1 #include "ceph_debug.h"
3 #include <linux/module.h>
5 #include <linux/smp_lock.h>
6 #include <linux/slab.h>
7 #include <linux/string.h>
8 #include <linux/uaccess.h>
9 #include <linux/kernel.h>
10 #include <linux/namei.h>
11 #include <linux/writeback.h>
12 #include <linux/vmalloc.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_inode_invalidate_pages(struct work_struct *work);
34 * find or create an inode, given the ceph ino number
36 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
39 ino_t t = ceph_vino_to_ino(vino);
41 inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
43 return ERR_PTR(-ENOMEM);
44 if (inode->i_state & I_NEW) {
45 dout("get_inode created new inode %p %llx.%llx ino %llx\n",
46 inode, ceph_vinop(inode), (u64)inode->i_ino);
47 unlock_new_inode(inode);
50 dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
56 * get/constuct snapdir inode for a given directory
58 struct inode *ceph_get_snapdir(struct inode *parent)
60 struct ceph_vino vino = {
61 .ino = ceph_ino(parent),
64 struct inode *inode = ceph_get_inode(parent->i_sb, vino);
65 struct ceph_inode_info *ci = ceph_inode(inode);
67 BUG_ON(!S_ISDIR(parent->i_mode));
69 return ERR_PTR(PTR_ERR(inode));
70 inode->i_mode = parent->i_mode;
71 inode->i_uid = parent->i_uid;
72 inode->i_gid = parent->i_gid;
73 inode->i_op = &ceph_dir_iops;
74 inode->i_fop = &ceph_dir_fops;
75 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
80 const struct inode_operations ceph_file_iops = {
81 .permission = ceph_permission,
82 .setattr = ceph_setattr,
83 .getattr = ceph_getattr,
84 .setxattr = ceph_setxattr,
85 .getxattr = ceph_getxattr,
86 .listxattr = ceph_listxattr,
87 .removexattr = ceph_removexattr,
92 * We use a 'frag tree' to keep track of the MDS's directory fragments
93 * for a given inode (usually there is just a single fragment). We
94 * need to know when a child frag is delegated to a new MDS, or when
95 * it is flagged as replicated, so we can direct our requests
100 * find/create a frag in the tree
102 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
106 struct rb_node *parent = NULL;
107 struct ceph_inode_frag *frag;
110 p = &ci->i_fragtree.rb_node;
113 frag = rb_entry(parent, struct ceph_inode_frag, node);
114 c = ceph_frag_compare(f, frag->frag);
123 frag = kmalloc(sizeof(*frag), GFP_NOFS);
125 pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
126 "frag %x\n", &ci->vfs_inode,
127 ceph_vinop(&ci->vfs_inode), f);
128 return ERR_PTR(-ENOMEM);
135 rb_link_node(&frag->node, parent, p);
136 rb_insert_color(&frag->node, &ci->i_fragtree);
138 dout("get_or_create_frag added %llx.%llx frag %x\n",
139 ceph_vinop(&ci->vfs_inode), f);
144 * find a specific frag @f
146 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
148 struct rb_node *n = ci->i_fragtree.rb_node;
151 struct ceph_inode_frag *frag =
152 rb_entry(n, struct ceph_inode_frag, node);
153 int c = ceph_frag_compare(f, frag->frag);
165 * Choose frag containing the given value @v. If @pfrag is
166 * specified, copy the frag delegation info to the caller if
169 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
170 struct ceph_inode_frag *pfrag,
173 u32 t = ceph_frag_make(0, 0);
174 struct ceph_inode_frag *frag;
181 mutex_lock(&ci->i_fragtree_mutex);
183 WARN_ON(!ceph_frag_contains_value(t, v));
184 frag = __ceph_find_frag(ci, t);
186 break; /* t is a leaf */
187 if (frag->split_by == 0) {
189 memcpy(pfrag, frag, sizeof(*pfrag));
196 nway = 1 << frag->split_by;
197 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
198 frag->split_by, nway);
199 for (i = 0; i < nway; i++) {
200 n = ceph_frag_make_child(t, frag->split_by, i);
201 if (ceph_frag_contains_value(n, v)) {
208 dout("choose_frag(%x) = %x\n", v, t);
210 mutex_unlock(&ci->i_fragtree_mutex);
215 * Process dirfrag (delegation) info from the mds. Include leaf
216 * fragment in tree ONLY if ndist > 0. Otherwise, only
217 * branches/splits are included in i_fragtree)
219 static int ceph_fill_dirfrag(struct inode *inode,
220 struct ceph_mds_reply_dirfrag *dirinfo)
222 struct ceph_inode_info *ci = ceph_inode(inode);
223 struct ceph_inode_frag *frag;
224 u32 id = le32_to_cpu(dirinfo->frag);
225 int mds = le32_to_cpu(dirinfo->auth);
226 int ndist = le32_to_cpu(dirinfo->ndist);
230 mutex_lock(&ci->i_fragtree_mutex);
232 /* no delegation info needed. */
233 frag = __ceph_find_frag(ci, id);
236 if (frag->split_by == 0) {
237 /* tree leaf, remove */
238 dout("fill_dirfrag removed %llx.%llx frag %x"
239 " (no ref)\n", ceph_vinop(inode), id);
240 rb_erase(&frag->node, &ci->i_fragtree);
243 /* tree branch, keep and clear */
244 dout("fill_dirfrag cleared %llx.%llx frag %x"
245 " referral\n", ceph_vinop(inode), id);
253 /* find/add this frag to store mds delegation info */
254 frag = __get_or_create_frag(ci, id);
256 /* this is not the end of the world; we can continue
257 with bad/inaccurate delegation info */
258 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
259 ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
265 frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
266 for (i = 0; i < frag->ndist; i++)
267 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
268 dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
269 ceph_vinop(inode), frag->frag, frag->ndist);
272 mutex_unlock(&ci->i_fragtree_mutex);
278 * initialize a newly allocated inode.
280 struct inode *ceph_alloc_inode(struct super_block *sb)
282 struct ceph_inode_info *ci;
285 ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
289 dout("alloc_inode %p\n", &ci->vfs_inode);
292 ci->i_time_warp_seq = 0;
293 ci->i_ceph_flags = 0;
294 ci->i_release_count = 0;
295 ci->i_symlink = NULL;
297 ci->i_fragtree = RB_ROOT;
298 mutex_init(&ci->i_fragtree_mutex);
300 ci->i_xattrs.blob = NULL;
301 ci->i_xattrs.prealloc_blob = NULL;
302 ci->i_xattrs.dirty = false;
303 ci->i_xattrs.index = RB_ROOT;
304 ci->i_xattrs.count = 0;
305 ci->i_xattrs.names_size = 0;
306 ci->i_xattrs.vals_size = 0;
307 ci->i_xattrs.version = 0;
308 ci->i_xattrs.index_version = 0;
310 ci->i_caps = RB_ROOT;
311 ci->i_auth_cap = NULL;
312 ci->i_dirty_caps = 0;
313 ci->i_flushing_caps = 0;
314 INIT_LIST_HEAD(&ci->i_dirty_item);
315 INIT_LIST_HEAD(&ci->i_flushing_item);
316 ci->i_cap_flush_seq = 0;
317 ci->i_cap_flush_last_tid = 0;
318 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
319 init_waitqueue_head(&ci->i_cap_wq);
320 ci->i_hold_caps_min = 0;
321 ci->i_hold_caps_max = 0;
322 INIT_LIST_HEAD(&ci->i_cap_delay_list);
323 ci->i_cap_exporting_mds = 0;
324 ci->i_cap_exporting_mseq = 0;
325 ci->i_cap_exporting_issued = 0;
326 INIT_LIST_HEAD(&ci->i_cap_snaps);
327 ci->i_head_snapc = NULL;
330 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
331 ci->i_nr_by_mode[i] = 0;
333 ci->i_truncate_seq = 0;
334 ci->i_truncate_size = 0;
335 ci->i_truncate_pending = 0;
338 ci->i_reported_size = 0;
339 ci->i_wanted_max_size = 0;
340 ci->i_requested_max_size = 0;
344 ci->i_rdcache_ref = 0;
346 ci->i_wrbuffer_ref = 0;
347 ci->i_wrbuffer_ref_head = 0;
348 ci->i_shared_gen = 0;
349 ci->i_rdcache_gen = 0;
350 ci->i_rdcache_revoking = 0;
352 INIT_LIST_HEAD(&ci->i_unsafe_writes);
353 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
354 spin_lock_init(&ci->i_unsafe_lock);
356 ci->i_snap_realm = NULL;
357 INIT_LIST_HEAD(&ci->i_snap_realm_item);
358 INIT_LIST_HEAD(&ci->i_snap_flush_item);
360 INIT_WORK(&ci->i_wb_work, ceph_inode_writeback);
361 INIT_WORK(&ci->i_pg_inv_work, ceph_inode_invalidate_pages);
363 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
365 return &ci->vfs_inode;
368 void ceph_destroy_inode(struct inode *inode)
370 struct ceph_inode_info *ci = ceph_inode(inode);
371 struct ceph_inode_frag *frag;
374 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
376 ceph_queue_caps_release(inode);
378 kfree(ci->i_symlink);
379 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
380 frag = rb_entry(n, struct ceph_inode_frag, node);
381 rb_erase(n, &ci->i_fragtree);
385 __ceph_destroy_xattrs(ci);
386 if (ci->i_xattrs.blob)
387 ceph_buffer_put(ci->i_xattrs.blob);
388 if (ci->i_xattrs.prealloc_blob)
389 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
391 kmem_cache_free(ceph_inode_cachep, ci);
396 * Helpers to fill in size, ctime, mtime, and atime. We have to be
397 * careful because either the client or MDS may have more up to date
398 * info, depending on which capabilities are held, and whether
399 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
400 * and size are monotonically increasing, except when utimes() or
401 * truncate() increments the corresponding _seq values.)
403 int ceph_fill_file_size(struct inode *inode, int issued,
404 u32 truncate_seq, u64 truncate_size, u64 size)
406 struct ceph_inode_info *ci = ceph_inode(inode);
409 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
410 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
411 dout("size %lld -> %llu\n", inode->i_size, size);
412 inode->i_size = size;
413 inode->i_blocks = (size + (1<<9) - 1) >> 9;
414 ci->i_reported_size = size;
415 if (truncate_seq != ci->i_truncate_seq) {
416 dout("truncate_seq %u -> %u\n",
417 ci->i_truncate_seq, truncate_seq);
418 ci->i_truncate_seq = truncate_seq;
419 if (issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
420 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
421 CEPH_CAP_FILE_EXCL)) {
422 ci->i_truncate_pending++;
427 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
428 ci->i_truncate_size != truncate_size) {
429 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
431 ci->i_truncate_size = truncate_size;
436 void ceph_fill_file_time(struct inode *inode, int issued,
437 u64 time_warp_seq, struct timespec *ctime,
438 struct timespec *mtime, struct timespec *atime)
440 struct ceph_inode_info *ci = ceph_inode(inode);
443 if (issued & (CEPH_CAP_FILE_EXCL|
445 CEPH_CAP_FILE_BUFFER)) {
446 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
447 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
448 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
449 ctime->tv_sec, ctime->tv_nsec);
450 inode->i_ctime = *ctime;
452 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
453 /* the MDS did a utimes() */
454 dout("mtime %ld.%09ld -> %ld.%09ld "
456 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
457 mtime->tv_sec, mtime->tv_nsec,
458 ci->i_time_warp_seq, (int)time_warp_seq);
460 inode->i_mtime = *mtime;
461 inode->i_atime = *atime;
462 ci->i_time_warp_seq = time_warp_seq;
463 } else if (time_warp_seq == ci->i_time_warp_seq) {
464 /* nobody did utimes(); take the max */
465 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
466 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
467 inode->i_mtime.tv_sec,
468 inode->i_mtime.tv_nsec,
469 mtime->tv_sec, mtime->tv_nsec);
470 inode->i_mtime = *mtime;
472 if (timespec_compare(atime, &inode->i_atime) > 0) {
473 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
474 inode->i_atime.tv_sec,
475 inode->i_atime.tv_nsec,
476 atime->tv_sec, atime->tv_nsec);
477 inode->i_atime = *atime;
479 } else if (issued & CEPH_CAP_FILE_EXCL) {
480 /* we did a utimes(); ignore mds values */
485 /* we have no write caps; whatever the MDS says is true */
486 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
487 inode->i_ctime = *ctime;
488 inode->i_mtime = *mtime;
489 inode->i_atime = *atime;
490 ci->i_time_warp_seq = time_warp_seq;
495 if (warn) /* time_warp_seq shouldn't go backwards */
496 dout("%p mds time_warp_seq %llu < %u\n",
497 inode, time_warp_seq, ci->i_time_warp_seq);
501 * Populate an inode based on info from mds. May be called on new or
504 static int fill_inode(struct inode *inode,
505 struct ceph_mds_reply_info_in *iinfo,
506 struct ceph_mds_reply_dirfrag *dirinfo,
507 struct ceph_mds_session *session,
508 unsigned long ttl_from, int cap_fmode,
509 struct ceph_cap_reservation *caps_reservation)
511 struct ceph_mds_reply_inode *info = iinfo->in;
512 struct ceph_inode_info *ci = ceph_inode(inode);
514 int issued, implemented;
515 struct timespec mtime, atime, ctime;
517 struct ceph_buffer *xattr_blob = NULL;
521 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
522 inode, ceph_vinop(inode), le64_to_cpu(info->version),
526 * prealloc xattr data, if it looks like we'll need it. only
527 * if len > 4 (meaning there are actually xattrs; the first 4
528 * bytes are the xattr count).
530 if (iinfo->xattr_len > 4) {
531 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
533 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
537 spin_lock(&inode->i_lock);
540 * provided version will be odd if inode value is projected,
541 * even if stable. skip the update if we have a newer info
542 * (e.g., due to inode info racing form multiple MDSs), or if
543 * we are getting projected (unstable) inode info.
545 if (le64_to_cpu(info->version) > 0 &&
546 (ci->i_version & ~1) > le64_to_cpu(info->version))
549 issued = __ceph_caps_issued(ci, &implemented);
550 issued |= implemented | __ceph_caps_dirty(ci);
553 ci->i_version = le64_to_cpu(info->version);
555 inode->i_rdev = le32_to_cpu(info->rdev);
557 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
558 inode->i_mode = le32_to_cpu(info->mode);
559 inode->i_uid = le32_to_cpu(info->uid);
560 inode->i_gid = le32_to_cpu(info->gid);
561 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
562 inode->i_uid, inode->i_gid);
565 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
566 inode->i_nlink = le32_to_cpu(info->nlink);
568 /* be careful with mtime, atime, size */
569 ceph_decode_timespec(&atime, &info->atime);
570 ceph_decode_timespec(&mtime, &info->mtime);
571 ceph_decode_timespec(&ctime, &info->ctime);
572 queue_trunc = ceph_fill_file_size(inode, issued,
573 le32_to_cpu(info->truncate_seq),
574 le64_to_cpu(info->truncate_size),
575 le64_to_cpu(info->size));
576 ceph_fill_file_time(inode, issued,
577 le32_to_cpu(info->time_warp_seq),
578 &ctime, &mtime, &atime);
580 ci->i_max_size = le64_to_cpu(info->max_size);
581 ci->i_layout = info->layout;
582 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
585 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
586 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
587 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
588 if (ci->i_xattrs.blob)
589 ceph_buffer_put(ci->i_xattrs.blob);
590 ci->i_xattrs.blob = xattr_blob;
592 memcpy(ci->i_xattrs.blob->vec.iov_base,
593 iinfo->xattr_data, iinfo->xattr_len);
594 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
597 inode->i_mapping->a_ops = &ceph_aops;
598 inode->i_mapping->backing_dev_info =
599 &ceph_client(inode->i_sb)->backing_dev_info;
601 switch (inode->i_mode & S_IFMT) {
606 init_special_inode(inode, inode->i_mode, inode->i_rdev);
607 inode->i_op = &ceph_file_iops;
610 inode->i_op = &ceph_file_iops;
611 inode->i_fop = &ceph_file_fops;
614 inode->i_op = &ceph_symlink_iops;
615 if (!ci->i_symlink) {
616 int symlen = iinfo->symlink_len;
619 BUG_ON(symlen != inode->i_size);
620 spin_unlock(&inode->i_lock);
623 sym = kmalloc(symlen+1, GFP_NOFS);
626 memcpy(sym, iinfo->symlink, symlen);
629 spin_lock(&inode->i_lock);
633 kfree(sym); /* lost a race */
637 inode->i_op = &ceph_dir_iops;
638 inode->i_fop = &ceph_dir_fops;
640 ci->i_files = le64_to_cpu(info->files);
641 ci->i_subdirs = le64_to_cpu(info->subdirs);
642 ci->i_rbytes = le64_to_cpu(info->rbytes);
643 ci->i_rfiles = le64_to_cpu(info->rfiles);
644 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
645 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
647 /* set dir completion flag? */
648 if (ci->i_files == 0 && ci->i_subdirs == 0 &&
649 ceph_snap(inode) == CEPH_NOSNAP &&
650 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED)) {
651 dout(" marking %p complete (empty)\n", inode);
652 ci->i_ceph_flags |= CEPH_I_COMPLETE;
653 ci->i_max_offset = 2;
656 /* it may be better to set st_size in getattr instead? */
657 if (ceph_test_opt(ceph_client(inode->i_sb), RBYTES))
658 inode->i_size = ci->i_rbytes;
661 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
662 ceph_vinop(inode), inode->i_mode);
666 spin_unlock(&inode->i_lock);
668 /* queue truncate if we saw i_size decrease */
670 if (queue_work(ceph_client(inode->i_sb)->trunc_wq,
671 &ci->i_vmtruncate_work))
674 /* populate frag tree */
675 /* FIXME: move me up, if/when version reflects fragtree changes */
676 nsplits = le32_to_cpu(info->fragtree.nsplits);
677 mutex_lock(&ci->i_fragtree_mutex);
678 for (i = 0; i < nsplits; i++) {
679 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
680 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
684 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
685 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
687 mutex_unlock(&ci->i_fragtree_mutex);
689 /* were we issued a capability? */
690 if (info->cap.caps) {
691 if (ceph_snap(inode) == CEPH_NOSNAP) {
692 ceph_add_cap(inode, session,
693 le64_to_cpu(info->cap.cap_id),
695 le32_to_cpu(info->cap.caps),
696 le32_to_cpu(info->cap.wanted),
697 le32_to_cpu(info->cap.seq),
698 le32_to_cpu(info->cap.mseq),
699 le64_to_cpu(info->cap.realm),
703 spin_lock(&inode->i_lock);
704 dout(" %p got snap_caps %s\n", inode,
705 ceph_cap_string(le32_to_cpu(info->cap.caps)));
706 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
708 __ceph_get_fmode(ci, cap_fmode);
709 spin_unlock(&inode->i_lock);
713 /* update delegation info? */
715 ceph_fill_dirfrag(inode, dirinfo);
721 ceph_buffer_put(xattr_blob);
726 * caller should hold session s_mutex.
728 static void update_dentry_lease(struct dentry *dentry,
729 struct ceph_mds_reply_lease *lease,
730 struct ceph_mds_session *session,
731 unsigned long from_time)
733 struct ceph_dentry_info *di = ceph_dentry(dentry);
734 long unsigned duration = le32_to_cpu(lease->duration_ms);
735 long unsigned ttl = from_time + (duration * HZ) / 1000;
736 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
739 /* only track leases on regular dentries */
740 if (dentry->d_op != &ceph_dentry_ops)
743 spin_lock(&dentry->d_lock);
744 dout("update_dentry_lease %p mask %d duration %lu ms ttl %lu\n",
745 dentry, le16_to_cpu(lease->mask), duration, ttl);
747 /* make lease_rdcache_gen match directory */
748 dir = dentry->d_parent->d_inode;
749 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
751 if (lease->mask == 0)
754 if (di->lease_gen == session->s_cap_gen &&
755 time_before(ttl, dentry->d_time))
756 goto out_unlock; /* we already have a newer lease. */
758 if (di->lease_session && di->lease_session != session)
761 ceph_dentry_lru_touch(dentry);
763 if (!di->lease_session)
764 di->lease_session = ceph_get_mds_session(session);
765 di->lease_gen = session->s_cap_gen;
766 di->lease_seq = le32_to_cpu(lease->seq);
767 di->lease_renew_after = half_ttl;
768 di->lease_renew_from = 0;
769 dentry->d_time = ttl;
771 spin_unlock(&dentry->d_lock);
776 * splice a dentry to an inode.
777 * caller must hold directory i_mutex for this to be safe.
779 * we will only rehash the resulting dentry if @prehash is
780 * true; @prehash will be set to false (for the benefit of
781 * the caller) if we fail.
783 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
786 struct dentry *realdn;
788 /* dn must be unhashed */
791 realdn = d_materialise_unique(dn, in);
792 if (IS_ERR(realdn)) {
793 pr_err("splice_dentry error %p inode %p ino %llx.%llx\n",
794 dn, in, ceph_vinop(in));
796 *prehash = false; /* don't rehash on error */
797 dn = realdn; /* note realdn contains the error */
800 dout("dn %p (%d) spliced with %p (%d) "
801 "inode %p ino %llx.%llx\n",
802 dn, atomic_read(&dn->d_count),
803 realdn, atomic_read(&realdn->d_count),
804 realdn->d_inode, ceph_vinop(realdn->d_inode));
808 BUG_ON(!ceph_dentry(dn));
810 dout("dn %p attached to %p ino %llx.%llx\n",
811 dn, dn->d_inode, ceph_vinop(dn->d_inode));
813 if ((!prehash || *prehash) && d_unhashed(dn))
820 * Set dentry's directory position based on the current dir's max, and
821 * order it in d_subdirs, so that dcache_readdir behaves.
823 static void ceph_set_dentry_offset(struct dentry *dn)
825 struct dentry *dir = dn->d_parent;
826 struct inode *inode = dn->d_parent->d_inode;
827 struct ceph_dentry_info *di;
831 di = ceph_dentry(dn);
833 spin_lock(&inode->i_lock);
834 di->offset = ceph_inode(inode)->i_max_offset++;
835 spin_unlock(&inode->i_lock);
837 spin_lock(&dcache_lock);
838 spin_lock(&dn->d_lock);
839 list_move_tail(&dir->d_subdirs, &dn->d_u.d_child);
840 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
841 dn->d_u.d_child.prev, dn->d_u.d_child.next);
842 spin_unlock(&dn->d_lock);
843 spin_unlock(&dcache_lock);
847 * Incorporate results into the local cache. This is either just
848 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
851 * A reply may contain
852 * a directory inode along with a dentry.
853 * and/or a target inode
855 * Called with snap_rwsem (read).
857 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
858 struct ceph_mds_session *session)
860 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
861 struct inode *in = NULL;
862 struct ceph_mds_reply_inode *ininfo;
863 struct ceph_vino vino;
867 dout("fill_trace %p is_dentry %d is_target %d\n", req,
868 rinfo->head->is_dentry, rinfo->head->is_target);
874 * If we resend completed ops to a recovering mds, we get no
875 * trace. Since that is very rare, pretend this is the case
876 * to ensure the 'no trace' handlers in the callers behave.
878 * Fill in inodes unconditionally to avoid breaking cap
881 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
882 pr_info("fill_trace faking empty trace on %lld %s\n",
883 req->r_tid, ceph_mds_op_name(rinfo->head->op));
884 if (rinfo->head->is_dentry) {
885 rinfo->head->is_dentry = 0;
886 err = fill_inode(req->r_locked_dir,
887 &rinfo->diri, rinfo->dirfrag,
888 session, req->r_request_started, -1);
890 if (rinfo->head->is_target) {
891 rinfo->head->is_target = 0;
892 ininfo = rinfo->targeti.in;
893 vino.ino = le64_to_cpu(ininfo->ino);
894 vino.snap = le64_to_cpu(ininfo->snapid);
895 in = ceph_get_inode(sb, vino);
896 err = fill_inode(in, &rinfo->targeti, NULL,
897 session, req->r_request_started,
904 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
905 dout("fill_trace reply is empty!\n");
906 if (rinfo->head->result == 0 && req->r_locked_dir) {
907 struct ceph_inode_info *ci =
908 ceph_inode(req->r_locked_dir);
909 dout(" clearing %p complete (empty trace)\n",
911 ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
912 ci->i_release_count++;
917 if (rinfo->head->is_dentry) {
918 struct inode *dir = req->r_locked_dir;
920 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
921 session, req->r_request_started, -1,
922 &req->r_caps_reservation);
927 if (rinfo->head->is_dentry && !req->r_aborted) {
929 * lookup link rename : null -> possibly existing inode
930 * mknod symlink mkdir : null -> new inode
931 * unlink : linked -> null
933 struct inode *dir = req->r_locked_dir;
934 struct dentry *dn = req->r_dentry;
935 bool have_dir_cap, have_lease;
939 BUG_ON(dn->d_parent->d_inode != dir);
940 BUG_ON(ceph_ino(dir) !=
941 le64_to_cpu(rinfo->diri.in->ino));
942 BUG_ON(ceph_snap(dir) !=
943 le64_to_cpu(rinfo->diri.in->snapid));
945 /* do we have a lease on the whole dir? */
947 (le32_to_cpu(rinfo->diri.in->cap.caps) &
948 CEPH_CAP_FILE_SHARED);
950 /* do we have a dn lease? */
951 have_lease = have_dir_cap ||
952 (le16_to_cpu(rinfo->dlease->mask) &
956 dout("fill_trace no dentry lease or dir cap\n");
959 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
960 dout(" src %p '%.*s' dst %p '%.*s'\n",
962 req->r_old_dentry->d_name.len,
963 req->r_old_dentry->d_name.name,
964 dn, dn->d_name.len, dn->d_name.name);
965 dout("fill_trace doing d_move %p -> %p\n",
966 req->r_old_dentry, dn);
967 d_move(req->r_old_dentry, dn);
968 dout(" src %p '%.*s' dst %p '%.*s'\n",
970 req->r_old_dentry->d_name.len,
971 req->r_old_dentry->d_name.name,
972 dn, dn->d_name.len, dn->d_name.name);
973 /* ensure target dentry is invalidated, despite
974 rehashing bug in vfs_rename_dir */
975 dn->d_time = jiffies;
976 ceph_dentry(dn)->lease_shared_gen = 0;
977 /* take overwritten dentry's readdir offset */
978 ceph_dentry(req->r_old_dentry)->offset =
979 ceph_dentry(dn)->offset;
980 dn = req->r_old_dentry; /* use old_dentry */
985 if (!rinfo->head->is_target) {
986 dout("fill_trace null dentry\n");
988 dout("d_delete %p\n", dn);
991 dout("d_instantiate %p NULL\n", dn);
992 d_instantiate(dn, NULL);
993 if (have_lease && d_unhashed(dn))
995 update_dentry_lease(dn, rinfo->dlease,
997 req->r_request_started);
1002 /* attach proper inode */
1003 ininfo = rinfo->targeti.in;
1004 vino.ino = le64_to_cpu(ininfo->ino);
1005 vino.snap = le64_to_cpu(ininfo->snapid);
1007 in = ceph_get_inode(sb, vino);
1009 pr_err("fill_trace bad get_inode "
1010 "%llx.%llx\n", vino.ino, vino.snap);
1015 dn = splice_dentry(dn, in, &have_lease);
1020 req->r_dentry = dn; /* may have spliced */
1021 ceph_set_dentry_offset(dn);
1023 } else if (ceph_ino(in) == vino.ino &&
1024 ceph_snap(in) == vino.snap) {
1027 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1028 dn, in, ceph_ino(in), ceph_snap(in),
1029 vino.ino, vino.snap);
1035 update_dentry_lease(dn, rinfo->dlease, session,
1036 req->r_request_started);
1037 dout(" final dn %p\n", dn);
1039 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1040 req->r_op == CEPH_MDS_OP_MKSNAP) {
1041 struct dentry *dn = req->r_dentry;
1043 /* fill out a snapdir LOOKUPSNAP dentry */
1045 BUG_ON(!req->r_locked_dir);
1046 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1047 ininfo = rinfo->targeti.in;
1048 vino.ino = le64_to_cpu(ininfo->ino);
1049 vino.snap = le64_to_cpu(ininfo->snapid);
1050 in = ceph_get_inode(sb, vino);
1052 pr_err("fill_inode get_inode badness %llx.%llx\n",
1053 vino.ino, vino.snap);
1058 dout(" linking snapped dir %p to dn %p\n", in, dn);
1059 dn = splice_dentry(dn, in, NULL);
1064 ceph_set_dentry_offset(dn);
1065 req->r_dentry = dn; /* may have spliced */
1067 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1070 if (rinfo->head->is_target) {
1071 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1072 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1074 if (in == NULL || ceph_ino(in) != vino.ino ||
1075 ceph_snap(in) != vino.snap) {
1076 in = ceph_get_inode(sb, vino);
1082 req->r_target_inode = in;
1084 err = fill_inode(in,
1085 &rinfo->targeti, NULL,
1086 session, req->r_request_started,
1087 (le32_to_cpu(rinfo->head->result) == 0) ?
1089 &req->r_caps_reservation);
1091 pr_err("fill_inode badness %p %llx.%llx\n",
1092 in, ceph_vinop(in));
1098 dout("fill_trace done err=%d\n", err);
1103 * Prepopulate our cache with readdir results, leases, etc.
1105 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1106 struct ceph_mds_session *session)
1108 struct dentry *parent = req->r_dentry;
1109 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1114 struct inode *snapdir = NULL;
1115 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1116 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1117 struct ceph_dentry_info *di;
1119 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1120 snapdir = ceph_get_snapdir(parent->d_inode);
1121 parent = d_find_alias(snapdir);
1122 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1123 rinfo->dir_nr, parent);
1125 dout("readdir_prepopulate %d items under dn %p\n",
1126 rinfo->dir_nr, parent);
1128 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1131 for (i = 0; i < rinfo->dir_nr; i++) {
1132 struct ceph_vino vino;
1134 dname.name = rinfo->dir_dname[i];
1135 dname.len = rinfo->dir_dname_len[i];
1136 dname.hash = full_name_hash(dname.name, dname.len);
1138 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1139 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1142 dn = d_lookup(parent, &dname);
1143 dout("d_lookup on parent=%p name=%.*s got %p\n",
1144 parent, dname.len, dname.name, dn);
1147 dn = d_alloc(parent, &dname);
1148 dout("d_alloc %p '%.*s' = %p\n", parent,
1149 dname.len, dname.name, dn);
1151 dout("d_alloc badness\n");
1155 err = ceph_init_dentry(dn);
1158 } else if (dn->d_inode &&
1159 (ceph_ino(dn->d_inode) != vino.ino ||
1160 ceph_snap(dn->d_inode) != vino.snap)) {
1161 dout(" dn %p points to wrong inode %p\n",
1167 /* reorder parent's d_subdirs */
1168 spin_lock(&dcache_lock);
1169 spin_lock(&dn->d_lock);
1170 list_move(&dn->d_u.d_child, &parent->d_subdirs);
1171 spin_unlock(&dn->d_lock);
1172 spin_unlock(&dcache_lock);
1176 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1182 in = ceph_get_inode(parent->d_sb, vino);
1184 dout("new_inode badness\n");
1190 dn = splice_dentry(dn, in, NULL);
1193 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1194 req->r_request_started, -1,
1195 &req->r_caps_reservation) < 0) {
1196 pr_err("fill_inode badness on %p\n", in);
1200 update_dentry_lease(dn, rinfo->dir_dlease[i],
1201 req->r_session, req->r_request_started);
1204 req->r_did_prepopulate = true;
1211 dout("readdir_prepopulate done\n");
1215 int ceph_inode_set_size(struct inode *inode, loff_t size)
1217 struct ceph_inode_info *ci = ceph_inode(inode);
1220 spin_lock(&inode->i_lock);
1221 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1222 inode->i_size = size;
1223 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1225 /* tell the MDS if we are approaching max_size */
1226 if ((size << 1) >= ci->i_max_size &&
1227 (ci->i_reported_size << 1) < ci->i_max_size)
1230 spin_unlock(&inode->i_lock);
1235 * Write back inode data in a worker thread. (This can't be done
1236 * in the message handler context.)
1238 void ceph_inode_writeback(struct work_struct *work)
1240 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1242 struct inode *inode = &ci->vfs_inode;
1244 dout("writeback %p\n", inode);
1245 filemap_fdatawrite(&inode->i_data);
1250 * Invalidate inode pages in a worker thread. (This can't be done
1251 * in the message handler context.)
1253 static void ceph_inode_invalidate_pages(struct work_struct *work)
1255 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1257 struct inode *inode = &ci->vfs_inode;
1261 spin_lock(&inode->i_lock);
1262 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1263 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1264 if (ci->i_rdcache_gen == 0 ||
1265 ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1266 BUG_ON(ci->i_rdcache_revoking > ci->i_rdcache_gen);
1268 ci->i_rdcache_revoking = 0;
1269 spin_unlock(&inode->i_lock);
1272 orig_gen = ci->i_rdcache_gen;
1273 spin_unlock(&inode->i_lock);
1275 truncate_inode_pages(&inode->i_data, 0);
1277 spin_lock(&inode->i_lock);
1278 if (orig_gen == ci->i_rdcache_gen) {
1279 dout("invalidate_pages %p gen %d successful\n", inode,
1281 ci->i_rdcache_gen = 0;
1282 ci->i_rdcache_revoking = 0;
1285 dout("invalidate_pages %p gen %d raced, gen now %d\n",
1286 inode, orig_gen, ci->i_rdcache_gen);
1288 spin_unlock(&inode->i_lock);
1291 ceph_check_caps(ci, 0, NULL);
1298 * called by trunc_wq; take i_mutex ourselves
1300 * We also truncate in a separate thread as well.
1302 void ceph_vmtruncate_work(struct work_struct *work)
1304 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1306 struct inode *inode = &ci->vfs_inode;
1308 dout("vmtruncate_work %p\n", inode);
1309 mutex_lock(&inode->i_mutex);
1310 __ceph_do_pending_vmtruncate(inode);
1311 mutex_unlock(&inode->i_mutex);
1316 * called with i_mutex held.
1318 * Make sure any pending truncation is applied before doing anything
1319 * that may depend on it.
1321 void __ceph_do_pending_vmtruncate(struct inode *inode)
1323 struct ceph_inode_info *ci = ceph_inode(inode);
1325 int wrbuffer_refs, wake = 0;
1328 spin_lock(&inode->i_lock);
1329 if (ci->i_truncate_pending == 0) {
1330 dout("__do_pending_vmtruncate %p none pending\n", inode);
1331 spin_unlock(&inode->i_lock);
1336 * make sure any dirty snapped pages are flushed before we
1337 * possibly truncate them.. so write AND block!
1339 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1340 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1342 spin_unlock(&inode->i_lock);
1343 filemap_write_and_wait_range(&inode->i_data, 0,
1344 inode->i_sb->s_maxbytes);
1348 to = ci->i_truncate_size;
1349 wrbuffer_refs = ci->i_wrbuffer_ref;
1350 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1351 ci->i_truncate_pending, to);
1352 spin_unlock(&inode->i_lock);
1354 truncate_inode_pages(inode->i_mapping, to);
1356 spin_lock(&inode->i_lock);
1357 ci->i_truncate_pending--;
1358 if (ci->i_truncate_pending == 0)
1360 spin_unlock(&inode->i_lock);
1362 if (wrbuffer_refs == 0)
1363 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1365 wake_up(&ci->i_cap_wq);
1372 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1374 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1375 nd_set_link(nd, ci->i_symlink);
1379 static const struct inode_operations ceph_symlink_iops = {
1380 .readlink = generic_readlink,
1381 .follow_link = ceph_sym_follow_link,
1387 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1389 struct inode *inode = dentry->d_inode;
1390 struct ceph_inode_info *ci = ceph_inode(inode);
1391 struct inode *parent_inode = dentry->d_parent->d_inode;
1392 const unsigned int ia_valid = attr->ia_valid;
1393 struct ceph_mds_request *req;
1394 struct ceph_mds_client *mdsc = &ceph_client(dentry->d_sb)->mdsc;
1396 int release = 0, dirtied = 0;
1399 int queue_trunc = 0;
1401 if (ceph_snap(inode) != CEPH_NOSNAP)
1404 __ceph_do_pending_vmtruncate(inode);
1406 err = inode_change_ok(inode, attr);
1410 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1413 return PTR_ERR(req);
1415 spin_lock(&inode->i_lock);
1416 issued = __ceph_caps_issued(ci, NULL);
1417 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1419 if (ia_valid & ATTR_UID) {
1420 dout("setattr %p uid %d -> %d\n", inode,
1421 inode->i_uid, attr->ia_uid);
1422 if (issued & CEPH_CAP_AUTH_EXCL) {
1423 inode->i_uid = attr->ia_uid;
1424 dirtied |= CEPH_CAP_AUTH_EXCL;
1425 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1426 attr->ia_uid != inode->i_uid) {
1427 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1428 mask |= CEPH_SETATTR_UID;
1429 release |= CEPH_CAP_AUTH_SHARED;
1432 if (ia_valid & ATTR_GID) {
1433 dout("setattr %p gid %d -> %d\n", inode,
1434 inode->i_gid, attr->ia_gid);
1435 if (issued & CEPH_CAP_AUTH_EXCL) {
1436 inode->i_gid = attr->ia_gid;
1437 dirtied |= CEPH_CAP_AUTH_EXCL;
1438 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1439 attr->ia_gid != inode->i_gid) {
1440 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1441 mask |= CEPH_SETATTR_GID;
1442 release |= CEPH_CAP_AUTH_SHARED;
1445 if (ia_valid & ATTR_MODE) {
1446 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1448 if (issued & CEPH_CAP_AUTH_EXCL) {
1449 inode->i_mode = attr->ia_mode;
1450 dirtied |= CEPH_CAP_AUTH_EXCL;
1451 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1452 attr->ia_mode != inode->i_mode) {
1453 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1454 mask |= CEPH_SETATTR_MODE;
1455 release |= CEPH_CAP_AUTH_SHARED;
1459 if (ia_valid & ATTR_ATIME) {
1460 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1461 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1462 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1463 if (issued & CEPH_CAP_FILE_EXCL) {
1464 ci->i_time_warp_seq++;
1465 inode->i_atime = attr->ia_atime;
1466 dirtied |= CEPH_CAP_FILE_EXCL;
1467 } else if ((issued & CEPH_CAP_FILE_WR) &&
1468 timespec_compare(&inode->i_atime,
1469 &attr->ia_atime) < 0) {
1470 inode->i_atime = attr->ia_atime;
1471 dirtied |= CEPH_CAP_FILE_WR;
1472 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1473 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1474 ceph_encode_timespec(&req->r_args.setattr.atime,
1476 mask |= CEPH_SETATTR_ATIME;
1477 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1481 if (ia_valid & ATTR_MTIME) {
1482 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1483 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1484 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1485 if (issued & CEPH_CAP_FILE_EXCL) {
1486 ci->i_time_warp_seq++;
1487 inode->i_mtime = attr->ia_mtime;
1488 dirtied |= CEPH_CAP_FILE_EXCL;
1489 } else if ((issued & CEPH_CAP_FILE_WR) &&
1490 timespec_compare(&inode->i_mtime,
1491 &attr->ia_mtime) < 0) {
1492 inode->i_mtime = attr->ia_mtime;
1493 dirtied |= CEPH_CAP_FILE_WR;
1494 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1495 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1496 ceph_encode_timespec(&req->r_args.setattr.mtime,
1498 mask |= CEPH_SETATTR_MTIME;
1499 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1503 if (ia_valid & ATTR_SIZE) {
1504 dout("setattr %p size %lld -> %lld\n", inode,
1505 inode->i_size, attr->ia_size);
1506 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1510 if ((issued & CEPH_CAP_FILE_EXCL) &&
1511 attr->ia_size > inode->i_size) {
1512 inode->i_size = attr->ia_size;
1513 if (attr->ia_size < inode->i_size) {
1514 ci->i_truncate_size = attr->ia_size;
1515 ci->i_truncate_pending++;
1519 (attr->ia_size + (1 << 9) - 1) >> 9;
1520 inode->i_ctime = attr->ia_ctime;
1521 ci->i_reported_size = attr->ia_size;
1522 dirtied |= CEPH_CAP_FILE_EXCL;
1523 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1524 attr->ia_size != inode->i_size) {
1525 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1526 req->r_args.setattr.old_size =
1527 cpu_to_le64(inode->i_size);
1528 mask |= CEPH_SETATTR_SIZE;
1529 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1534 /* these do nothing */
1535 if (ia_valid & ATTR_CTIME) {
1536 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1537 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1538 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1539 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1540 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1541 only ? "ctime only" : "ignored");
1542 inode->i_ctime = attr->ia_ctime;
1545 * if kernel wants to dirty ctime but nothing else,
1546 * we need to choose a cap to dirty under, or do
1547 * a almost-no-op setattr
1549 if (issued & CEPH_CAP_AUTH_EXCL)
1550 dirtied |= CEPH_CAP_AUTH_EXCL;
1551 else if (issued & CEPH_CAP_FILE_EXCL)
1552 dirtied |= CEPH_CAP_FILE_EXCL;
1553 else if (issued & CEPH_CAP_XATTR_EXCL)
1554 dirtied |= CEPH_CAP_XATTR_EXCL;
1556 mask |= CEPH_SETATTR_CTIME;
1559 if (ia_valid & ATTR_FILE)
1560 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1563 __ceph_mark_dirty_caps(ci, dirtied);
1564 inode->i_ctime = CURRENT_TIME;
1568 spin_unlock(&inode->i_lock);
1571 __ceph_do_pending_vmtruncate(inode);
1574 req->r_inode = igrab(inode);
1575 req->r_inode_drop = release;
1576 req->r_args.setattr.mask = cpu_to_le32(mask);
1577 req->r_num_caps = 1;
1578 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1580 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1581 ceph_cap_string(dirtied), mask);
1583 ceph_mdsc_put_request(req);
1584 __ceph_do_pending_vmtruncate(inode);
1587 spin_unlock(&inode->i_lock);
1588 ceph_mdsc_put_request(req);
1593 * Verify that we have a lease on the given mask. If not,
1594 * do a getattr against an mds.
1596 int ceph_do_getattr(struct inode *inode, int mask)
1598 struct ceph_client *client = ceph_sb_to_client(inode->i_sb);
1599 struct ceph_mds_client *mdsc = &client->mdsc;
1600 struct ceph_mds_request *req;
1603 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1604 dout("do_getattr inode %p SNAPDIR\n", inode);
1608 dout("do_getattr inode %p mask %s\n", inode, ceph_cap_string(mask));
1609 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1612 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1614 return PTR_ERR(req);
1615 req->r_inode = igrab(inode);
1616 req->r_num_caps = 1;
1617 req->r_args.getattr.mask = cpu_to_le32(mask);
1618 err = ceph_mdsc_do_request(mdsc, NULL, req);
1619 ceph_mdsc_put_request(req);
1620 dout("do_getattr result=%d\n", err);
1626 * Check inode permissions. We verify we have a valid value for
1627 * the AUTH cap, then call the generic handler.
1629 int ceph_permission(struct inode *inode, int mask)
1631 int err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1634 err = generic_permission(inode, mask, NULL);
1639 * Get all attributes. Hopefully somedata we'll have a statlite()
1640 * and can limit the fields we require to be accurate.
1642 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1645 struct inode *inode = dentry->d_inode;
1646 struct ceph_inode_info *ci = ceph_inode(inode);
1649 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1651 generic_fillattr(inode, stat);
1652 stat->ino = inode->i_ino;
1653 if (ceph_snap(inode) != CEPH_NOSNAP)
1654 stat->dev = ceph_snap(inode);
1657 if (S_ISDIR(inode->i_mode)) {
1658 stat->size = ci->i_rbytes;
1660 stat->blksize = 65536;