true;
1302 dout("readdir_prepopulate done\n");
1306 int ceph_inode_set_size(struct inode *inode, loff_t size)
1308 struct ceph_inode_info *ci = ceph_inode(inode);
1311 spin_lock(&inode->i_lock);
1312 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1313 inode->i_size = size;
1314 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1316 /* tell the MDS if we are approaching max_size */
1317 if ((size << 1) >= ci->i_max_size &&
1318 (ci->i_reported_size << 1) < ci->i_max_size)
1321 spin_unlock(&inode->i_lock);
1326 * Write back inode data in a worker thread. (This can't be done
1327 * in the message handler context.)
1329 void ceph_queue_writeback(struct inode *inode)
1331 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1332 &ceph_inode(inode)->i_wb_work)) {
1333 dout("ceph_queue_writeback %p\n", inode);
1336 dout("ceph_queue_writeback %p failed\n", inode);
1340 static void ceph_writeback_work(struct work_struct *work)
1342 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1344 struct inode *inode = &ci->vfs_inode;
1346 dout("writeback %p\n", inode);
1347 filemap_fdatawrite(&inode->i_data);
1352 * queue an async invalidation
1354 void ceph_queue_invalidate(struct inode *inode)
1356 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1357 &ceph_inode(inode)->i_pg_inv_work)) {
1358 dout("ceph_queue_invalidate %p\n", inode);
1361 dout("ceph_queue_invalidate %p failed\n", inode);
1366 * Invalidate inode pages in a worker thread. (This can't be done
1367 * in the message handler context.)
1369 static void ceph_invalidate_work(struct work_struct *work)
1371 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1373 struct inode *inode = &ci->vfs_inode;
1377 spin_lock(&inode->i_lock);
1378 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1379 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1380 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1382 spin_unlock(&inode->i_lock);
1385 orig_gen = ci->i_rdcache_gen;
1386 spin_unlock(&inode->i_lock);
1388 truncate_inode_pages(&inode->i_data, 0);
1390 spin_lock(&inode->i_lock);
1391 if (orig_gen == ci->i_rdcache_gen &&
1392 orig_gen == ci->i_rdcache_revoking) {
1393 dout("invalidate_pages %p gen %d successful\n", inode,
1395 ci->i_rdcache_revoking--;
1398 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1399 inode, orig_gen, ci->i_rdcache_gen,
1400 ci->i_rdcache_revoking);
1402 spin_unlock(&inode->i_lock);
1405 ceph_check_caps(ci, 0, NULL);
1412 * called by trunc_wq; take i_mutex ourselves
1414 * We also truncate in a separate thread as well.
1416 static void ceph_vmtruncate_work(struct work_struct *work)
1418 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1420 struct inode *inode = &ci->vfs_inode;
1422 dout("vmtruncate_work %p\n", inode);
1423 mutex_lock(&inode->i_mutex);
1424 __ceph_do_pending_vmtruncate(inode);
1425 mutex_unlock(&inode->i_mutex);
1430 * Queue an async vmtruncate. If we fail to queue work, we will handle
1431 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1433 void ceph_queue_vmtruncate(struct inode *inode)
1435 struct ceph_inode_info *ci = ceph_inode(inode);
1437 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1438 &ci->i_vmtruncate_work)) {
1439 dout("ceph_queue_vmtruncate %p\n", inode);
1442 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1443 inode, ci->i_truncate_pending);
1448 * called with i_mutex held.
1450 * Make sure any pending truncation is applied before doing anything
1451 * that may depend on it.
1453 void __ceph_do_pending_vmtruncate(struct inode *inode)
1455 struct ceph_inode_info *ci = ceph_inode(inode);
1457 int wrbuffer_refs, wake = 0;
1460 spin_lock(&inode->i_lock);
1461 if (ci->i_truncate_pending == 0) {
1462 dout("__do_pending_vmtruncate %p none pending\n", inode);
1463 spin_unlock(&inode->i_lock);
1468 * make sure any dirty snapped pages are flushed before we
1469 * possibly truncate them.. so write AND block!
1471 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1472 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1474 spin_unlock(&inode->i_lock);
1475 filemap_write_and_wait_range(&inode->i_data, 0,
1476 inode->i_sb->s_maxbytes);
1480 to = ci->i_truncate_size;
1481 wrbuffer_refs = ci->i_wrbuffer_ref;
1482 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1483 ci->i_truncate_pending, to);
1484 spin_unlock(&inode->i_lock);
1486 truncate_inode_pages(inode->i_mapping, to);
1488 spin_lock(&inode->i_lock);
1489 ci->i_truncate_pending--;
1490 if (ci->i_truncate_pending == 0)
1492 spin_unlock(&inode->i_lock);
1494 if (wrbuffer_refs == 0)
1495 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1497 wake_up_all(&ci->i_cap_wq);
1504 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1506 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1507 nd_set_link(nd, ci->i_symlink);
1511 static const struct inode_operations ceph_symlink_iops = {
1512 .readlink = generic_readlink,
1513 .follow_link = ceph_sym_follow_link,
1519 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1521 struct inode *inode = dentry->d_inode;
1522 struct ceph_inode_info *ci = ceph_inode(inode);
1523 struct inode *parent_inode;
1524 const unsigned int ia_valid = attr->ia_valid;
1525 struct ceph_mds_request *req;
1526 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1528 int release = 0, dirtied = 0;
1531 int inode_dirty_flags = 0;
1533 if (ceph_snap(inode) != CEPH_NOSNAP)
1536 __ceph_do_pending_vmtruncate(inode);
1538 err = inode_change_ok(inode, attr);
1542 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1545 return PTR_ERR(req);
1547 spin_lock(&inode->i_lock);
1548 issued = __ceph_caps_issued(ci, NULL);
1549 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1551 if (ia_valid & ATTR_UID) {
1552 dout("setattr %p uid %d -> %d\n", inode,
1553 inode->i_uid, attr->ia_uid);
1554 if (issued & CEPH_CAP_AUTH_EXCL) {
1555 inode->i_uid = attr->ia_uid;
1556 dirtied |= CEPH_CAP_AUTH_EXCL;
1557 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1558 attr->ia_uid != inode->i_uid) {
1559 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1560 mask |= CEPH_SETATTR_UID;
1561 release |= CEPH_CAP_AUTH_SHARED;
1564 if (ia_valid & ATTR_GID) {
1565 dout("setattr %p gid %d -> %d\n", inode,
1566 inode->i_gid, attr->ia_gid);
1567 if (issued & CEPH_CAP_AUTH_EXCL) {
1568 inode->i_gid = attr->ia_gid;
1569 dirtied |= CEPH_CAP_AUTH_EXCL;
1570 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1571 attr->ia_gid != inode->i_gid) {
1572 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1573 mask |= CEPH_SETATTR_GID;
1574 release |= CEPH_CAP_AUTH_SHARED;
1577 if (ia_valid & ATTR_MODE) {
1578 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1580 if (issued & CEPH_CAP_AUTH_EXCL) {
1581 inode->i_mode = attr->ia_mode;
1582 dirtied |= CEPH_CAP_AUTH_EXCL;
1583 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1584 attr->ia_mode != inode->i_mode) {
1585 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1586 mask |= CEPH_SETATTR_MODE;
1587 release |= CEPH_CAP_AUTH_SHARED;
1591 if (ia_valid & ATTR_ATIME) {
1592 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1593 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1594 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1595 if (issued & CEPH_CAP_FILE_EXCL) {
1596 ci->i_time_warp_seq++;
1597 inode->i_atime = attr->ia_atime;
1598 dirtied |= CEPH_CAP_FILE_EXCL;
1599 } else if ((issued & CEPH_CAP_FILE_WR) &&
1600 timespec_compare(&inode->i_atime,
1601 &attr->ia_atime) < 0) {
1602 inode->i_atime = attr->ia_atime;
1603 dirtied |= CEPH_CAP_FILE_WR;
1604 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1605 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1606 ceph_encode_timespec(&req->r_args.setattr.atime,
1608 mask |= CEPH_SETATTR_ATIME;
1609 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1613 if (ia_valid & ATTR_MTIME) {
1614 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1615 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1616 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1617 if (issued & CEPH_CAP_FILE_EXCL) {
1618 ci->i_time_warp_seq++;
1619 inode->i_mtime = attr->ia_mtime;
1620 dirtied |= CEPH_CAP_FILE_EXCL;
1621 } else if ((issued & CEPH_CAP_FILE_WR) &&
1622 timespec_compare(&inode->i_mtime,
1623 &attr->ia_mtime) < 0) {
1624 inode->i_mtime = attr->ia_mtime;
1625 dirtied |= CEPH_CAP_FILE_WR;
1626 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1627 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1628 ceph_encode_timespec(&req->r_args.setattr.mtime,
1630 mask |= CEPH_SETATTR_MTIME;
1631 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1635 if (ia_valid & ATTR_SIZE) {
1636 dout("setattr %p size %lld -> %lld\n", inode,
1637 inode->i_size, attr->ia_size);
1638 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1642 if ((issued & CEPH_CAP_FILE_EXCL) &&
1643 attr->ia_size > inode->i_size) {
1644 inode->i_size = attr->ia_size;
1646 (attr->ia_size + (1 << 9) - 1) >> 9;
1647 inode->i_ctime = attr->ia_ctime;
1648 ci->i_reported_size = attr->ia_size;
1649 dirtied |= CEPH_CAP_FILE_EXCL;
1650 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1651 attr->ia_size != inode->i_size) {
1652 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1653 req->r_args.setattr.old_size =
1654 cpu_to_le64(inode->i_size);
1655 mask |= CEPH_SETATTR_SIZE;
1656 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1661 /* these do nothing */
1662 if (ia_valid & ATTR_CTIME) {
1663 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1664 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1665 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1666 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1667 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1668 only ? "ctime only" : "ignored");
1669 inode->i_ctime = attr->ia_ctime;
1672 * if kernel wants to dirty ctime but nothing else,
1673 * we need to choose a cap to dirty under, or do
1674 * a almost-no-op setattr
1676 if (issued & CEPH_CAP_AUTH_EXCL)
1677 dirtied |= CEPH_CAP_AUTH_EXCL;
1678 else if (issued & CEPH_CAP_FILE_EXCL)
1679 dirtied |= CEPH_CAP_FILE_EXCL;
1680 else if (issued & CEPH_CAP_XATTR_EXCL)
1681 dirtied |= CEPH_CAP_XATTR_EXCL;
1683 mask |= CEPH_SETATTR_CTIME;
1686 if (ia_valid & ATTR_FILE)
1687 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1690 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1691 inode->i_ctime = CURRENT_TIME;
1695 spin_unlock(&inode->i_lock);
1697 if (inode_dirty_flags)
1698 __mark_inode_dirty(inode, inode_dirty_flags);
1701 req->r_inode = inode;
1703 req->r_inode_drop = release;
1704 req->r_args.setattr.mask = cpu_to_le32(mask);
1705 req->r_num_caps = 1;
1706 parent_inode = ceph_get_dentry_parent_inode(dentry);
1707 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1710 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1711 ceph_cap_string(dirtied), mask);
1713 ceph_mdsc_put_request(req);
1714 __ceph_do_pending_vmtruncate(inode);
1717 spin_unlock(&inode->i_lock);
1718 ceph_mdsc_put_request(req);
1723 * Verify that we have a lease on the given mask. If not,
1724 * do a getattr against an mds.
1726 int ceph_do_getattr(struct inode *inode, int mask)
1728 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1729 struct ceph_mds_client *mdsc = fsc->mdsc;
1730 struct ceph_mds_request *req;
1733 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1734 dout("do_getattr inode %p SNAPDIR\n", inode);
1738 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1739 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1742 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1744 return PTR_ERR(req);
1745 req->r_inode = inode;
1747 req->r_num_caps = 1;
1748 req->r_args.getattr.mask = cpu_to_le32(mask);
1749 err = ceph_mdsc_do_request(mdsc, NULL, req);
1750 ceph_mdsc_put_request(req);
1751 dout("do_getattr result=%d\n", err);
1757 * Check inode permissions. We verify we have a valid value for
1758 * the AUTH cap, then call the generic handler.
1760 int ceph_permission(struct inode *inode, int mask)
1764 if (mask & MAY_NOT_BLOCK)
1767 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1770 err = generic_permission(inode, mask);
1775 * Get all attributes. Hopefully somedata we'll have a statlite()
1776 * and can limit the fields we require to be accurate.
1778 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1781 struct inode *inode = dentry->d_inode;
1782 struct ceph_inode_info *ci = ceph_inode(inode);
1785 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1787 generic_fillattr(inode, stat);
1788 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1789 if (ceph_snap(inode) != CEPH_NOSNAP)
1790 stat->dev = ceph_snap(inode);
1793 if (S_ISDIR(inode->i_mode)) {
1794 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1796 stat->size = ci->i_rbytes;
1798 stat->size = ci->i_files + ci->i_subdirs;
1800 stat->blksize = 65536;