1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/module.h>
4#include <linux/fs.h>
5#include <linux/slab.h>
6#include <linux/string.h>
7#include <linux/uaccess.h>
8#include <linux/kernel.h>
9#include <linux/namei.h>
10#include <linux/writeback.h>
11#include <linux/vmalloc.h>
12#include <linux/posix_acl.h>
13#include <linux/random.h>
14
15#include "super.h"
16#include "mds_client.h"
17#include "cache.h"
18#include <linux/ceph/decode.h>
19
20/*
21 * Ceph inode operations
22 *
23 * Implement basic inode helpers (get, alloc) and inode ops (getattr,
24 * setattr, etc.), xattr helpers, and helpers for assimilating
25 * metadata returned by the MDS into our cache.
26 *
27 * Also define helpers for doing asynchronous writeback, invalidation,
28 * and truncation for the benefit of those who can't afford to block
29 * (typically because they are in the message handler path).
30 */
31
32static const struct inode_operations ceph_symlink_iops;
33
34static void ceph_invalidate_work(struct work_struct *work);
35static void ceph_writeback_work(struct work_struct *work);
36static void ceph_vmtruncate_work(struct work_struct *work);
37
38/*
39 * find or create an inode, given the ceph ino number
40 */
41static int ceph_set_ino_cb(struct inode *inode, void *data)
42{
43	ceph_inode(inode)->i_vino = *(struct ceph_vino *)data;
44	inode->i_ino = ceph_vino_to_ino(*(struct ceph_vino *)data);
45	return 0;
46}
47
48struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
49{
50	struct inode *inode;
51	ino_t t = ceph_vino_to_ino(vino);
52
53	inode = iget5_locked(sb, t, ceph_ino_compare, ceph_set_ino_cb, &vino);
54	if (inode == NULL)
55		return ERR_PTR(-ENOMEM);
56	if (inode->i_state & I_NEW) {
57		dout("get_inode created new inode %p %llx.%llx ino %llx\n",
58		     inode, ceph_vinop(inode), (u64)inode->i_ino);
59		unlock_new_inode(inode);
60	}
61
62	dout("get_inode on %lu=%llx.%llx got %p\n", inode->i_ino, vino.ino,
63	     vino.snap, inode);
64	return inode;
65}
66
67/*
68 * get/constuct snapdir inode for a given directory
69 */
70struct inode *ceph_get_snapdir(struct inode *parent)
71{
72	struct ceph_vino vino = {
73		.ino = ceph_ino(parent),
74		.snap = CEPH_SNAPDIR,
75	};
76	struct inode *inode = ceph_get_inode(parent->i_sb, vino);
77	struct ceph_inode_info *ci = ceph_inode(inode);
78
79	BUG_ON(!S_ISDIR(parent->i_mode));
80	if (IS_ERR(inode))
81		return inode;
82	inode->i_mode = parent->i_mode;
83	inode->i_uid = parent->i_uid;
84	inode->i_gid = parent->i_gid;
85	inode->i_op = &ceph_snapdir_iops;
86	inode->i_fop = &ceph_snapdir_fops;
87	ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
88	ci->i_rbytes = 0;
89	return inode;
90}
91
92const struct inode_operations ceph_file_iops = {
93	.permission = ceph_permission,
94	.setattr = ceph_setattr,
95	.getattr = ceph_getattr,
96	.setxattr = ceph_setxattr,
97	.getxattr = ceph_getxattr,
98	.listxattr = ceph_listxattr,
99	.removexattr = ceph_removexattr,
100	.get_acl = ceph_get_acl,
101	.set_acl = ceph_set_acl,
102};
103
104
105/*
106 * We use a 'frag tree' to keep track of the MDS's directory fragments
107 * for a given inode (usually there is just a single fragment).  We
108 * need to know when a child frag is delegated to a new MDS, or when
109 * it is flagged as replicated, so we can direct our requests
110 * accordingly.
111 */
112
113/*
114 * find/create a frag in the tree
115 */
116static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
117						    u32 f)
118{
119	struct rb_node **p;
120	struct rb_node *parent = NULL;
121	struct ceph_inode_frag *frag;
122	int c;
123
124	p = &ci->i_fragtree.rb_node;
125	while (*p) {
126		parent = *p;
127		frag = rb_entry(parent, struct ceph_inode_frag, node);
128		c = ceph_frag_compare(f, frag->frag);
129		if (c < 0)
130			p = &(*p)->rb_left;
131		else if (c > 0)
132			p = &(*p)->rb_right;
133		else
134			return frag;
135	}
136
137	frag = kmalloc(sizeof(*frag), GFP_NOFS);
138	if (!frag) {
139		pr_err("__get_or_create_frag ENOMEM on %p %llx.%llx "
140		       "frag %x\n", &ci->vfs_inode,
141		       ceph_vinop(&ci->vfs_inode), f);
142		return ERR_PTR(-ENOMEM);
143	}
144	frag->frag = f;
145	frag->split_by = 0;
146	frag->mds = -1;
147	frag->ndist = 0;
148
149	rb_link_node(&frag->node, parent, p);
150	rb_insert_color(&frag->node, &ci->i_fragtree);
151
152	dout("get_or_create_frag added %llx.%llx frag %x\n",
153	     ceph_vinop(&ci->vfs_inode), f);
154	return frag;
155}
156
157/*
158 * find a specific frag @f
159 */
160struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
161{
162	struct rb_node *n = ci->i_fragtree.rb_node;
163
164	while (n) {
165		struct ceph_inode_frag *frag =
166			rb_entry(n, struct ceph_inode_frag, node);
167		int c = ceph_frag_compare(f, frag->frag);
168		if (c < 0)
169			n = n->rb_left;
170		else if (c > 0)
171			n = n->rb_right;
172		else
173			return frag;
174	}
175	return NULL;
176}
177
178/*
179 * Choose frag containing the given value @v.  If @pfrag is
180 * specified, copy the frag delegation info to the caller if
181 * it is present.
182 */
183static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
184			      struct ceph_inode_frag *pfrag, int *found)
185{
186	u32 t = ceph_frag_make(0, 0);
187	struct ceph_inode_frag *frag;
188	unsigned nway, i;
189	u32 n;
190
191	if (found)
192		*found = 0;
193
194	while (1) {
195		WARN_ON(!ceph_frag_contains_value(t, v));
196		frag = __ceph_find_frag(ci, t);
197		if (!frag)
198			break; /* t is a leaf */
199		if (frag->split_by == 0) {
200			if (pfrag)
201				memcpy(pfrag, frag, sizeof(*pfrag));
202			if (found)
203				*found = 1;
204			break;
205		}
206
207		/* choose child */
208		nway = 1 << frag->split_by;
209		dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
210		     frag->split_by, nway);
211		for (i = 0; i < nway; i++) {
212			n = ceph_frag_make_child(t, frag->split_by, i);
213			if (ceph_frag_contains_value(n, v)) {
214				t = n;
215				break;
216			}
217		}
218		BUG_ON(i == nway);
219	}
220	dout("choose_frag(%x) = %x\n", v, t);
221
222	return t;
223}
224
225u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
226		     struct ceph_inode_frag *pfrag, int *found)
227{
228	u32 ret;
229	mutex_lock(&ci->i_fragtree_mutex);
230	ret = __ceph_choose_frag(ci, v, pfrag, found);
231	mutex_unlock(&ci->i_fragtree_mutex);
232	return ret;
233}
234
235/*
236 * Process dirfrag (delegation) info from the mds.  Include leaf
237 * fragment in tree ONLY if ndist > 0.  Otherwise, only
238 * branches/splits are included in i_fragtree)
239 */
240static int ceph_fill_dirfrag(struct inode *inode,
241			     struct ceph_mds_reply_dirfrag *dirinfo)
242{
243	struct ceph_inode_info *ci = ceph_inode(inode);
244	struct ceph_inode_frag *frag;
245	u32 id = le32_to_cpu(dirinfo->frag);
246	int mds = le32_to_cpu(dirinfo->auth);
247	int ndist = le32_to_cpu(dirinfo->ndist);
248	int diri_auth = -1;
249	int i;
250	int err = 0;
251
252	spin_lock(&ci->i_ceph_lock);
253	if (ci->i_auth_cap)
254		diri_auth = ci->i_auth_cap->mds;
255	spin_unlock(&ci->i_ceph_lock);
256
257	mutex_lock(&ci->i_fragtree_mutex);
258	if (ndist == 0 && mds == diri_auth) {
259		/* no delegation info needed. */
260		frag = __ceph_find_frag(ci, id);
261		if (!frag)
262			goto out;
263		if (frag->split_by == 0) {
264			/* tree leaf, remove */
265			dout("fill_dirfrag removed %llx.%llx frag %x"
266			     " (no ref)\n", ceph_vinop(inode), id);
267			rb_erase(&frag->node, &ci->i_fragtree);
268			kfree(frag);
269		} else {
270			/* tree branch, keep and clear */
271			dout("fill_dirfrag cleared %llx.%llx frag %x"
272			     " referral\n", ceph_vinop(inode), id);
273			frag->mds = -1;
274			frag->ndist = 0;
275		}
276		goto out;
277	}
278
279
280	/* find/add this frag to store mds delegation info */
281	frag = __get_or_create_frag(ci, id);
282	if (IS_ERR(frag)) {
283		/* this is not the end of the world; we can continue
284		   with bad/inaccurate delegation info */
285		pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
286		       ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
287		err = -ENOMEM;
288		goto out;
289	}
290
291	frag->mds = mds;
292	frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
293	for (i = 0; i < frag->ndist; i++)
294		frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
295	dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
296	     ceph_vinop(inode), frag->frag, frag->ndist);
297
298out:
299	mutex_unlock(&ci->i_fragtree_mutex);
300	return err;
301}
302
303static int ceph_fill_fragtree(struct inode *inode,
304			      struct ceph_frag_tree_head *fragtree,
305			      struct ceph_mds_reply_dirfrag *dirinfo)
306{
307	struct ceph_inode_info *ci = ceph_inode(inode);
308	struct ceph_inode_frag *frag;
309	struct rb_node *rb_node;
310	int i;
311	u32 id, nsplits;
312	bool update = false;
313
314	mutex_lock(&ci->i_fragtree_mutex);
315	nsplits = le32_to_cpu(fragtree->nsplits);
316	if (nsplits) {
317		i = prandom_u32() % nsplits;
318		id = le32_to_cpu(fragtree->splits[i].frag);
319		if (!__ceph_find_frag(ci, id))
320			update = true;
321	} else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
322		rb_node = rb_first(&ci->i_fragtree);
323		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
324		if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
325			update = true;
326	}
327	if (!update && dirinfo) {
328		id = le32_to_cpu(dirinfo->frag);
329		if (id != __ceph_choose_frag(ci, id, NULL, NULL))
330			update = true;
331	}
332	if (!update)
333		goto out_unlock;
334
335	dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
336	rb_node = rb_first(&ci->i_fragtree);
337	for (i = 0; i < nsplits; i++) {
338		id = le32_to_cpu(fragtree->splits[i].frag);
339		frag = NULL;
340		while (rb_node) {
341			frag = rb_entry(rb_node, struct ceph_inode_frag, node);
342			if (ceph_frag_compare(frag->frag, id) >= 0) {
343				if (frag->frag != id)
344					frag = NULL;
345				else
346					rb_node = rb_next(rb_node);
347				break;
348			}
349			rb_node = rb_next(rb_node);
350			rb_erase(&frag->node, &ci->i_fragtree);
351			kfree(frag);
352			frag = NULL;
353		}
354		if (!frag) {
355			frag = __get_or_create_frag(ci, id);
356			if (IS_ERR(frag))
357				continue;
358		}
359		frag->split_by = le32_to_cpu(fragtree->splits[i].by);
360		dout(" frag %x split by %d\n", frag->frag, frag->split_by);
361	}
362	while (rb_node) {
363		frag = rb_entry(rb_node, struct ceph_inode_frag, node);
364		rb_node = rb_next(rb_node);
365		rb_erase(&frag->node, &ci->i_fragtree);
366		kfree(frag);
367	}
368out_unlock:
369	mutex_unlock(&ci->i_fragtree_mutex);
370	return 0;
371}
372
373/*
374 * initialize a newly allocated inode.
375 */
376struct inode *ceph_alloc_inode(struct super_block *sb)
377{
378	struct ceph_inode_info *ci;
379	int i;
380
381	ci = kmem_cache_alloc(ceph_inode_cachep, GFP_NOFS);
382	if (!ci)
383		return NULL;
384
385	dout("alloc_inode %p\n", &ci->vfs_inode);
386
387	spin_lock_init(&ci->i_ceph_lock);
388
389	ci->i_version = 0;
390	ci->i_inline_version = 0;
391	ci->i_time_warp_seq = 0;
392	ci->i_ceph_flags = 0;
393	ci->i_ordered_count = 0;
394	atomic_set(&ci->i_release_count, 1);
395	atomic_set(&ci->i_complete_count, 0);
396	ci->i_symlink = NULL;
397
398	memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
399
400	ci->i_fragtree = RB_ROOT;
401	mutex_init(&ci->i_fragtree_mutex);
402
403	ci->i_xattrs.blob = NULL;
404	ci->i_xattrs.prealloc_blob = NULL;
405	ci->i_xattrs.dirty = false;
406	ci->i_xattrs.index = RB_ROOT;
407	ci->i_xattrs.count = 0;
408	ci->i_xattrs.names_size = 0;
409	ci->i_xattrs.vals_size = 0;
410	ci->i_xattrs.version = 0;
411	ci->i_xattrs.index_version = 0;
412
413	ci->i_caps = RB_ROOT;
414	ci->i_auth_cap = NULL;
415	ci->i_dirty_caps = 0;
416	ci->i_flushing_caps = 0;
417	INIT_LIST_HEAD(&ci->i_dirty_item);
418	INIT_LIST_HEAD(&ci->i_flushing_item);
419	ci->i_cap_flush_seq = 0;
420	ci->i_cap_flush_last_tid = 0;
421	memset(&ci->i_cap_flush_tid, 0, sizeof(ci->i_cap_flush_tid));
422	init_waitqueue_head(&ci->i_cap_wq);
423	ci->i_hold_caps_min = 0;
424	ci->i_hold_caps_max = 0;
425	INIT_LIST_HEAD(&ci->i_cap_delay_list);
426	INIT_LIST_HEAD(&ci->i_cap_snaps);
427	ci->i_head_snapc = NULL;
428	ci->i_snap_caps = 0;
429
430	for (i = 0; i < CEPH_FILE_MODE_NUM; i++)
431		ci->i_nr_by_mode[i] = 0;
432
433	mutex_init(&ci->i_truncate_mutex);
434	ci->i_truncate_seq = 0;
435	ci->i_truncate_size = 0;
436	ci->i_truncate_pending = 0;
437
438	ci->i_max_size = 0;
439	ci->i_reported_size = 0;
440	ci->i_wanted_max_size = 0;
441	ci->i_requested_max_size = 0;
442
443	ci->i_pin_ref = 0;
444	ci->i_rd_ref = 0;
445	ci->i_rdcache_ref = 0;
446	ci->i_wr_ref = 0;
447	ci->i_wb_ref = 0;
448	ci->i_wrbuffer_ref = 0;
449	ci->i_wrbuffer_ref_head = 0;
450	ci->i_shared_gen = 0;
451	ci->i_rdcache_gen = 0;
452	ci->i_rdcache_revoking = 0;
453
454	INIT_LIST_HEAD(&ci->i_unsafe_writes);
455	INIT_LIST_HEAD(&ci->i_unsafe_dirops);
456	spin_lock_init(&ci->i_unsafe_lock);
457
458	ci->i_snap_realm = NULL;
459	INIT_LIST_HEAD(&ci->i_snap_realm_item);
460	INIT_LIST_HEAD(&ci->i_snap_flush_item);
461
462	INIT_WORK(&ci->i_wb_work, ceph_writeback_work);
463	INIT_WORK(&ci->i_pg_inv_work, ceph_invalidate_work);
464
465	INIT_WORK(&ci->i_vmtruncate_work, ceph_vmtruncate_work);
466
467	ceph_fscache_inode_init(ci);
468
469	return &ci->vfs_inode;
470}
471
472static void ceph_i_callback(struct rcu_head *head)
473{
474	struct inode *inode = container_of(head, struct inode, i_rcu);
475	struct ceph_inode_info *ci = ceph_inode(inode);
476
477	kmem_cache_free(ceph_inode_cachep, ci);
478}
479
480void ceph_destroy_inode(struct inode *inode)
481{
482	struct ceph_inode_info *ci = ceph_inode(inode);
483	struct ceph_inode_frag *frag;
484	struct rb_node *n;
485
486	dout("destroy_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
487
488	ceph_fscache_unregister_inode_cookie(ci);
489
490	ceph_queue_caps_release(inode);
491
492	/*
493	 * we may still have a snap_realm reference if there are stray
494	 * caps in i_snap_caps.
495	 */
496	if (ci->i_snap_realm) {
497		struct ceph_mds_client *mdsc =
498			ceph_sb_to_client(ci->vfs_inode.i_sb)->mdsc;
499		struct ceph_snap_realm *realm = ci->i_snap_realm;
500
501		dout(" dropping residual ref to snap realm %p\n", realm);
502		spin_lock(&realm->inodes_with_caps_lock);
503		list_del_init(&ci->i_snap_realm_item);
504		spin_unlock(&realm->inodes_with_caps_lock);
505		ceph_put_snap_realm(mdsc, realm);
506	}
507
508	kfree(ci->i_symlink);
509	while ((n = rb_first(&ci->i_fragtree)) != NULL) {
510		frag = rb_entry(n, struct ceph_inode_frag, node);
511		rb_erase(n, &ci->i_fragtree);
512		kfree(frag);
513	}
514
515	__ceph_destroy_xattrs(ci);
516	if (ci->i_xattrs.blob)
517		ceph_buffer_put(ci->i_xattrs.blob);
518	if (ci->i_xattrs.prealloc_blob)
519		ceph_buffer_put(ci->i_xattrs.prealloc_blob);
520
521	call_rcu(&inode->i_rcu, ceph_i_callback);
522}
523
524int ceph_drop_inode(struct inode *inode)
525{
526	/*
527	 * Positve dentry and corresponding inode are always accompanied
528	 * in MDS reply. So no need to keep inode in the cache after
529	 * dropping all its aliases.
530	 */
531	return 1;
532}
533
534/*
535 * Helpers to fill in size, ctime, mtime, and atime.  We have to be
536 * careful because either the client or MDS may have more up to date
537 * info, depending on which capabilities are held, and whether
538 * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
539 * and size are monotonically increasing, except when utimes() or
540 * truncate() increments the corresponding _seq values.)
541 */
542int ceph_fill_file_size(struct inode *inode, int issued,
543			u32 truncate_seq, u64 truncate_size, u64 size)
544{
545	struct ceph_inode_info *ci = ceph_inode(inode);
546	int queue_trunc = 0;
547
548	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
549	    (truncate_seq == ci->i_truncate_seq && size > inode->i_size)) {
550		dout("size %lld -> %llu\n", inode->i_size, size);
551		inode->i_size = size;
552		inode->i_blocks = (size + (1<<9) - 1) >> 9;
553		ci->i_reported_size = size;
554		if (truncate_seq != ci->i_truncate_seq) {
555			dout("truncate_seq %u -> %u\n",
556			     ci->i_truncate_seq, truncate_seq);
557			ci->i_truncate_seq = truncate_seq;
558
559			/* the MDS should have revoked these caps */
560			WARN_ON_ONCE(issued & (CEPH_CAP_FILE_EXCL |
561					       CEPH_CAP_FILE_RD |
562					       CEPH_CAP_FILE_WR |
563					       CEPH_CAP_FILE_LAZYIO));
564			/*
565			 * If we hold relevant caps, or in the case where we're
566			 * not the only client referencing this file and we
567			 * don't hold those caps, then we need to check whether
568			 * the file is either opened or mmaped
569			 */
570			if ((issued & (CEPH_CAP_FILE_CACHE|
571				       CEPH_CAP_FILE_BUFFER)) ||
572			    mapping_mapped(inode->i_mapping) ||
573			    __ceph_caps_file_wanted(ci)) {
574				ci->i_truncate_pending++;
575				queue_trunc = 1;
576			}
577		}
578	}
579	if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
580	    ci->i_truncate_size != truncate_size) {
581		dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
582		     truncate_size);
583		ci->i_truncate_size = truncate_size;
584	}
585
586	if (queue_trunc)
587		ceph_fscache_invalidate(inode);
588
589	return queue_trunc;
590}
591
592void ceph_fill_file_time(struct inode *inode, int issued,
593			 u64 time_warp_seq, struct timespec *ctime,
594			 struct timespec *mtime, struct timespec *atime)
595{
596	struct ceph_inode_info *ci = ceph_inode(inode);
597	int warn = 0;
598
599	if (issued & (CEPH_CAP_FILE_EXCL|
600		      CEPH_CAP_FILE_WR|
601		      CEPH_CAP_FILE_BUFFER|
602		      CEPH_CAP_AUTH_EXCL|
603		      CEPH_CAP_XATTR_EXCL)) {
604		if (timespec_compare(ctime, &inode->i_ctime) > 0) {
605			dout("ctime %ld.%09ld -> %ld.%09ld inc w/ cap\n",
606			     inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
607			     ctime->tv_sec, ctime->tv_nsec);
608			inode->i_ctime = *ctime;
609		}
610		if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
611			/* the MDS did a utimes() */
612			dout("mtime %ld.%09ld -> %ld.%09ld "
613			     "tw %d -> %d\n",
614			     inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
615			     mtime->tv_sec, mtime->tv_nsec,
616			     ci->i_time_warp_seq, (int)time_warp_seq);
617
618			inode->i_mtime = *mtime;
619			inode->i_atime = *atime;
620			ci->i_time_warp_seq = time_warp_seq;
621		} else if (time_warp_seq == ci->i_time_warp_seq) {
622			/* nobody did utimes(); take the max */
623			if (timespec_compare(mtime, &inode->i_mtime) > 0) {
624				dout("mtime %ld.%09ld -> %ld.%09ld inc\n",
625				     inode->i_mtime.tv_sec,
626				     inode->i_mtime.tv_nsec,
627				     mtime->tv_sec, mtime->tv_nsec);
628				inode->i_mtime = *mtime;
629			}
630			if (timespec_compare(atime, &inode->i_atime) > 0) {
631				dout("atime %ld.%09ld -> %ld.%09ld inc\n",
632				     inode->i_atime.tv_sec,
633				     inode->i_atime.tv_nsec,
634				     atime->tv_sec, atime->tv_nsec);
635				inode->i_atime = *atime;
636			}
637		} else if (issued & CEPH_CAP_FILE_EXCL) {
638			/* we did a utimes(); ignore mds values */
639		} else {
640			warn = 1;
641		}
642	} else {
643		/* we have no write|excl caps; whatever the MDS says is true */
644		if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
645			inode->i_ctime = *ctime;
646			inode->i_mtime = *mtime;
647			inode->i_atime = *atime;
648			ci->i_time_warp_seq = time_warp_seq;
649		} else {
650			warn = 1;
651		}
652	}
653	if (warn) /* time_warp_seq shouldn't go backwards */
654		dout("%p mds time_warp_seq %llu < %u\n",
655		     inode, time_warp_seq, ci->i_time_warp_seq);
656}
657
658/*
659 * Populate an inode based on info from mds.  May be called on new or
660 * existing inodes.
661 */
662static int fill_inode(struct inode *inode, struct page *locked_page,
663		      struct ceph_mds_reply_info_in *iinfo,
664		      struct ceph_mds_reply_dirfrag *dirinfo,
665		      struct ceph_mds_session *session,
666		      unsigned long ttl_from, int cap_fmode,
667		      struct ceph_cap_reservation *caps_reservation)
668{
669	struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
670	struct ceph_mds_reply_inode *info = iinfo->in;
671	struct ceph_inode_info *ci = ceph_inode(inode);
672	int issued = 0, implemented, new_issued;
673	struct timespec mtime, atime, ctime;
674	struct ceph_buffer *xattr_blob = NULL;
675	struct ceph_cap *new_cap = NULL;
676	int err = 0;
677	bool wake = false;
678	bool queue_trunc = false;
679	bool new_version = false;
680	bool fill_inline = false;
681
682	dout("fill_inode %p ino %llx.%llx v %llu had %llu\n",
683	     inode, ceph_vinop(inode), le64_to_cpu(info->version),
684	     ci->i_version);
685
686	/* prealloc new cap struct */
687	if (info->cap.caps && ceph_snap(inode) == CEPH_NOSNAP)
688		new_cap = ceph_get_cap(mdsc, caps_reservation);
689
690	/*
691	 * prealloc xattr data, if it looks like we'll need it.  only
692	 * if len > 4 (meaning there are actually xattrs; the first 4
693	 * bytes are the xattr count).
694	 */
695	if (iinfo->xattr_len > 4) {
696		xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
697		if (!xattr_blob)
698			pr_err("fill_inode ENOMEM xattr blob %d bytes\n",
699			       iinfo->xattr_len);
700	}
701
702	spin_lock(&ci->i_ceph_lock);
703
704	/*
705	 * provided version will be odd if inode value is projected,
706	 * even if stable.  skip the update if we have newer stable
707	 * info (ours>=theirs, e.g. due to racing mds replies), unless
708	 * we are getting projected (unstable) info (in which case the
709	 * version is odd, and we want ours>theirs).
710	 *   us   them
711	 *   2    2     skip
712	 *   3    2     skip
713	 *   3    3     update
714	 */
715	if (ci->i_version == 0 ||
716	    ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
717	     le64_to_cpu(info->version) > (ci->i_version & ~1)))
718		new_version = true;
719
720	issued = __ceph_caps_issued(ci, &implemented);
721	issued |= implemented | __ceph_caps_dirty(ci);
722	new_issued = ~issued & le32_to_cpu(info->cap.caps);
723
724	/* update inode */
725	ci->i_version = le64_to_cpu(info->version);
726	inode->i_version++;
727	inode->i_rdev = le32_to_cpu(info->rdev);
728	inode->i_blkbits = fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
729
730	if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
731	    (issued & CEPH_CAP_AUTH_EXCL) == 0) {
732		inode->i_mode = le32_to_cpu(info->mode);
733		inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
734		inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
735		dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
736		     from_kuid(&init_user_ns, inode->i_uid),
737		     from_kgid(&init_user_ns, inode->i_gid));
738	}
739
740	if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
741	    (issued & CEPH_CAP_LINK_EXCL) == 0)
742		set_nlink(inode, le32_to_cpu(info->nlink));
743
744	if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
745		/* be careful with mtime, atime, size */
746		ceph_decode_timespec(&atime, &info->atime);
747		ceph_decode_timespec(&mtime, &info->mtime);
748		ceph_decode_timespec(&ctime, &info->ctime);
749		ceph_fill_file_time(inode, issued,
750				le32_to_cpu(info->time_warp_seq),
751				&ctime, &mtime, &atime);
752	}
753
754	if (new_version ||
755	    (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
756		ci->i_layout = info->layout;
757		queue_trunc = ceph_fill_file_size(inode, issued,
758					le32_to_cpu(info->truncate_seq),
759					le64_to_cpu(info->truncate_size),
760					le64_to_cpu(info->size));
761		/* only update max_size on auth cap */
762		if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
763		    ci->i_max_size != le64_to_cpu(info->max_size)) {
764			dout("max_size %lld -> %llu\n", ci->i_max_size,
765					le64_to_cpu(info->max_size));
766			ci->i_max_size = le64_to_cpu(info->max_size);
767		}
768	}
769
770	/* xattrs */
771	/* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
772	if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
773	    le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
774		if (ci->i_xattrs.blob)
775			ceph_buffer_put(ci->i_xattrs.blob);
776		ci->i_xattrs.blob = xattr_blob;
777		if (xattr_blob)
778			memcpy(ci->i_xattrs.blob->vec.iov_base,
779			       iinfo->xattr_data, iinfo->xattr_len);
780		ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
781		ceph_forget_all_cached_acls(inode);
782		xattr_blob = NULL;
783	}
784
785	inode->i_mapping->a_ops = &ceph_aops;
786
787	switch (inode->i_mode & S_IFMT) {
788	case S_IFIFO:
789	case S_IFBLK:
790	case S_IFCHR:
791	case S_IFSOCK:
792		init_special_inode(inode, inode->i_mode, inode->i_rdev);
793		inode->i_op = &ceph_file_iops;
794		break;
795	case S_IFREG:
796		inode->i_op = &ceph_file_iops;
797		inode->i_fop = &ceph_file_fops;
798		break;
799	case S_IFLNK:
800		inode->i_op = &ceph_symlink_iops;
801		if (!ci->i_symlink) {
802			u32 symlen = iinfo->symlink_len;
803			char *sym;
804
805			spin_unlock(&ci->i_ceph_lock);
806
807			err = -EINVAL;
808			if (WARN_ON(symlen != inode->i_size))
809				goto out;
810
811			err = -ENOMEM;
812			sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
813			if (!sym)
814				goto out;
815
816			spin_lock(&ci->i_ceph_lock);
817			if (!ci->i_symlink)
818				ci->i_symlink = sym;
819			else
820				kfree(sym); /* lost a race */
821		}
822		break;
823	case S_IFDIR:
824		inode->i_op = &ceph_dir_iops;
825		inode->i_fop = &ceph_dir_fops;
826
827		ci->i_dir_layout = iinfo->dir_layout;
828
829		ci->i_files = le64_to_cpu(info->files);
830		ci->i_subdirs = le64_to_cpu(info->subdirs);
831		ci->i_rbytes = le64_to_cpu(info->rbytes);
832		ci->i_rfiles = le64_to_cpu(info->rfiles);
833		ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
834		ceph_decode_timespec(&ci->i_rctime, &info->rctime);
835		break;
836	default:
837		pr_err("fill_inode %llx.%llx BAD mode 0%o\n",
838		       ceph_vinop(inode), inode->i_mode);
839	}
840
841	/* were we issued a capability? */
842	if (info->cap.caps) {
843		if (ceph_snap(inode) == CEPH_NOSNAP) {
844			unsigned caps = le32_to_cpu(info->cap.caps);
845			ceph_add_cap(inode, session,
846				     le64_to_cpu(info->cap.cap_id),
847				     cap_fmode, caps,
848				     le32_to_cpu(info->cap.wanted),
849				     le32_to_cpu(info->cap.seq),
850				     le32_to_cpu(info->cap.mseq),
851				     le64_to_cpu(info->cap.realm),
852				     info->cap.flags, &new_cap);
853
854			/* set dir completion flag? */
855			if (S_ISDIR(inode->i_mode) &&
856			    ci->i_files == 0 && ci->i_subdirs == 0 &&
857			    (caps & CEPH_CAP_FILE_SHARED) &&
858			    (issued & CEPH_CAP_FILE_EXCL) == 0 &&
859			    !__ceph_dir_is_complete(ci)) {
860				dout(" marking %p complete (empty)\n", inode);
861				__ceph_dir_set_complete(ci,
862					atomic_read(&ci->i_release_count),
863					ci->i_ordered_count);
864			}
865
866			wake = true;
867		} else {
868			dout(" %p got snap_caps %s\n", inode,
869			     ceph_cap_string(le32_to_cpu(info->cap.caps)));
870			ci->i_snap_caps |= le32_to_cpu(info->cap.caps);
871			if (cap_fmode >= 0)
872				__ceph_get_fmode(ci, cap_fmode);
873		}
874	} else if (cap_fmode >= 0) {
875		pr_warn("mds issued no caps on %llx.%llx\n",
876			   ceph_vinop(inode));
877		__ceph_get_fmode(ci, cap_fmode);
878	}
879
880	if (iinfo->inline_version > 0 &&
881	    iinfo->inline_version >= ci->i_inline_version) {
882		int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
883		ci->i_inline_version = iinfo->inline_version;
884		if (ci->i_inline_version != CEPH_INLINE_NONE &&
885		    (locked_page ||
886		     (le32_to_cpu(info->cap.caps) & cache_caps)))
887			fill_inline = true;
888	}
889
890	spin_unlock(&ci->i_ceph_lock);
891
892	if (fill_inline)
893		ceph_fill_inline_data(inode, locked_page,
894				      iinfo->inline_data, iinfo->inline_len);
895
896	if (wake)
897		wake_up_all(&ci->i_cap_wq);
898
899	/* queue truncate if we saw i_size decrease */
900	if (queue_trunc)
901		ceph_queue_vmtruncate(inode);
902
903	/* populate frag tree */
904	if (S_ISDIR(inode->i_mode))
905		ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
906
907	/* update delegation info? */
908	if (dirinfo)
909		ceph_fill_dirfrag(inode, dirinfo);
910
911	err = 0;
912out:
913	if (new_cap)
914		ceph_put_cap(mdsc, new_cap);
915	if (xattr_blob)
916		ceph_buffer_put(xattr_blob);
917	return err;
918}
919
920/*
921 * caller should hold session s_mutex.
922 */
923static void update_dentry_lease(struct dentry *dentry,
924				struct ceph_mds_reply_lease *lease,
925				struct ceph_mds_session *session,
926				unsigned long from_time)
927{
928	struct ceph_dentry_info *di = ceph_dentry(dentry);
929	long unsigned duration = le32_to_cpu(lease->duration_ms);
930	long unsigned ttl = from_time + (duration * HZ) / 1000;
931	long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
932	struct inode *dir;
933
934	/* only track leases on regular dentries */
935	if (dentry->d_op != &ceph_dentry_ops)
936		return;
937
938	spin_lock(&dentry->d_lock);
939	dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
940	     dentry, duration, ttl);
941
942	/* make lease_rdcache_gen match directory */
943	dir = d_inode(dentry->d_parent);
944	di->lease_shared_gen = ceph_inode(dir)->i_shared_gen;
945
946	if (duration == 0)
947		goto out_unlock;
948
949	if (di->lease_gen == session->s_cap_gen &&
950	    time_before(ttl, dentry->d_time))
951		goto out_unlock;  /* we already have a newer lease. */
952
953	if (di->lease_session && di->lease_session != session)
954		goto out_unlock;
955
956	ceph_dentry_lru_touch(dentry);
957
958	if (!di->lease_session)
959		di->lease_session = ceph_get_mds_session(session);
960	di->lease_gen = session->s_cap_gen;
961	di->lease_seq = le32_to_cpu(lease->seq);
962	di->lease_renew_after = half_ttl;
963	di->lease_renew_from = 0;
964	dentry->d_time = ttl;
965out_unlock:
966	spin_unlock(&dentry->d_lock);
967	return;
968}
969
970/*
971 * splice a dentry to an inode.
972 * caller must hold directory i_mutex for this to be safe.
973 *
974 * we will only rehash the resulting dentry if @prehash is
975 * true; @prehash will be set to false (for the benefit of
976 * the caller) if we fail.
977 */
978static struct dentry *splice_dentry(struct dentry *dn, struct inode *in,
979				    bool *prehash)
980{
981	struct dentry *realdn;
982
983	BUG_ON(d_inode(dn));
984
985	/* dn must be unhashed */
986	if (!d_unhashed(dn))
987		d_drop(dn);
988	realdn = d_splice_alias(in, dn);
989	if (IS_ERR(realdn)) {
990		pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
991		       PTR_ERR(realdn), dn, in, ceph_vinop(in));
992		if (prehash)
993			*prehash = false; /* don't rehash on error */
994		dn = realdn; /* note realdn contains the error */
995		goto out;
996	} else if (realdn) {
997		dout("dn %p (%d) spliced with %p (%d) "
998		     "inode %p ino %llx.%llx\n",
999		     dn, d_count(dn),
1000		     realdn, d_count(realdn),
1001		     d_inode(realdn), ceph_vinop(d_inode(realdn)));
1002		dput(dn);
1003		dn = realdn;
1004	} else {
1005		BUG_ON(!ceph_dentry(dn));
1006		dout("dn %p attached to %p ino %llx.%llx\n",
1007		     dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1008	}
1009	if ((!prehash || *prehash) && d_unhashed(dn))
1010		d_rehash(dn);
1011out:
1012	return dn;
1013}
1014
1015/*
1016 * Incorporate results into the local cache.  This is either just
1017 * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1018 * after a lookup).
1019 *
1020 * A reply may contain
1021 *         a directory inode along with a dentry.
1022 *  and/or a target inode
1023 *
1024 * Called with snap_rwsem (read).
1025 */
1026int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req,
1027		    struct ceph_mds_session *session)
1028{
1029	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1030	struct inode *in = NULL;
1031	struct ceph_vino vino;
1032	struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1033	int err = 0;
1034
1035	dout("fill_trace %p is_dentry %d is_target %d\n", req,
1036	     rinfo->head->is_dentry, rinfo->head->is_target);
1037
1038#if 0
1039	/*
1040	 * Debugging hook:
1041	 *
1042	 * If we resend completed ops to a recovering mds, we get no
1043	 * trace.  Since that is very rare, pretend this is the case
1044	 * to ensure the 'no trace' handlers in the callers behave.
1045	 *
1046	 * Fill in inodes unconditionally to avoid breaking cap
1047	 * invariants.
1048	 */
1049	if (rinfo->head->op & CEPH_MDS_OP_WRITE) {
1050		pr_info("fill_trace faking empty trace on %lld %s\n",
1051			req->r_tid, ceph_mds_op_name(rinfo->head->op));
1052		if (rinfo->head->is_dentry) {
1053			rinfo->head->is_dentry = 0;
1054			err = fill_inode(req->r_locked_dir,
1055					 &rinfo->diri, rinfo->dirfrag,
1056					 session, req->r_request_started, -1);
1057		}
1058		if (rinfo->head->is_target) {
1059			rinfo->head->is_target = 0;
1060			ininfo = rinfo->targeti.in;
1061			vino.ino = le64_to_cpu(ininfo->ino);
1062			vino.snap = le64_to_cpu(ininfo->snapid);
1063			in = ceph_get_inode(sb, vino);
1064			err = fill_inode(in, &rinfo->targeti, NULL,
1065					 session, req->r_request_started,
1066					 req->r_fmode);
1067			iput(in);
1068		}
1069	}
1070#endif
1071
1072	if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1073		dout("fill_trace reply is empty!\n");
1074		if (rinfo->head->result == 0 && req->r_locked_dir)
1075			ceph_invalidate_dir_request(req);
1076		return 0;
1077	}
1078
1079	if (rinfo->head->is_dentry) {
1080		struct inode *dir = req->r_locked_dir;
1081
1082		if (dir) {
1083			err = fill_inode(dir, NULL,
1084					 &rinfo->diri, rinfo->dirfrag,
1085					 session, req->r_request_started, -1,
1086					 &req->r_caps_reservation);
1087			if (err < 0)
1088				goto done;
1089		} else {
1090			WARN_ON_ONCE(1);
1091		}
1092
1093		if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME) {
1094			struct qstr dname;
1095			struct dentry *dn, *parent;
1096
1097			BUG_ON(!rinfo->head->is_target);
1098			BUG_ON(req->r_dentry);
1099
1100			parent = d_find_any_alias(dir);
1101			BUG_ON(!parent);
1102
1103			dname.name = rinfo->dname;
1104			dname.len = rinfo->dname_len;
1105			dname.hash = full_name_hash(dname.name, dname.len);
1106			vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1107			vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1108retry_lookup:
1109			dn = d_lookup(parent, &dname);
1110			dout("d_lookup on parent=%p name=%.*s got %p\n",
1111			     parent, dname.len, dname.name, dn);
1112
1113			if (!dn) {
1114				dn = d_alloc(parent, &dname);
1115				dout("d_alloc %p '%.*s' = %p\n", parent,
1116				     dname.len, dname.name, dn);
1117				if (dn == NULL) {
1118					dput(parent);
1119					err = -ENOMEM;
1120					goto done;
1121				}
1122				err = ceph_init_dentry(dn);
1123				if (err < 0) {
1124					dput(dn);
1125					dput(parent);
1126					goto done;
1127				}
1128			} else if (d_really_is_positive(dn) &&
1129				   (ceph_ino(d_inode(dn)) != vino.ino ||
1130				    ceph_snap(d_inode(dn)) != vino.snap)) {
1131				dout(" dn %p points to wrong inode %p\n",
1132				     dn, d_inode(dn));
1133				d_delete(dn);
1134				dput(dn);
1135				goto retry_lookup;
1136			}
1137
1138			req->r_dentry = dn;
1139			dput(parent);
1140		}
1141	}
1142
1143	if (rinfo->head->is_target) {
1144		vino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1145		vino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1146
1147		in = ceph_get_inode(sb, vino);
1148		if (IS_ERR(in)) {
1149			err = PTR_ERR(in);
1150			goto done;
1151		}
1152		req->r_target_inode = in;
1153
1154		err = fill_inode(in, req->r_locked_page, &rinfo->targeti, NULL,
1155				session, req->r_request_started,
1156				(!req->r_aborted && rinfo->head->result == 0) ?
1157				req->r_fmode : -1,
1158				&req->r_caps_reservation);
1159		if (err < 0) {
1160			pr_err("fill_inode badness %p %llx.%llx\n",
1161				in, ceph_vinop(in));
1162			goto done;
1163		}
1164	}
1165
1166	/*
1167	 * ignore null lease/binding on snapdir ENOENT, or else we
1168	 * will have trouble splicing in the virtual snapdir later
1169	 */
1170	if (rinfo->head->is_dentry && !req->r_aborted &&
1171	    req->r_locked_dir &&
1172	    (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1173					       fsc->mount_options->snapdir_name,
1174					       req->r_dentry->d_name.len))) {
1175		/*
1176		 * lookup link rename   : null -> possibly existing inode
1177		 * mknod symlink mkdir  : null -> new inode
1178		 * unlink               : linked -> null
1179		 */
1180		struct inode *dir = req->r_locked_dir;
1181		struct dentry *dn = req->r_dentry;
1182		bool have_dir_cap, have_lease;
1183
1184		BUG_ON(!dn);
1185		BUG_ON(!dir);
1186		BUG_ON(d_inode(dn->d_parent) != dir);
1187		BUG_ON(ceph_ino(dir) !=
1188		       le64_to_cpu(rinfo->diri.in->ino));
1189		BUG_ON(ceph_snap(dir) !=
1190		       le64_to_cpu(rinfo->diri.in->snapid));
1191
1192		/* do we have a lease on the whole dir? */
1193		have_dir_cap =
1194			(le32_to_cpu(rinfo->diri.in->cap.caps) &
1195			 CEPH_CAP_FILE_SHARED);
1196
1197		/* do we have a dn lease? */
1198		have_lease = have_dir_cap ||
1199			le32_to_cpu(rinfo->dlease->duration_ms);
1200		if (!have_lease)
1201			dout("fill_trace  no dentry lease or dir cap\n");
1202
1203		/* rename? */
1204		if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1205			struct inode *olddir = req->r_old_dentry_dir;
1206			BUG_ON(!olddir);
1207
1208			dout(" src %p '%pd' dst %p '%pd'\n",
1209			     req->r_old_dentry,
1210			     req->r_old_dentry,
1211			     dn, dn);
1212			dout("fill_trace doing d_move %p -> %p\n",
1213			     req->r_old_dentry, dn);
1214
1215			d_move(req->r_old_dentry, dn);
1216			dout(" src %p '%pd' dst %p '%pd'\n",
1217			     req->r_old_dentry,
1218			     req->r_old_dentry,
1219			     dn, dn);
1220
1221			/* ensure target dentry is invalidated, despite
1222			   rehashing bug in vfs_rename_dir */
1223			ceph_invalidate_dentry_lease(dn);
1224
1225			/* d_move screws up sibling dentries' offsets */
1226			ceph_dir_clear_ordered(dir);
1227			ceph_dir_clear_ordered(olddir);
1228
1229			dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1230			     ceph_dentry(req->r_old_dentry)->offset);
1231
1232			dn = req->r_old_dentry;  /* use old_dentry */
1233		}
1234
1235		/* null dentry? */
1236		if (!rinfo->head->is_target) {
1237			dout("fill_trace null dentry\n");
1238			if (d_really_is_positive(dn)) {
1239				ceph_dir_clear_ordered(dir);
1240				dout("d_delete %p\n", dn);
1241				d_delete(dn);
1242			} else {
1243				dout("d_instantiate %p NULL\n", dn);
1244				d_instantiate(dn, NULL);
1245				if (have_lease && d_unhashed(dn))
1246					d_rehash(dn);
1247				update_dentry_lease(dn, rinfo->dlease,
1248						    session,
1249						    req->r_request_started);
1250			}
1251			goto done;
1252		}
1253
1254		/* attach proper inode */
1255		if (d_really_is_negative(dn)) {
1256			ceph_dir_clear_ordered(dir);
1257			ihold(in);
1258			dn = splice_dentry(dn, in, &have_lease);
1259			if (IS_ERR(dn)) {
1260				err = PTR_ERR(dn);
1261				goto done;
1262			}
1263			req->r_dentry = dn;  /* may have spliced */
1264		} else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1265			dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1266			     dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1267			     ceph_vinop(in));
1268			have_lease = false;
1269		}
1270
1271		if (have_lease)
1272			update_dentry_lease(dn, rinfo->dlease, session,
1273					    req->r_request_started);
1274		dout(" final dn %p\n", dn);
1275	} else if (!req->r_aborted &&
1276		   (req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1277		    req->r_op == CEPH_MDS_OP_MKSNAP)) {
1278		struct dentry *dn = req->r_dentry;
1279		struct inode *dir = req->r_locked_dir;
1280
1281		/* fill out a snapdir LOOKUPSNAP dentry */
1282		BUG_ON(!dn);
1283		BUG_ON(!dir);
1284		BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1285		dout(" linking snapped dir %p to dn %p\n", in, dn);
1286		ceph_dir_clear_ordered(dir);
1287		ihold(in);
1288		dn = splice_dentry(dn, in, NULL);
1289		if (IS_ERR(dn)) {
1290			err = PTR_ERR(dn);
1291			goto done;
1292		}
1293		req->r_dentry = dn;  /* may have spliced */
1294	}
1295done:
1296	dout("fill_trace done err=%d\n", err);
1297	return err;
1298}
1299
1300/*
1301 * Prepopulate our cache with readdir results, leases, etc.
1302 */
1303static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1304					   struct ceph_mds_session *session)
1305{
1306	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1307	int i, err = 0;
1308
1309	for (i = 0; i < rinfo->dir_nr; i++) {
1310		struct ceph_vino vino;
1311		struct inode *in;
1312		int rc;
1313
1314		vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1315		vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1316
1317		in = ceph_get_inode(req->r_dentry->d_sb, vino);
1318		if (IS_ERR(in)) {
1319			err = PTR_ERR(in);
1320			dout("new_inode badness got %d\n", err);
1321			continue;
1322		}
1323		rc = fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1324				req->r_request_started, -1,
1325				&req->r_caps_reservation);
1326		if (rc < 0) {
1327			pr_err("fill_inode badness on %p got %d\n", in, rc);
1328			err = rc;
1329			continue;
1330		}
1331	}
1332
1333	return err;
1334}
1335
1336int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1337			     struct ceph_mds_session *session)
1338{
1339	struct dentry *parent = req->r_dentry;
1340	struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1341	struct qstr dname;
1342	struct dentry *dn;
1343	struct inode *in;
1344	int err = 0, ret, i;
1345	struct inode *snapdir = NULL;
1346	struct ceph_mds_request_head *rhead = req->r_request->front.iov_base;
1347	struct ceph_dentry_info *di;
1348	u64 r_readdir_offset = req->r_readdir_offset;
1349	u32 frag = le32_to_cpu(rhead->args.readdir.frag);
1350
1351	if (rinfo->dir_dir &&
1352	    le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1353		dout("readdir_prepopulate got new frag %x -> %x\n",
1354		     frag, le32_to_cpu(rinfo->dir_dir->frag));
1355		frag = le32_to_cpu(rinfo->dir_dir->frag);
1356		if (ceph_frag_is_leftmost(frag))
1357			r_readdir_offset = 2;
1358		else
1359			r_readdir_offset = 0;
1360	}
1361
1362	if (req->r_aborted)
1363		return readdir_prepopulate_inodes_only(req, session);
1364
1365	if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1366		snapdir = ceph_get_snapdir(d_inode(parent));
1367		parent = d_find_alias(snapdir);
1368		dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1369		     rinfo->dir_nr, parent);
1370	} else {
1371		dout("readdir_prepopulate %d items under dn %p\n",
1372		     rinfo->dir_nr, parent);
1373		if (rinfo->dir_dir)
1374			ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1375	}
1376
1377	/* FIXME: release caps/leases if error occurs */
1378	for (i = 0; i < rinfo->dir_nr; i++) {
1379		struct ceph_vino vino;
1380
1381		dname.name = rinfo->dir_dname[i];
1382		dname.len = rinfo->dir_dname_len[i];
1383		dname.hash = full_name_hash(dname.name, dname.len);
1384
1385		vino.ino = le64_to_cpu(rinfo->dir_in[i].in->ino);
1386		vino.snap = le64_to_cpu(rinfo->dir_in[i].in->snapid);
1387
1388retry_lookup:
1389		dn = d_lookup(parent, &dname);
1390		dout("d_lookup on parent=%p name=%.*s got %p\n",
1391		     parent, dname.len, dname.name, dn);
1392
1393		if (!dn) {
1394			dn = d_alloc(parent, &dname);
1395			dout("d_alloc %p '%.*s' = %p\n", parent,
1396			     dname.len, dname.name, dn);
1397			if (dn == NULL) {
1398				dout("d_alloc badness\n");
1399				err = -ENOMEM;
1400				goto out;
1401			}
1402			ret = ceph_init_dentry(dn);
1403			if (ret < 0) {
1404				dput(dn);
1405				err = ret;
1406				goto out;
1407			}
1408		} else if (d_really_is_positive(dn) &&
1409			   (ceph_ino(d_inode(dn)) != vino.ino ||
1410			    ceph_snap(d_inode(dn)) != vino.snap)) {
1411			dout(" dn %p points to wrong inode %p\n",
1412			     dn, d_inode(dn));
1413			d_delete(dn);
1414			dput(dn);
1415			goto retry_lookup;
1416		} else {
1417			/* reorder parent's d_subdirs */
1418			spin_lock(&parent->d_lock);
1419			spin_lock_nested(&dn->d_lock, DENTRY_D_LOCK_NESTED);
1420			list_move(&dn->d_child, &parent->d_subdirs);
1421			spin_unlock(&dn->d_lock);
1422			spin_unlock(&parent->d_lock);
1423		}
1424
1425		/* inode */
1426		if (d_really_is_positive(dn)) {
1427			in = d_inode(dn);
1428		} else {
1429			in = ceph_get_inode(parent->d_sb, vino);
1430			if (IS_ERR(in)) {
1431				dout("new_inode badness\n");
1432				d_drop(dn);
1433				dput(dn);
1434				err = PTR_ERR(in);
1435				goto out;
1436			}
1437		}
1438
1439		if (fill_inode(in, NULL, &rinfo->dir_in[i], NULL, session,
1440			       req->r_request_started, -1,
1441			       &req->r_caps_reservation) < 0) {
1442			pr_err("fill_inode badness on %p\n", in);
1443			if (d_really_is_negative(dn))
1444				iput(in);
1445			d_drop(dn);
1446			goto next_item;
1447		}
1448
1449		if (d_really_is_negative(dn)) {
1450			struct dentry *realdn = splice_dentry(dn, in, NULL);
1451			if (IS_ERR(realdn)) {
1452				err = PTR_ERR(realdn);
1453				d_drop(dn);
1454				dn = NULL;
1455				goto next_item;
1456			}
1457			dn = realdn;
1458		}
1459
1460		di = dn->d_fsdata;
1461		di->offset = ceph_make_fpos(frag, i + r_readdir_offset);
1462
1463		update_dentry_lease(dn, rinfo->dir_dlease[i],
1464				    req->r_session,
1465				    req->r_request_started);
1466next_item:
1467		if (dn)
1468			dput(dn);
1469	}
1470	if (err == 0)
1471		req->r_did_prepopulate = true;
1472
1473out:
1474	if (snapdir) {
1475		iput(snapdir);
1476		dput(parent);
1477	}
1478	dout("readdir_prepopulate done\n");
1479	return err;
1480}
1481
1482int ceph_inode_set_size(struct inode *inode, loff_t size)
1483{
1484	struct ceph_inode_info *ci = ceph_inode(inode);
1485	int ret = 0;
1486
1487	spin_lock(&ci->i_ceph_lock);
1488	dout("set_size %p %llu -> %llu\n", inode, inode->i_size, size);
1489	inode->i_size = size;
1490	inode->i_blocks = (size + (1 << 9) - 1) >> 9;
1491
1492	/* tell the MDS if we are approaching max_size */
1493	if ((size << 1) >= ci->i_max_size &&
1494	    (ci->i_reported_size << 1) < ci->i_max_size)
1495		ret = 1;
1496
1497	spin_unlock(&ci->i_ceph_lock);
1498	return ret;
1499}
1500
1501/*
1502 * Write back inode data in a worker thread.  (This can't be done
1503 * in the message handler context.)
1504 */
1505void ceph_queue_writeback(struct inode *inode)
1506{
1507	ihold(inode);
1508	if (queue_work(ceph_inode_to_client(inode)->wb_wq,
1509		       &ceph_inode(inode)->i_wb_work)) {
1510		dout("ceph_queue_writeback %p\n", inode);
1511	} else {
1512		dout("ceph_queue_writeback %p failed\n", inode);
1513		iput(inode);
1514	}
1515}
1516
1517static void ceph_writeback_work(struct work_struct *work)
1518{
1519	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1520						  i_wb_work);
1521	struct inode *inode = &ci->vfs_inode;
1522
1523	dout("writeback %p\n", inode);
1524	filemap_fdatawrite(&inode->i_data);
1525	iput(inode);
1526}
1527
1528/*
1529 * queue an async invalidation
1530 */
1531void ceph_queue_invalidate(struct inode *inode)
1532{
1533	ihold(inode);
1534	if (queue_work(ceph_inode_to_client(inode)->pg_inv_wq,
1535		       &ceph_inode(inode)->i_pg_inv_work)) {
1536		dout("ceph_queue_invalidate %p\n", inode);
1537	} else {
1538		dout("ceph_queue_invalidate %p failed\n", inode);
1539		iput(inode);
1540	}
1541}
1542
1543/*
1544 * Invalidate inode pages in a worker thread.  (This can't be done
1545 * in the message handler context.)
1546 */
1547static void ceph_invalidate_work(struct work_struct *work)
1548{
1549	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1550						  i_pg_inv_work);
1551	struct inode *inode = &ci->vfs_inode;
1552	u32 orig_gen;
1553	int check = 0;
1554
1555	mutex_lock(&ci->i_truncate_mutex);
1556	spin_lock(&ci->i_ceph_lock);
1557	dout("invalidate_pages %p gen %d revoking %d\n", inode,
1558	     ci->i_rdcache_gen, ci->i_rdcache_revoking);
1559	if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1560		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1561			check = 1;
1562		spin_unlock(&ci->i_ceph_lock);
1563		mutex_unlock(&ci->i_truncate_mutex);
1564		goto out;
1565	}
1566	orig_gen = ci->i_rdcache_gen;
1567	spin_unlock(&ci->i_ceph_lock);
1568
1569	truncate_pagecache(inode, 0);
1570
1571	spin_lock(&ci->i_ceph_lock);
1572	if (orig_gen == ci->i_rdcache_gen &&
1573	    orig_gen == ci->i_rdcache_revoking) {
1574		dout("invalidate_pages %p gen %d successful\n", inode,
1575		     ci->i_rdcache_gen);
1576		ci->i_rdcache_revoking--;
1577		check = 1;
1578	} else {
1579		dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1580		     inode, orig_gen, ci->i_rdcache_gen,
1581		     ci->i_rdcache_revoking);
1582		if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1583			check = 1;
1584	}
1585	spin_unlock(&ci->i_ceph_lock);
1586	mutex_unlock(&ci->i_truncate_mutex);
1587out:
1588	if (check)
1589		ceph_check_caps(ci, 0, NULL);
1590	iput(inode);
1591}
1592
1593
1594/*
1595 * called by trunc_wq;
1596 *
1597 * We also truncate in a separate thread as well.
1598 */
1599static void ceph_vmtruncate_work(struct work_struct *work)
1600{
1601	struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1602						  i_vmtruncate_work);
1603	struct inode *inode = &ci->vfs_inode;
1604
1605	dout("vmtruncate_work %p\n", inode);
1606	__ceph_do_pending_vmtruncate(inode);
1607	iput(inode);
1608}
1609
1610/*
1611 * Queue an async vmtruncate.  If we fail to queue work, we will handle
1612 * the truncation the next time we call __ceph_do_pending_vmtruncate.
1613 */
1614void ceph_queue_vmtruncate(struct inode *inode)
1615{
1616	struct ceph_inode_info *ci = ceph_inode(inode);
1617
1618	ihold(inode);
1619
1620	if (queue_work(ceph_sb_to_client(inode->i_sb)->trunc_wq,
1621		       &ci->i_vmtruncate_work)) {
1622		dout("ceph_queue_vmtruncate %p\n", inode);
1623	} else {
1624		dout("ceph_queue_vmtruncate %p failed, pending=%d\n",
1625		     inode, ci->i_truncate_pending);
1626		iput(inode);
1627	}
1628}
1629
1630/*
1631 * Make sure any pending truncation is applied before doing anything
1632 * that may depend on it.
1633 */
1634void __ceph_do_pending_vmtruncate(struct inode *inode)
1635{
1636	struct ceph_inode_info *ci = ceph_inode(inode);
1637	u64 to;
1638	int wrbuffer_refs, finish = 0;
1639
1640	mutex_lock(&ci->i_truncate_mutex);
1641retry:
1642	spin_lock(&ci->i_ceph_lock);
1643	if (ci->i_truncate_pending == 0) {
1644		dout("__do_pending_vmtruncate %p none pending\n", inode);
1645		spin_unlock(&ci->i_ceph_lock);
1646		mutex_unlock(&ci->i_truncate_mutex);
1647		return;
1648	}
1649
1650	/*
1651	 * make sure any dirty snapped pages are flushed before we
1652	 * possibly truncate them.. so write AND block!
1653	 */
1654	if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1655		dout("__do_pending_vmtruncate %p flushing snaps first\n",
1656		     inode);
1657		spin_unlock(&ci->i_ceph_lock);
1658		filemap_write_and_wait_range(&inode->i_data, 0,
1659					     inode->i_sb->s_maxbytes);
1660		goto retry;
1661	}
1662
1663	/* there should be no reader or writer */
1664	WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1665
1666	to = ci->i_truncate_size;
1667	wrbuffer_refs = ci->i_wrbuffer_ref;
1668	dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1669	     ci->i_truncate_pending, to);
1670	spin_unlock(&ci->i_ceph_lock);
1671
1672	truncate_pagecache(inode, to);
1673
1674	spin_lock(&ci->i_ceph_lock);
1675	if (to == ci->i_truncate_size) {
1676		ci->i_truncate_pending = 0;
1677		finish = 1;
1678	}
1679	spin_unlock(&ci->i_ceph_lock);
1680	if (!finish)
1681		goto retry;
1682
1683	mutex_unlock(&ci->i_truncate_mutex);
1684
1685	if (wrbuffer_refs == 0)
1686		ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL);
1687
1688	wake_up_all(&ci->i_cap_wq);
1689}
1690
1691/*
1692 * symlinks
1693 */
1694static void *ceph_sym_follow_link(struct dentry *dentry, struct nameidata *nd)
1695{
1696	struct ceph_inode_info *ci = ceph_inode(d_inode(dentry));
1697	nd_set_link(nd, ci->i_symlink);
1698	return NULL;
1699}
1700
1701static const struct inode_operations ceph_symlink_iops = {
1702	.readlink = generic_readlink,
1703	.follow_link = ceph_sym_follow_link,
1704	.setattr = ceph_setattr,
1705	.getattr = ceph_getattr,
1706	.setxattr = ceph_setxattr,
1707	.getxattr = ceph_getxattr,
1708	.listxattr = ceph_listxattr,
1709	.removexattr = ceph_removexattr,
1710};
1711
1712/*
1713 * setattr
1714 */
1715int ceph_setattr(struct dentry *dentry, struct iattr *attr)
1716{
1717	struct inode *inode = d_inode(dentry);
1718	struct ceph_inode_info *ci = ceph_inode(inode);
1719	const unsigned int ia_valid = attr->ia_valid;
1720	struct ceph_mds_request *req;
1721	struct ceph_mds_client *mdsc = ceph_sb_to_client(dentry->d_sb)->mdsc;
1722	int issued;
1723	int release = 0, dirtied = 0;
1724	int mask = 0;
1725	int err = 0;
1726	int inode_dirty_flags = 0;
1727
1728	if (ceph_snap(inode) != CEPH_NOSNAP)
1729		return -EROFS;
1730
1731	err = inode_change_ok(inode, attr);
1732	if (err != 0)
1733		return err;
1734
1735	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
1736				       USE_AUTH_MDS);
1737	if (IS_ERR(req))
1738		return PTR_ERR(req);
1739
1740	spin_lock(&ci->i_ceph_lock);
1741	issued = __ceph_caps_issued(ci, NULL);
1742	dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
1743
1744	if (ia_valid & ATTR_UID) {
1745		dout("setattr %p uid %d -> %d\n", inode,
1746		     from_kuid(&init_user_ns, inode->i_uid),
1747		     from_kuid(&init_user_ns, attr->ia_uid));
1748		if (issued & CEPH_CAP_AUTH_EXCL) {
1749			inode->i_uid = attr->ia_uid;
1750			dirtied |= CEPH_CAP_AUTH_EXCL;
1751		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1752			   !uid_eq(attr->ia_uid, inode->i_uid)) {
1753			req->r_args.setattr.uid = cpu_to_le32(
1754				from_kuid(&init_user_ns, attr->ia_uid));
1755			mask |= CEPH_SETATTR_UID;
1756			release |= CEPH_CAP_AUTH_SHARED;
1757		}
1758	}
1759	if (ia_valid & ATTR_GID) {
1760		dout("setattr %p gid %d -> %d\n", inode,
1761		     from_kgid(&init_user_ns, inode->i_gid),
1762		     from_kgid(&init_user_ns, attr->ia_gid));
1763		if (issued & CEPH_CAP_AUTH_EXCL) {
1764			inode->i_gid = attr->ia_gid;
1765			dirtied |= CEPH_CAP_AUTH_EXCL;
1766		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1767			   !gid_eq(attr->ia_gid, inode->i_gid)) {
1768			req->r_args.setattr.gid = cpu_to_le32(
1769				from_kgid(&init_user_ns, attr->ia_gid));
1770			mask |= CEPH_SETATTR_GID;
1771			release |= CEPH_CAP_AUTH_SHARED;
1772		}
1773	}
1774	if (ia_valid & ATTR_MODE) {
1775		dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
1776		     attr->ia_mode);
1777		if (issued & CEPH_CAP_AUTH_EXCL) {
1778			inode->i_mode = attr->ia_mode;
1779			dirtied |= CEPH_CAP_AUTH_EXCL;
1780		} else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
1781			   attr->ia_mode != inode->i_mode) {
1782			inode->i_mode = attr->ia_mode;
1783			req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
1784			mask |= CEPH_SETATTR_MODE;
1785			release |= CEPH_CAP_AUTH_SHARED;
1786		}
1787	}
1788
1789	if (ia_valid & ATTR_ATIME) {
1790		dout("setattr %p atime %ld.%ld -> %ld.%ld\n", inode,
1791		     inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
1792		     attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
1793		if (issued & CEPH_CAP_FILE_EXCL) {
1794			ci->i_time_warp_seq++;
1795			inode->i_atime = attr->ia_atime;
1796			dirtied |= CEPH_CAP_FILE_EXCL;
1797		} else if ((issued & CEPH_CAP_FILE_WR) &&
1798			   timespec_compare(&inode->i_atime,
1799					    &attr->ia_atime) < 0) {
1800			inode->i_atime = attr->ia_atime;
1801			dirtied |= CEPH_CAP_FILE_WR;
1802		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1803			   !timespec_equal(&inode->i_atime, &attr->ia_atime)) {
1804			ceph_encode_timespec(&req->r_args.setattr.atime,
1805					     &attr->ia_atime);
1806			mask |= CEPH_SETATTR_ATIME;
1807			release |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD |
1808				CEPH_CAP_FILE_WR;
1809		}
1810	}
1811	if (ia_valid & ATTR_MTIME) {
1812		dout("setattr %p mtime %ld.%ld -> %ld.%ld\n", inode,
1813		     inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
1814		     attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
1815		if (issued & CEPH_CAP_FILE_EXCL) {
1816			ci->i_time_warp_seq++;
1817			inode->i_mtime = attr->ia_mtime;
1818			dirtied |= CEPH_CAP_FILE_EXCL;
1819		} else if ((issued & CEPH_CAP_FILE_WR) &&
1820			   timespec_compare(&inode->i_mtime,
1821					    &attr->ia_mtime) < 0) {
1822			inode->i_mtime = attr->ia_mtime;
1823			dirtied |= CEPH_CAP_FILE_WR;
1824		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1825			   !timespec_equal(&inode->i_mtime, &attr->ia_mtime)) {
1826			ceph_encode_timespec(&req->r_args.setattr.mtime,
1827					     &attr->ia_mtime);
1828			mask |= CEPH_SETATTR_MTIME;
1829			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1830				CEPH_CAP_FILE_WR;
1831		}
1832	}
1833	if (ia_valid & ATTR_SIZE) {
1834		dout("setattr %p size %lld -> %lld\n", inode,
1835		     inode->i_size, attr->ia_size);
1836		if ((issued & CEPH_CAP_FILE_EXCL) &&
1837		    attr->ia_size > inode->i_size) {
1838			inode->i_size = attr->ia_size;
1839			inode->i_blocks =
1840				(attr->ia_size + (1 << 9) - 1) >> 9;
1841			inode->i_ctime = attr->ia_ctime;
1842			ci->i_reported_size = attr->ia_size;
1843			dirtied |= CEPH_CAP_FILE_EXCL;
1844		} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
1845			   attr->ia_size != inode->i_size) {
1846			req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
1847			req->r_args.setattr.old_size =
1848				cpu_to_le64(inode->i_size);
1849			mask |= CEPH_SETATTR_SIZE;
1850			release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD |
1851				CEPH_CAP_FILE_WR;
1852		}
1853	}
1854
1855	/* these do nothing */
1856	if (ia_valid & ATTR_CTIME) {
1857		bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
1858					 ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
1859		dout("setattr %p ctime %ld.%ld -> %ld.%ld (%s)\n", inode,
1860		     inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
1861		     attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
1862		     only ? "ctime only" : "ignored");
1863		inode->i_ctime = attr->ia_ctime;
1864		if (only) {
1865			/*
1866			 * if kernel wants to dirty ctime but nothing else,
1867			 * we need to choose a cap to dirty under, or do
1868			 * a almost-no-op setattr
1869			 */
1870			if (issued & CEPH_CAP_AUTH_EXCL)
1871				dirtied |= CEPH_CAP_AUTH_EXCL;
1872			else if (issued & CEPH_CAP_FILE_EXCL)
1873				dirtied |= CEPH_CAP_FILE_EXCL;
1874			else if (issued & CEPH_CAP_XATTR_EXCL)
1875				dirtied |= CEPH_CAP_XATTR_EXCL;
1876			else
1877				mask |= CEPH_SETATTR_CTIME;
1878		}
1879	}
1880	if (ia_valid & ATTR_FILE)
1881		dout("setattr %p ATTR_FILE ... hrm!\n", inode);
1882
1883	if (dirtied) {
1884		inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied);
1885		inode->i_ctime = CURRENT_TIME;
1886	}
1887
1888	release &= issued;
1889	spin_unlock(&ci->i_ceph_lock);
1890
1891	if (inode_dirty_flags)
1892		__mark_inode_dirty(inode, inode_dirty_flags);
1893
1894	if (ia_valid & ATTR_MODE) {
1895		err = posix_acl_chmod(inode, attr->ia_mode);
1896		if (err)
1897			goto out_put;
1898	}
1899
1900	if (mask) {
1901		req->r_inode = inode;
1902		ihold(inode);
1903		req->r_inode_drop = release;
1904		req->r_args.setattr.mask = cpu_to_le32(mask);
1905		req->r_num_caps = 1;
1906		err = ceph_mdsc_do_request(mdsc, NULL, req);
1907	}
1908	dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
1909	     ceph_cap_string(dirtied), mask);
1910
1911	ceph_mdsc_put_request(req);
1912	if (mask & CEPH_SETATTR_SIZE)
1913		__ceph_do_pending_vmtruncate(inode);
1914	return err;
1915out_put:
1916	ceph_mdsc_put_request(req);
1917	return err;
1918}
1919
1920/*
1921 * Verify that we have a lease on the given mask.  If not,
1922 * do a getattr against an mds.
1923 */
1924int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
1925		      int mask, bool force)
1926{
1927	struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
1928	struct ceph_mds_client *mdsc = fsc->mdsc;
1929	struct ceph_mds_request *req;
1930	int err;
1931
1932	if (ceph_snap(inode) == CEPH_SNAPDIR) {
1933		dout("do_getattr inode %p SNAPDIR\n", inode);
1934		return 0;
1935	}
1936
1937	dout("do_getattr inode %p mask %s mode 0%o\n",
1938	     inode, ceph_cap_string(mask), inode->i_mode);
1939	if (!force && ceph_caps_issued_mask(ceph_inode(inode), mask, 1))
1940		return 0;
1941
1942	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, USE_ANY_MDS);
1943	if (IS_ERR(req))
1944		return PTR_ERR(req);
1945	req->r_inode = inode;
1946	ihold(inode);
1947	req->r_num_caps = 1;
1948	req->r_args.getattr.mask = cpu_to_le32(mask);
1949	req->r_locked_page = locked_page;
1950	err = ceph_mdsc_do_request(mdsc, NULL, req);
1951	if (locked_page && err == 0) {
1952		u64 inline_version = req->r_reply_info.targeti.inline_version;
1953		if (inline_version == 0) {
1954			/* the reply is supposed to contain inline data */
1955			err = -EINVAL;
1956		} else if (inline_version == CEPH_INLINE_NONE) {
1957			err = -ENODATA;
1958		} else {
1959			err = req->r_reply_info.targeti.inline_len;
1960		}
1961	}
1962	ceph_mdsc_put_request(req);
1963	dout("do_getattr result=%d\n", err);
1964	return err;
1965}
1966
1967
1968/*
1969 * Check inode permissions.  We verify we have a valid value for
1970 * the AUTH cap, then call the generic handler.
1971 */
1972int ceph_permission(struct inode *inode, int mask)
1973{
1974	int err;
1975
1976	if (mask & MAY_NOT_BLOCK)
1977		return -ECHILD;
1978
1979	err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
1980
1981	if (!err)
1982		err = generic_permission(inode, mask);
1983	return err;
1984}
1985
1986/*
1987 * Get all attributes.  Hopefully somedata we'll have a statlite()
1988 * and can limit the fields we require to be accurate.
1989 */
1990int ceph_getattr(struct vfsmount *mnt, struct dentry *dentry,
1991		 struct kstat *stat)
1992{
1993	struct inode *inode = d_inode(dentry);
1994	struct ceph_inode_info *ci = ceph_inode(inode);
1995	int err;
1996
1997	err = ceph_do_getattr(inode, CEPH_STAT_CAP_INODE_ALL, false);
1998	if (!err) {
1999		generic_fillattr(inode, stat);
2000		stat->ino = ceph_translate_ino(inode->i_sb, inode->i_ino);
2001		if (ceph_snap(inode) != CEPH_NOSNAP)
2002			stat->dev = ceph_snap(inode);
2003		else
2004			stat->dev = 0;
2005		if (S_ISDIR(inode->i_mode)) {
2006			if (ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb),
2007						RBYTES))
2008				stat->size = ci->i_rbytes;
2009			else
2010				stat->size = ci->i_files + ci->i_subdirs;
2011			stat->blocks = 0;
2012			stat->blksize = 65536;
2013		}
2014	}
2015	return err;
2016}
2017