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);
300 spin_lock_init(&ci->i_ceph_lock);
303 ci->i_time_warp_seq = 0;
304 ci->i_ceph_flags = 0;
305 ci->i_release_count = 0;
306 ci->i_symlink = NULL;
308 memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
310 ci->i_fragtree = RB_ROOT;
311 mutex_init(&ci->i_fragtree_mutex);
313 ci->i_xattrs.blob = NULL;
314 ci->i_xattrs.prealloc_blob = NULL;
315 ci->i_xattrs.dirty = false;
316 ci->i_xattrs.index = RB_ROOT;
317 ci->i_xattrs.count = 0;
318 ci->i_xattrs.names_size = 0;
319 ci->i_xattrs.vals_size = 0;
320 ci->i_xattrs.version = 0;
321 ci->i_xattrs.index_version = 0;
323 ci->i_caps = RB_ROOT;
324 ci->i_auth_cap = NULL;
325 ci->i_dirty_caps = 0;
326 ci->i_flushing_caps = 0;
327 INIT_LIST_HEAD(&ci->i_dirty_item);
328 INIT_LIST_HEAD(&ci->i_flushing_item);
329 ci->i_cap_flush_seq = 0;
330 ci->i_cap_flush_last_tid = 0;
331 memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
332 init_waitqueue_head(&ci->i_cap_wq);
333 ci->i_hold_caps_min = 0;
334 ci->i_hold_caps_max = 0;
335 INIT_LIST_HEAD(&ci->i_cap_delay_list);
336 ci->i_cap_exporting_mds = 0;
337 ci->i_cap_exporting_mseq = 0;
338 ci->i_cap_exporting_issued = 0;
339 INIT_LIST_HEAD(&ci->i_cap_snaps);
340 ci->i_head_snapc = NULL;
343 for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
344 ci->i_nr_by_mode[i] = 0;
346 ci->i_truncate_seq = 0;
347 ci->i_truncate_size = 0;
348 ci->i_truncate_pending = 0;
351 ci->i_reported_size = 0;
352 ci->i_wanted_max_size = 0;
353 ci->i_requested_max_size = 0;
357 ci->i_rdcache_ref = 0;
360 ci->i_wrbuffer_ref = 0;
361 ci->i_wrbuffer_ref_head = 0;
362 ci->i_shared_gen = 0;
363 ci->i_rdcache_gen = 0;
364 ci->i_rdcache_revoking = 0;
366 INIT_LIST_HEAD(&ci->i_unsafe_writes);
367 INIT_LIST_HEAD(&ci->i_unsafe_dirops);
368 spin_lock_init(&ci->i_unsafe_lock);
370 ci->i_snap_realm = NULL;
371 INIT_LIST_HEAD(&ci->i_snap_realm_item);
372 INIT_LIST_HEAD(&ci->i_snap_flush_item);
374 INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
375 INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
377 INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
379 return &ci->vfs_inode;
382 static void ceph_i_callback(struct rcu_head *head)
384 struct inode *inode = container_of(head, struct inode, i_rcu);
385 struct ceph_inode_info *ci = ceph_inode(inode);
387 INIT_LIST_HEAD(&inode->i_dentry);
388 kmem_cache_free(ceph_inode_cachep, ci);
391 void ceph_destroy_inode(struct inode *inode)
393 struct ceph_inode_info *ci = ceph_inode(inode);
394 struct ceph_inode_frag *frag;
397 dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
399 ceph_queue_caps_release(inode);
402 * we may still have a snap_realm reference if there are stray
403 * caps in i_cap_exporting_issued or i_snap_caps.
405 if (ci->i_snap_realm) {
406 struct ceph_mds_client *mdsc =
407 ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
408 struct ceph_snap_realm *realm = ci->i_snap_realm;
410 dout(" dropping residual ref to snap realm %p\n", realm);
411 spin_lock(&realm->inodes_with_caps_lock);
412 list_del_init(&ci->i_snap_realm_item);
413 spin_unlock(&realm->inodes_with_caps_lock);
414 ceph_put_snap_realm(mdsc, realm);
417 kfree(ci->i_symlink);
418 while ((n = rb_first(&ci->i_fragtree)) != NULL) {
419 frag = rb_entry(n, struct ceph_inode_frag, node);
420 rb_erase(n, &ci->i_fragtree);
424 __ceph_destroy_xattrs(ci);
425 if (ci->i_xattrs.blob)
426 ceph_buffer_put(ci->i_xattrs.blob);
427 if (ci->i_xattrs.prealloc_blob)
428 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
430 call_rcu(&inode->i_rcu, ceph_i_callback);
435 * Helpers to fill in size, ctime, mtime, and atime. We have to be
436 * careful because either the client or MDS may have more up to date
437 * info, depending on which capabilities are held, and whether
438 * time_warp_seq or truncate_seq have increased. (Ordinarily, mtime
439 * and size are monotonically increasing, except when utimes() or
440 * truncate() increments the corresponding _seq values.)
442 int ceph_fill_file_size(struct inode *inode, int issued,
443 u32 truncate_seq, u64 truncate_size, u64 size)
445 struct ceph_inode_info *ci = ceph_inode(inode);
448 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
449 (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
450 dout("size %lld -> %llu\n", inode->i_size, size);
451 inode->i_size = size;
452 inode->i_blocks = (size + (1<<9) - 1) >> 9;
453 ci->i_reported_size = size;
454 if (truncate_seq != ci->i_truncate_seq) {
455 dout("truncate_seq %u -> %u\n",
456 ci->i_truncate_seq, truncate_seq);
457 ci->i_truncate_seq = truncate_seq;
459 * If we hold relevant caps, or in the case where we're
460 * not the only client referencing this file and we
461 * don't hold those caps, then we need to check whether
462 * the file is either opened or mmaped
464 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_RD|
465 CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER|
467 CEPH_CAP_FILE_LAZYIO)) ||
468 mapping_mapped(inode->i_mapping) ||
469 __ceph_caps_file_wanted(ci)) {
470 ci->i_truncate_pending++;
475 if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
476 ci->i_truncate_size != truncate_size) {
477 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
479 ci->i_truncate_size = truncate_size;
484 void ceph_fill_file_time(struct inode *inode, int issued,
485 u64 time_warp_seq, struct timespec *ctime,
486 struct timespec *mtime, struct timespec *atime)
488 struct ceph_inode_info *ci = ceph_inode(inode);
491 if (issued & (CEPH_CAP_FILE_EXCL|
493 CEPH_CAP_FILE_BUFFER|
495 CEPH_CAP_XATTR_EXCL)) {
496 if (timespec_compare(ctime, &inode->i_ctime) > 0) {
497 dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
498 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
499 ctime->tv_sec, ctime->tv_nsec);
500 inode->i_ctime = *ctime;
502 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
503 /* the MDS did a utimes() */
504 dout("mtime %ld.%09ld -> %ld.%09ld "
506 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
507 mtime->tv_sec, mtime->tv_nsec,
508 ci->i_time_warp_seq, (int)time_warp_seq);
510 inode->i_mtime = *mtime;
511 inode->i_atime = *atime;
512 ci->i_time_warp_seq = time_warp_seq;
513 } else if (time_warp_seq == ci->i_time_warp_seq) {
514 /* nobody did utimes(); take the max */
515 if (timespec_compare(mtime, &inode->i_mtime) > 0) {
516 dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
517 inode->i_mtime.tv_sec,
518 inode->i_mtime.tv_nsec,
519 mtime->tv_sec, mtime->tv_nsec);
520 inode->i_mtime = *mtime;
522 if (timespec_compare(atime, &inode->i_atime) > 0) {
523 dout("atime %ld.%09ld -> %ld.%09ld inc\n",
524 inode->i_atime.tv_sec,
525 inode->i_atime.tv_nsec,
526 atime->tv_sec, atime->tv_nsec);
527 inode->i_atime = *atime;
529 } else if (issued & CEPH_CAP_FILE_EXCL) {
530 /* we did a utimes(); ignore mds values */
535 /* we have no write|excl caps; whatever the MDS says is true */
536 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
537 inode->i_ctime = *ctime;
538 inode->i_mtime = *mtime;
539 inode->i_atime = *atime;
540 ci->i_time_warp_seq = time_warp_seq;
545 if (warn) /* time_warp_seq shouldn't go backwards */
546 dout("%p mds time_warp_seq %llu < %u\n",
547 inode, time_warp_seq, ci->i_time_warp_seq);
551 * Populate an inode based on info from mds. May be called on new or
554 static int fill_inode(struct inode *inode,
555 struct ceph_mds_reply_info_in *iinfo,
556 struct ceph_mds_reply_dirfrag *dirinfo,
557 struct ceph_mds_session *session,
558 unsigned long ttl_from, int cap_fmode,
559 struct ceph_cap_reservation *caps_reservation)
561 struct ceph_mds_reply_inode *info = iinfo->in;
562 struct ceph_inode_info *ci = ceph_inode(inode);
564 int issued = 0, implemented;
565 int updating_inode = 0;
566 struct timespec mtime, atime, ctime;
568 struct ceph_buffer *xattr_blob = NULL;
572 dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
573 inode, ceph_vinop(inode), le64_to_cpu(info->version),
577 * prealloc xattr data, if it looks like we'll need it. only
578 * if len > 4 (meaning there are actually xattrs; the first 4
579 * bytes are the xattr count).
581 if (iinfo->xattr_len > 4) {
582 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
584 pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
588 spin_lock(&ci->i_ceph_lock);
591 * provided version will be odd if inode value is projected,
592 * even if stable. skip the update if we have newer stable
593 * info (ours>=theirs, e.g. due to racing mds replies), unless
594 * we are getting projected (unstable) info (in which case the
595 * version is odd, and we want ours>theirs).
601 if (le64_to_cpu(info->version) > 0 &&
602 (ci->i_version & ~1) >= le64_to_cpu(info->version))
606 issued = __ceph_caps_issued(ci, &implemented);
607 issued |= implemented | __ceph_caps_dirty(ci);
610 ci->i_version = le64_to_cpu(info->version);
612 inode->i_rdev = le32_to_cpu(info->rdev);
614 if ((issued & CEPH_CAP_AUTH_EXCL) == 0) {
615 inode->i_mode = le32_to_cpu(info->mode);
616 inode->i_uid = le32_to_cpu(info->uid);
617 inode->i_gid = le32_to_cpu(info->gid);
618 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
619 inode->i_uid, inode->i_gid);
622 if ((issued & CEPH_CAP_LINK_EXCL) == 0)
623 set_nlink(inode, le32_to_cpu(info->nlink));
625 /* be careful with mtime, atime, size */
626 ceph_decode_timespec(&atime, &info->atime);
627 ceph_decode_timespec(&mtime, &info->mtime);
628 ceph_decode_timespec(&ctime, &info->ctime);
629 queue_trunc = ceph_fill_file_size(inode, issued,
630 le32_to_cpu(info->truncate_seq),
631 le64_to_cpu(info->truncate_size),
632 le64_to_cpu(info->size));
633 ceph_fill_file_time(inode, issued,
634 le32_to_cpu(info->time_warp_seq),
635 &ctime, &mtime, &atime);
637 /* only update max_size on auth cap */
638 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
639 ci->i_max_size != le64_to_cpu(info->max_size)) {
640 dout("max_size %lld -> %llu\n", ci->i_max_size,
641 le64_to_cpu(info->max_size));
642 ci->i_max_size = le64_to_cpu(info->max_size);
645 ci->i_layout = info->layout;
646 inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
649 /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
650 if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
651 le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
652 if (ci->i_xattrs.blob)
653 ceph_buffer_put(ci->i_xattrs.blob);
654 ci->i_xattrs.blob = xattr_blob;
656 memcpy(ci->i_xattrs.blob->vec.iov_base,
657 iinfo->xattr_data, iinfo->xattr_len);
658 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
662 inode->i_mapping->a_ops = &ceph_aops;
663 inode->i_mapping->backing_dev_info =
664 &ceph_sb_to_client(inode->i_sb)->backing_dev_info;
666 switch (inode->i_mode & S_IFMT) {
671 init_special_inode(inode, inode->i_mode, inode->i_rdev);
672 inode->i_op = &ceph_file_iops;
675 inode->i_op = &ceph_file_iops;
676 inode->i_fop = &ceph_file_fops;
679 inode->i_op = &ceph_symlink_iops;
680 if (!ci->i_symlink) {
681 int symlen = iinfo->symlink_len;
684 BUG_ON(symlen != inode->i_size);
685 spin_unlock(&ci->i_ceph_lock);
688 sym = kmalloc(symlen+1, GFP_NOFS);
691 memcpy(sym, iinfo->symlink, symlen);
694 spin_lock(&ci->i_ceph_lock);
698 kfree(sym); /* lost a race */
702 inode->i_op = &ceph_dir_iops;
703 inode->i_fop = &ceph_dir_fops;
705 ci->i_dir_layout = iinfo->dir_layout;
707 ci->i_files = le64_to_cpu(info->files);
708 ci->i_subdirs = le64_to_cpu(info->subdirs);
709 ci->i_rbytes = le64_to_cpu(info->rbytes);
710 ci->i_rfiles = le64_to_cpu(info->rfiles);
711 ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
712 ceph_decode_timespec(&ci->i_rctime, &info->rctime);
715 pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
716 ceph_vinop(inode), inode->i_mode);
720 spin_unlock(&ci->i_ceph_lock);
722 /* queue truncate if we saw i_size decrease */
724 ceph_queue_vmtruncate(inode);
726 /* populate frag tree */
727 /* FIXME: move me up, if/when version reflects fragtree changes */
728 nsplits = le32_to_cpu(info->fragtree.nsplits);
729 mutex_lock(&ci->i_fragtree_mutex);
730 for (i = 0; i < nsplits; i++) {
731 u32 id = le32_to_cpu(info->fragtree.splits[i].frag);
732 struct ceph_inode_frag *frag = __get_or_create_frag(ci, id);
736 frag->split_by = le32_to_cpu(info->fragtree.splits[i].by);
737 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
739 mutex_unlock(&ci->i_fragtree_mutex);
741 /* were we issued a capability? */
742 if (info->cap.caps) {
743 if (ceph_snap(inode) == CEPH_NOSNAP) {
744 ceph_add_cap(inode, session,
745 le64_to_cpu(info->cap.cap_id),
747 le32_to_cpu(info->cap.caps),
748 le32_to_cpu(info->cap.wanted),
749 le32_to_cpu(info->cap.seq),
750 le32_to_cpu(info->cap.mseq),
751 le64_to_cpu(info->cap.realm),
755 spin_lock(&ci->i_ceph_lock);
756 dout(" %p got snap_caps %s\n", inode,
757 ceph_cap_string(le32_to_cpu(info->cap.caps)));
758 ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
760 __ceph_get_fmode(ci, cap_fmode);
761 spin_unlock(&ci->i_ceph_lock);
763 } else if (cap_fmode >= 0) {
764 pr_warning("mds issued no caps on %llx.%llx\n",
766 __ceph_get_fmode(ci, cap_fmode);
769 /* set dir completion flag? */
770 if (S_ISDIR(inode->i_mode) &&
771 updating_inode && /* didn't jump to no_change */
772 ci->i_files == 0 && ci->i_subdirs == 0 &&
773 ceph_snap(inode) == CEPH_NOSNAP &&
774 (le32_to_cpu(info->cap.caps) & CEPH_CAP_FILE_SHARED) &&
775 (issued & CEPH_CAP_FILE_EXCL) == 0 &&
776 !ceph_dir_test_complete(inode)) {
777 dout(" marking %p complete (empty)\n", inode);
778 ceph_dir_set_complete(inode);
779 ci->i_max_offset = 2;
782 /* update delegation info? */
784 ceph_fill_dirfrag(inode, dirinfo);
790 ceph_buffer_put(xattr_blob);
795 * caller should hold session s_mutex.
797 static void update_dentry_lease(struct dentry *dentry,
798 struct ceph_mds_reply_lease *lease,
799 struct ceph_mds_session *session,
800 unsigned long from_time)
802 struct ceph_dentry_info *di = ceph_dentry(dentry);
803 long unsigned duration = le32_to_cpu(lease->duration_ms);
804 long unsigned ttl = from_time + (duration * HZ) / 1000;
805 long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
808 /* only track leases on regular dentries */
809 if (dentry->d_op != &ceph_dentry_ops)
812 spin_lock(&dentry->d_lock);
813 dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
814 dentry, duration, ttl);
816 /* make lease_rdcache_gen match directory */
817 dir = dentry->d_parent->d_inode;
818 di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
823 if (di->lease_gen == session->s_cap_gen &&
824 time_before(ttl, dentry->d_time))
825 goto out_unlock; /* we already have a newer lease. */
827 if (di->lease_session && di->lease_session != session)
830 ceph_dentry_lru_touch(dentry);
832 if (!di->lease_session)
833 di->lease_session = ceph_get_mds_session(session);
834 di->lease_gen = session->s_cap_gen;
835 di->lease_seq = le32_to_cpu(lease->seq);
836 di->lease_renew_after = half_ttl;
837 di->lease_renew_from = 0;
838 dentry->d_time = ttl;
840 spin_unlock(&dentry->d_lock);
845 * Set dentry's directory position based on the current dir's max, and
846 * order it in d_subdirs, so that dcache_readdir behaves.
848 * Always called under directory's i_mutex.
850 static void ceph_set_dentry_offset(struct dentry *dn)
852 struct dentry *dir = dn->d_parent;
853 struct inode *inode = dir->d_inode;
854 struct ceph_inode_info *ci = ceph_inode(inode);
855 struct ceph_dentry_info *di;
859 di = ceph_dentry(dn);
861 spin_lock(&ci->i_ceph_lock);
862 if (!ceph_dir_test_complete(inode)) {
863 spin_unlock(&ci->i_ceph_lock);
866 di->offset = ceph_inode(inode)->i_max_offset++;
867 spin_unlock(&ci->i_ceph_lock);
869 spin_lock(&dir->d_lock);
870 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
871 list_move(&dn->d_child, &dir->d_subdirs);
872 dout("set_dentry_offset %p %lld (%p %p)\n", dn, di->offset,
873 dn->d_child.prev, dn->d_child.next);
874 spin_unlock(&dn->d_lock);
875 spin_unlock(&dir->d_lock);
879 * splice a dentry to an inode.
880 * caller must hold directory i_mutex for this to be safe.
882 * we will only rehash the resulting dentry if @prehash is
883 * true; @prehash will be set to false (for the benefit of
884 * the caller) if we fail.
886 static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
887 bool *prehash, bool set_offset)
889 struct dentry *realdn;
893 /* dn must be unhashed */
896 realdn = d_materialise_unique(dn, in);
897 if (IS_ERR(realdn)) {
898 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
899 PTR_ERR(realdn), dn, in, ceph_vinop(in));
901 *prehash = false; /* don't rehash on error */
902 dn = realdn; /* note realdn contains the error */
905 dout("dn %p (%d) spliced with %p (%d) "
906 "inode %p ino %llx.%llx\n",
908 realdn, realdn->d_count,
909 realdn->d_inode, ceph_vinop(realdn->d_inode));
913 BUG_ON(!ceph_dentry(dn));
914 dout("dn %p attached to %p ino %llx.%llx\n",
915 dn, dn->d_inode, ceph_vinop(dn->d_inode));
917 if ((!prehash || *prehash) && d_unhashed(dn))
920 ceph_set_dentry_offset(dn);
926 * Incorporate results into the local cache. This is either just
927 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
930 * A reply may contain
931 * a directory inode along with a dentry.
932 * and/or a target inode
934 * Called with snap_rwsem (read).
936 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
937 struct ceph_mds_session *session)
939 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
940 struct inode *in = NULL;
941 struct ceph_mds_reply_inode *ininfo;
942 struct ceph_vino vino;
943 struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
947 dout("fill_trace %p is_dentry %d is_target %d\n", req,
948 rinfo->head->is_dentry, rinfo->head->is_target);
954 * If we resend completed ops to a recovering mds, we get no
955 * trace. Since that is very rare, pretend this is the case
956 * to ensure the 'no trace' handlers in the callers behave.
958 * Fill in inodes unconditionally to avoid breaking cap
961 if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
962 pr_info("fill_trace faking empty trace on %lld %s\n",
963 req->r_tid, ceph_mds_op_name(rinfo->head->op));
964 if (rinfo->head->is_dentry) {
965 rinfo->head->is_dentry = 0;
966 err = fill_inode(req->r_locked_dir,
967 &rinfo->diri, rinfo->dirfrag,
968 session, req->r_request_started, -1);
970 if (rinfo->head->is_target) {
971 rinfo->head->is_target = 0;
972 ininfo = rinfo->targeti.in;
973 vino.ino = le64_to_cpu(ininfo->ino);
974 vino.snap = le64_to_cpu(ininfo->snapid);
975 in = ceph_get_inode(sb, vino);
976 err = fill_inode(in, &rinfo->targeti, NULL,
977 session, req->r_request_started,
984 if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
985 dout("fill_trace reply is empty!\n");
986 if (rinfo->head->result == 0 && req->r_locked_dir)
987 ceph_invalidate_dir_request(req);
991 if (rinfo->head->is_dentry) {
992 struct inode *dir = req->r_locked_dir;
994 err = fill_inode(dir, &rinfo->diri, rinfo->dirfrag,
995 session, req->r_request_started, -1,
996 &req->r_caps_reservation);
1002 * ignore null lease/binding on snapdir ENOENT, or else we
1003 * will have trouble splicing in the virtual snapdir later
1005 if (rinfo->head->is_dentry && !req->r_aborted &&
1006 (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1007 fsc->mount_options->snapdir_name,
1008 req->r_dentry->d_name.len))) {
1010 * lookup link rename : null -> possibly existing inode
1011 * mknod symlink mkdir : null -> new inode
1012 * unlink : linked -> null
1014 struct inode *dir = req->r_locked_dir;
1015 struct dentry *dn = req->r_dentry;
1016 bool have_dir_cap, have_lease;
1020 BUG_ON(dn->d_parent->d_inode != dir);
1021 BUG_ON(ceph_ino(dir) !=
1022 le64_to_cpu(rinfo->diri.in->ino));
1023 BUG_ON(ceph_snap(dir) !=
1024 le64_to_cpu(rinfo->diri.in->snapid));
1026 /* do we have a lease on the whole dir? */
1028 (le32_to_cpu(rinfo->diri.in->cap.caps) &
1029 CEPH_CAP_FILE_SHARED);
1031 /* do we have a dn lease? */
1032 have_lease = have_dir_cap ||
1033 le32_to_cpu(rinfo->dlease->duration_ms);
1035 dout("fill_trace no dentry lease or dir cap\n");
1038 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1039 dout(" src %p '%.*s' dst %p '%.*s'\n",
1041 req->r_old_dentry->d_name.len,
1042 req->r_old_dentry->d_name.name,
1043 dn, dn->d_name.len, dn->d_name.name);
1044 dout("fill_trace doing d_move %p -> %p\n",
1045 req->r_old_dentry, dn);
1047 d_move(req->r_old_dentry, dn);
1048 dout(" src %p '%.*s' dst %p '%.*s'\n",
1050 req->r_old_dentry->d_name.len,
1051 req->r_old_dentry->d_name.name,
1052 dn, dn->d_name.len, dn->d_name.name);
1054 /* ensure target dentry is invalidated, despite
1055 rehashing bug in vfs_rename_dir */
1056 ceph_invalidate_dentry_lease(dn);
1059 * d_move() puts the renamed dentry at the end of
1060 * d_subdirs. We need to assign it an appropriate
1061 * directory offset so we can behave when holding
1064 ceph_set_dentry_offset(req->r_old_dentry);
1065 dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1066 ceph_dentry(req->r_old_dentry)->offset);
1068 dn = req->r_old_dentry; /* use old_dentry */
1073 if (!rinfo->head->is_target) {
1074 dout("fill_trace null dentry\n");
1076 dout("d_delete %p\n", dn);
1079 dout("d_instantiate %p NULL\n", dn);
1080 d_instantiate(dn, NULL);
1081 if (have_lease && d_unhashed(dn))
1083 update_dentry_lease(dn, rinfo->dlease,
1085 req->r_request_started);
1090 /* attach proper inode */
1091 ininfo = rinfo->targeti.in;
1092 vino.ino = le64_to_cpu(ininfo->ino);
1093 vino.snap = le64_to_cpu(ininfo->snapid);
1096 in = ceph_get_inode(sb, vino);
1098 pr_err("fill_trace bad get_inode "
1099 "%llx.%llx\n", vino.ino, vino.snap);
1104 dn = splice_dentry(dn, in, &have_lease, true);
1109 req->r_dentry = dn; /* may have spliced */
1111 } else if (ceph_ino(in) == vino.ino &&
1112 ceph_snap(in) == vino.snap) {
1115 dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1116 dn, in, ceph_ino(in), ceph_snap(in),
1117 vino.ino, vino.snap);
1123 update_dentry_lease(dn, rinfo->dlease, session,
1124 req->r_request_started);
1125 dout(" final dn %p\n", dn);
1127 } else if (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1128 req->r_op == CEPH_MDS_OP_MKSNAP) {
1129 struct dentry *dn = req->r_dentry;
1131 /* fill out a snapdir LOOKUPSNAP dentry */
1133 BUG_ON(!req->r_locked_dir);
1134 BUG_ON(ceph_snap(req->r_locked_dir) != CEPH_SNAPDIR);
1135 ininfo = rinfo->targeti.in;
1136 vino.ino = le64_to_cpu(ininfo->ino);
1137 vino.snap = le64_to_cpu(ininfo->snapid);
1138 in = ceph_get_inode(sb, vino);
1140 pr_err("fill_inode get_inode badness %llx.%llx\n",
1141 vino.ino, vino.snap);
1146 dout(" linking snapped dir %p to dn %p\n", in, dn);
1147 dn = splice_dentry(dn, in, NULL, true);
1152 req->r_dentry = dn; /* may have spliced */
1154 rinfo->head->is_dentry = 1; /* fool notrace handlers */
1157 if (rinfo->head->is_target) {
1158 vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1159 vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1161 if (in == NULL || ceph_ino(in) != vino.ino ||
1162 ceph_snap(in) != vino.snap) {
1163 in = ceph_get_inode(sb, vino);
1169 req->r_target_inode = in;
1171 err = fill_inode(in,
1172 &rinfo->targeti, NULL,
1173 session, req->r_request_started,
1174 (le32_to_cpu(rinfo->head->result) == 0) ?
1176 &req->r_caps_reservation);
1178 pr_err("fill_inode badness %p %llx.%llx\n",
1179 in, ceph_vinop(in));
1185 dout("fill_trace done err=%d\n", err);
1190 * Prepopulate our cache with readdir results, leases, etc.
1192 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1193 struct ceph_mds_session *session)
1195 struct dentry *parent = req->r_dentry;
1196 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1201 struct inode *snapdir = NULL;
1202 struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1203 u64 frag = le32_to_cpu(rhead->args.readdir.frag);
1204 struct ceph_dentry_info *di;
1206 if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1207 snapdir = ceph_get_snapdir(parent->d_inode);
1208 parent = d_find_alias(snapdir);
1209 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1210 rinfo->dir_nr, parent);
1212 dout("readdir_prepopulate %d items under dn %p\n",
1213 rinfo->dir_nr, parent);
1215 ceph_fill_dirfrag(parent->d_inode, rinfo->dir_dir);
1218 for (i = 0; i < rinfo->dir_nr; i++) {
1219 struct ceph_vino vino;
1221 dname.name = rinfo->dir_dname[i];
1222 dname.len = rinfo->dir_dname_len[i];
1223 dname.hash = full_name_hash(dname.name, dname.len);
1225 vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1226 vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1229 dn = d_lookup(parent, &dname);
1230 dout("d_lookup on parent=%p name=%.*s got %p\n",
1231 parent, dname.len, dname.name, dn);
1234 dn = d_alloc(parent, &dname);
1235 dout("d_alloc %p '%.*s' = %p\n", parent,
1236 dname.len, dname.name, dn);
1238 dout("d_alloc badness\n");
1242 err = ceph_init_dentry(dn);
1247 } else if (dn->d_inode &&
1248 (ceph_ino(dn->d_inode) != vino.ino ||
1249 ceph_snap(dn->d_inode) != vino.snap)) {
1250 dout(" dn %p points to wrong inode %p\n",
1256 /* reorder parent's d_subdirs */
1257 spin_lock(&parent->d_lock);
1258 spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1259 list_move(&dn->d_child, &parent->d_subdirs);
1260 spin_unlock(&dn->d_lock);
1261 spin_unlock(&parent->d_lock);
1265 di->offset = ceph_make_fpos(frag, i + req->r_readdir_offset);
1271 in = ceph_get_inode(parent->d_sb, vino);
1273 dout("new_inode badness\n");
1279 dn = splice_dentry(dn, in, NULL, false);
1284 if (fill_inode(in, &rinfo->dir_in[i], NULL, session,
1285 req->r_request_started, -1,
1286 &req->r_caps_reservation) < 0) {
1287 pr_err("fill_inode badness on %p\n", in);
1291 update_dentry_lease(dn, rinfo->dir_dlease[i],
1293 req->r_request_started);
1298 req->r_did_prepopulate = true;
1305 dout("readdir_prepopulate done\n");
1309 int ceph_inode_set_size(struct inode *inode, loff_t size)
1311 struct ceph_inode_info *ci = ceph_inode(inode);
1314 spin_lock(&ci->i_ceph_lock);
1315 dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1316 inode->i_size = size;
1317 inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1319 /* tell the MDS if we are approaching max_size */
1320 if ((size << 1) >= ci->i_max_size &&
1321 (ci->i_reported_size << 1) < ci->i_max_size)
1324 spin_unlock(&ci->i_ceph_lock);
1329 * Write back inode data in a worker thread. (This can't be done
1330 * in the message handler context.)
1332 void ceph_queue_writeback(struct inode *inode)
1335 if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1336 &ceph_inode(inode)->i_wb_work)) {
1337 dout("ceph_queue_writeback %p\n", inode);
1339 dout("ceph_queue_writeback %p failed\n", inode);
1344 static void ceph_writeback_work(struct work_struct *work)
1346 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1348 struct inode *inode = &ci->vfs_inode;
1350 dout("writeback %p\n", inode);
1351 filemap_fdatawrite(&inode->i_data);
1356 * queue an async invalidation
1358 void ceph_queue_invalidate(struct inode *inode)
1361 if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1362 &ceph_inode(inode)->i_pg_inv_work)) {
1363 dout("ceph_queue_invalidate %p\n", inode);
1365 dout("ceph_queue_invalidate %p failed\n", inode);
1371 * Invalidate inode pages in a worker thread. (This can't be done
1372 * in the message handler context.)
1374 static void ceph_invalidate_work(struct work_struct *work)
1376 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1378 struct inode *inode = &ci->vfs_inode;
1382 spin_lock(&ci->i_ceph_lock);
1383 dout("invalidate_pages %p gen %d revoking %d\n", inode,
1384 ci->i_rdcache_gen, ci->i_rdcache_revoking);
1385 if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1387 spin_unlock(&ci->i_ceph_lock);
1390 orig_gen = ci->i_rdcache_gen;
1391 spin_unlock(&ci->i_ceph_lock);
1393 truncate_inode_pages(&inode->i_data, 0);
1395 spin_lock(&ci->i_ceph_lock);
1396 if (orig_gen == ci->i_rdcache_gen &&
1397 orig_gen == ci->i_rdcache_revoking) {
1398 dout("invalidate_pages %p gen %d successful\n", inode,
1400 ci->i_rdcache_revoking--;
1403 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1404 inode, orig_gen, ci->i_rdcache_gen,
1405 ci->i_rdcache_revoking);
1407 spin_unlock(&ci->i_ceph_lock);
1410 ceph_check_caps(ci, 0, NULL);
1417 * called by trunc_wq; take i_mutex ourselves
1419 * We also truncate in a separate thread as well.
1421 static void ceph_vmtruncate_work(struct work_struct *work)
1423 struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1425 struct inode *inode = &ci->vfs_inode;
1427 dout("vmtruncate_work %p\n", inode);
1428 mutex_lock(&inode->i_mutex);
1429 __ceph_do_pending_vmtruncate(inode);
1430 mutex_unlock(&inode->i_mutex);
1435 * Queue an async vmtruncate. If we fail to queue work, we will handle
1436 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1438 void ceph_queue_vmtruncate(struct inode *inode)
1440 struct ceph_inode_info *ci = ceph_inode(inode);
1443 if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1444 &ci->i_vmtruncate_work)) {
1445 dout("ceph_queue_vmtruncate %p\n", inode);
1447 dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1448 inode, ci->i_truncate_pending);
1454 * called with i_mutex held.
1456 * Make sure any pending truncation is applied before doing anything
1457 * that may depend on it.
1459 void __ceph_do_pending_vmtruncate(struct inode *inode)
1461 struct ceph_inode_info *ci = ceph_inode(inode);
1463 int wrbuffer_refs, wake = 0;
1466 spin_lock(&ci->i_ceph_lock);
1467 if (ci->i_truncate_pending == 0) {
1468 dout("__do_pending_vmtruncate %p none pending\n", inode);
1469 spin_unlock(&ci->i_ceph_lock);
1474 * make sure any dirty snapped pages are flushed before we
1475 * possibly truncate them.. so write AND block!
1477 if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1478 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1480 spin_unlock(&ci->i_ceph_lock);
1481 filemap_write_and_wait_range(&inode->i_data, 0,
1482 inode->i_sb->s_maxbytes);
1486 to = ci->i_truncate_size;
1487 wrbuffer_refs = ci->i_wrbuffer_ref;
1488 dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1489 ci->i_truncate_pending, to);
1490 spin_unlock(&ci->i_ceph_lock);
1492 truncate_inode_pages(inode->i_mapping, to);
1494 spin_lock(&ci->i_ceph_lock);
1495 ci->i_truncate_pending--;
1496 if (ci->i_truncate_pending == 0)
1498 spin_unlock(&ci->i_ceph_lock);
1500 if (wrbuffer_refs == 0)
1501 ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1503 wake_up_all(&ci->i_cap_wq);
1510 static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1512 struct ceph_inode_info *ci = ceph_inode(dentry->d_inode);
1513 nd_set_link(nd, ci->i_symlink);
1517 static const struct inode_operations ceph_symlink_iops = {
1518 .readlink = generic_readlink,
1519 .follow_link = ceph_sym_follow_link,
1525 int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1527 struct inode *inode = dentry->d_inode;
1528 struct ceph_inode_info *ci = ceph_inode(inode);
1529 struct inode *parent_inode;
1530 const unsigned int ia_valid = attr->ia_valid;
1531 struct ceph_mds_request *req;
1532 struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1534 int release = 0, dirtied = 0;
1537 int inode_dirty_flags = 0;
1539 if (ceph_snap(inode) != CEPH_NOSNAP)
1542 __ceph_do_pending_vmtruncate(inode);
1544 err = inode_change_ok(inode, attr);
1548 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1551 return PTR_ERR(req);
1553 spin_lock(&ci->i_ceph_lock);
1554 issued = __ceph_caps_issued(ci, NULL);
1555 dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1557 if (ia_valid & ATTR_UID) {
1558 dout("setattr %p uid %d -> %d\n", inode,
1559 inode->i_uid, attr->ia_uid);
1560 if (issued & CEPH_CAP_AUTH_EXCL) {
1561 inode->i_uid = attr->ia_uid;
1562 dirtied |= CEPH_CAP_AUTH_EXCL;
1563 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1564 attr->ia_uid != inode->i_uid) {
1565 req->r_args.setattr.uid = cpu_to_le32(attr->ia_uid);
1566 mask |= CEPH_SETATTR_UID;
1567 release |= CEPH_CAP_AUTH_SHARED;
1570 if (ia_valid & ATTR_GID) {
1571 dout("setattr %p gid %d -> %d\n", inode,
1572 inode->i_gid, attr->ia_gid);
1573 if (issued & CEPH_CAP_AUTH_EXCL) {
1574 inode->i_gid = attr->ia_gid;
1575 dirtied |= CEPH_CAP_AUTH_EXCL;
1576 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1577 attr->ia_gid != inode->i_gid) {
1578 req->r_args.setattr.gid = cpu_to_le32(attr->ia_gid);
1579 mask |= CEPH_SETATTR_GID;
1580 release |= CEPH_CAP_AUTH_SHARED;
1583 if (ia_valid & ATTR_MODE) {
1584 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1586 if (issued & CEPH_CAP_AUTH_EXCL) {
1587 inode->i_mode = attr->ia_mode;
1588 dirtied |= CEPH_CAP_AUTH_EXCL;
1589 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1590 attr->ia_mode != inode->i_mode) {
1591 req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1592 mask |= CEPH_SETATTR_MODE;
1593 release |= CEPH_CAP_AUTH_SHARED;
1597 if (ia_valid & ATTR_ATIME) {
1598 dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1599 inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1600 attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1601 if (issued & CEPH_CAP_FILE_EXCL) {
1602 ci->i_time_warp_seq++;
1603 inode->i_atime = attr->ia_atime;
1604 dirtied |= CEPH_CAP_FILE_EXCL;
1605 } else if ((issued & CEPH_CAP_FILE_WR) &&
1606 timespec_compare(&inode->i_atime,
1607 &attr->ia_atime) < 0) {
1608 inode->i_atime = attr->ia_atime;
1609 dirtied |= CEPH_CAP_FILE_WR;
1610 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1611 !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1612 ceph_encode_timespec(&req->r_args.setattr.atime,
1614 mask |= CEPH_SETATTR_ATIME;
1615 release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1619 if (ia_valid & ATTR_MTIME) {
1620 dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1621 inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1622 attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1623 if (issued & CEPH_CAP_FILE_EXCL) {
1624 ci->i_time_warp_seq++;
1625 inode->i_mtime = attr->ia_mtime;
1626 dirtied |= CEPH_CAP_FILE_EXCL;
1627 } else if ((issued & CEPH_CAP_FILE_WR) &&
1628 timespec_compare(&inode->i_mtime,
1629 &attr->ia_mtime) < 0) {
1630 inode->i_mtime = attr->ia_mtime;
1631 dirtied |= CEPH_CAP_FILE_WR;
1632 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1633 !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1634 ceph_encode_timespec(&req->r_args.setattr.mtime,
1636 mask |= CEPH_SETATTR_MTIME;
1637 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1641 if (ia_valid & ATTR_SIZE) {
1642 dout("setattr %p size %lld -> %lld\n", inode,
1643 inode->i_size, attr->ia_size);
1644 if (attr->ia_size > inode->i_sb->s_maxbytes) {
1648 if ((issued & CEPH_CAP_FILE_EXCL) &&
1649 attr->ia_size > inode->i_size) {
1650 inode->i_size = attr->ia_size;
1652 (attr->ia_size + (1 << 9) - 1) >> 9;
1653 inode->i_ctime = attr->ia_ctime;
1654 ci->i_reported_size = attr->ia_size;
1655 dirtied |= CEPH_CAP_FILE_EXCL;
1656 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1657 attr->ia_size != inode->i_size) {
1658 req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1659 req->r_args.setattr.old_size =
1660 cpu_to_le64(inode->i_size);
1661 mask |= CEPH_SETATTR_SIZE;
1662 release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1667 /* these do nothing */
1668 if (ia_valid & ATTR_CTIME) {
1669 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1670 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1671 dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1672 inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1673 attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1674 only ? "ctime only" : "ignored");
1675 inode->i_ctime = attr->ia_ctime;
1678 * if kernel wants to dirty ctime but nothing else,
1679 * we need to choose a cap to dirty under, or do
1680 * a almost-no-op setattr
1682 if (issued & CEPH_CAP_AUTH_EXCL)
1683 dirtied |= CEPH_CAP_AUTH_EXCL;
1684 else if (issued & CEPH_CAP_FILE_EXCL)
1685 dirtied |= CEPH_CAP_FILE_EXCL;
1686 else if (issued & CEPH_CAP_XATTR_EXCL)
1687 dirtied |= CEPH_CAP_XATTR_EXCL;
1689 mask |= CEPH_SETATTR_CTIME;
1692 if (ia_valid & ATTR_FILE)
1693 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1696 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1697 inode->i_ctime = CURRENT_TIME;
1701 spin_unlock(&ci->i_ceph_lock);
1703 if (inode_dirty_flags)
1704 __mark_inode_dirty(inode, inode_dirty_flags);
1707 req->r_inode = inode;
1709 req->r_inode_drop = release;
1710 req->r_args.setattr.mask = cpu_to_le32(mask);
1711 req->r_num_caps = 1;
1712 parent_inode = ceph_get_dentry_parent_inode(dentry);
1713 err = ceph_mdsc_do_request(mdsc, parent_inode, req);
1716 dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1717 ceph_cap_string(dirtied), mask);
1719 ceph_mdsc_put_request(req);
1720 __ceph_do_pending_vmtruncate(inode);
1723 spin_unlock(&ci->i_ceph_lock);
1724 ceph_mdsc_put_request(req);
1729 * Verify that we have a lease on the given mask. If not,
1730 * do a getattr against an mds.
1732 int ceph_do_getattr(struct inode *inode, int mask)
1734 struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1735 struct ceph_mds_client *mdsc = fsc->mdsc;
1736 struct ceph_mds_request *req;
1739 if (ceph_snap(inode) == CEPH_SNAPDIR) {
1740 dout("do_getattr inode %p SNAPDIR\n", inode);
1744 dout("do_getattr inode %p mask %s mode 0%o\n", inode, ceph_cap_string(mask), inode->i_mode);
1745 if (ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1748 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1750 return PTR_ERR(req);
1751 req->r_inode = inode;
1753 req->r_num_caps = 1;
1754 req->r_args.getattr.mask = cpu_to_le32(mask);
1755 err = ceph_mdsc_do_request(mdsc, NULL, req);
1756 ceph_mdsc_put_request(req);
1757 dout("do_getattr result=%d\n", err);
1763 * Check inode permissions. We verify we have a valid value for
1764 * the AUTH cap, then call the generic handler.
1766 int ceph_permission(struct inode *inode, int mask)
1770 if (mask & MAY_NOT_BLOCK)
1773 err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED);
1776 err = generic_permission(inode, mask);
1781 * Get all attributes. Hopefully somedata we'll have a statlite()
1782 * and can limit the fields we require to be accurate.
1784 int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1787 struct inode *inode = dentry->d_inode;
1788 struct ceph_inode_info *ci = ceph_inode(inode);
1791 err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL);
1793 generic_fillattr(inode, stat);
1794 stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
1795 if (ceph_snap(inode) != CEPH_NOSNAP)
1796 stat->dev = ceph_snap(inode);
1799 if (S_ISDIR(inode->i_mode)) {
1800 if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
1802 stat->size = ci->i_rbytes;
1804 stat->size = ci->i_files + ci->i_subdirs;
1806 stat->blksize = 65536;