xref: /linux/fs/ceph/dir.c (revision 48c36c8f9a3e881953bb72deb55623a53795a684)
1 #include <linux/ceph/ceph_debug.h>
2 
3 #include <linux/spinlock.h>
4 #include <linux/fs_struct.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
8 
9 #include "super.h"
10 #include "mds_client.h"
11 
12 /*
13  * Directory operations: readdir, lookup, create, link, unlink,
14  * rename, etc.
15  */
16 
17 /*
18  * Ceph MDS operations are specified in terms of a base ino and
19  * relative path.  Thus, the client can specify an operation on a
20  * specific inode (e.g., a getattr due to fstat(2)), or as a path
21  * relative to, say, the root directory.
22  *
23  * Normally, we limit ourselves to strict inode ops (no path component)
24  * or dentry operations (a single path component relative to an ino).  The
25  * exception to this is open_root_dentry(), which will open the mount
26  * point by name.
27  */
28 
29 const struct inode_operations ceph_dir_iops;
30 const struct file_operations ceph_dir_fops;
31 const struct dentry_operations ceph_dentry_ops;
32 
33 /*
34  * Initialize ceph dentry state.
35  */
36 int ceph_init_dentry(struct dentry *dentry)
37 {
38 	struct ceph_dentry_info *di;
39 
40 	if (dentry->d_fsdata)
41 		return 0;
42 
43 	if (ceph_snap(dentry->d_parent->d_inode) == CEPH_NOSNAP)
44 		dentry->d_op = &ceph_dentry_ops;
45 	else if (ceph_snap(dentry->d_parent->d_inode) == CEPH_SNAPDIR)
46 		dentry->d_op = &ceph_snapdir_dentry_ops;
47 	else
48 		dentry->d_op = &ceph_snap_dentry_ops;
49 
50 	di = kmem_cache_alloc(ceph_dentry_cachep, GFP_NOFS | __GFP_ZERO);
51 	if (!di)
52 		return -ENOMEM;          /* oh well */
53 
54 	spin_lock(&dentry->d_lock);
55 	if (dentry->d_fsdata) {
56 		/* lost a race */
57 		kmem_cache_free(ceph_dentry_cachep, di);
58 		goto out_unlock;
59 	}
60 	di->dentry = dentry;
61 	di->lease_session = NULL;
62 	dentry->d_fsdata = di;
63 	dentry->d_time = jiffies;
64 	ceph_dentry_lru_add(dentry);
65 out_unlock:
66 	spin_unlock(&dentry->d_lock);
67 	return 0;
68 }
69 
70 
71 
72 /*
73  * for readdir, we encode the directory frag and offset within that
74  * frag into f_pos.
75  */
76 static unsigned fpos_frag(loff_t p)
77 {
78 	return p >> 32;
79 }
80 static unsigned fpos_off(loff_t p)
81 {
82 	return p & 0xffffffff;
83 }
84 
85 /*
86  * When possible, we try to satisfy a readdir by peeking at the
87  * dcache.  We make this work by carefully ordering dentries on
88  * d_u.d_child when we initially get results back from the MDS, and
89  * falling back to a "normal" sync readdir if any dentries in the dir
90  * are dropped.
91  *
92  * I_COMPLETE tells indicates we have all dentries in the dir.  It is
93  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
94  * the MDS if/when the directory is modified).
95  */
96 static int __dcache_readdir(struct file *filp,
97 			    void *dirent, filldir_t filldir)
98 {
99 	struct ceph_file_info *fi = filp->private_data;
100 	struct dentry *parent = filp->f_dentry;
101 	struct inode *dir = parent->d_inode;
102 	struct list_head *p;
103 	struct dentry *dentry, *last;
104 	struct ceph_dentry_info *di;
105 	int err = 0;
106 
107 	/* claim ref on last dentry we returned */
108 	last = fi->dentry;
109 	fi->dentry = NULL;
110 
111 	dout("__dcache_readdir %p at %llu (last %p)\n", dir, filp->f_pos,
112 	     last);
113 
114 	spin_lock(&dcache_lock);
115 
116 	/* start at beginning? */
117 	if (filp->f_pos == 2 || (last &&
118 				 filp->f_pos < ceph_dentry(last)->offset)) {
119 		if (list_empty(&parent->d_subdirs))
120 			goto out_unlock;
121 		p = parent->d_subdirs.prev;
122 		dout(" initial p %p/%p\n", p->prev, p->next);
123 	} else {
124 		p = last->d_u.d_child.prev;
125 	}
126 
127 more:
128 	dentry = list_entry(p, struct dentry, d_u.d_child);
129 	di = ceph_dentry(dentry);
130 	while (1) {
131 		dout(" p %p/%p %s d_subdirs %p/%p\n", p->prev, p->next,
132 		     d_unhashed(dentry) ? "!hashed" : "hashed",
133 		     parent->d_subdirs.prev, parent->d_subdirs.next);
134 		if (p == &parent->d_subdirs) {
135 			fi->at_end = 1;
136 			goto out_unlock;
137 		}
138 		if (!d_unhashed(dentry) && dentry->d_inode &&
139 		    ceph_snap(dentry->d_inode) != CEPH_SNAPDIR &&
140 		    ceph_ino(dentry->d_inode) != CEPH_INO_CEPH &&
141 		    filp->f_pos <= di->offset)
142 			break;
143 		dout(" skipping %p %.*s at %llu (%llu)%s%s\n", dentry,
144 		     dentry->d_name.len, dentry->d_name.name, di->offset,
145 		     filp->f_pos, d_unhashed(dentry) ? " unhashed" : "",
146 		     !dentry->d_inode ? " null" : "");
147 		p = p->prev;
148 		dentry = list_entry(p, struct dentry, d_u.d_child);
149 		di = ceph_dentry(dentry);
150 	}
151 
152 	atomic_inc(&dentry->d_count);
153 	spin_unlock(&dcache_lock);
154 
155 	dout(" %llu (%llu) dentry %p %.*s %p\n", di->offset, filp->f_pos,
156 	     dentry, dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
157 	filp->f_pos = di->offset;
158 	err = filldir(dirent, dentry->d_name.name,
159 		      dentry->d_name.len, di->offset,
160 		      dentry->d_inode->i_ino,
161 		      dentry->d_inode->i_mode >> 12);
162 
163 	if (last) {
164 		if (err < 0) {
165 			/* remember our position */
166 			fi->dentry = last;
167 			fi->next_offset = di->offset;
168 		} else {
169 			dput(last);
170 		}
171 	}
172 	last = dentry;
173 
174 	if (err < 0)
175 		goto out;
176 
177 	filp->f_pos++;
178 
179 	/* make sure a dentry wasn't dropped while we didn't have dcache_lock */
180 	if (!ceph_i_test(dir, CEPH_I_COMPLETE)) {
181 		dout(" lost I_COMPLETE on %p; falling back to mds\n", dir);
182 		err = -EAGAIN;
183 		goto out;
184 	}
185 
186 	spin_lock(&dcache_lock);
187 	p = p->prev;	/* advance to next dentry */
188 	goto more;
189 
190 out_unlock:
191 	spin_unlock(&dcache_lock);
192 out:
193 	if (last)
194 		dput(last);
195 	return err;
196 }
197 
198 /*
199  * make note of the last dentry we read, so we can
200  * continue at the same lexicographical point,
201  * regardless of what dir changes take place on the
202  * server.
203  */
204 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
205 			    int len)
206 {
207 	kfree(fi->last_name);
208 	fi->last_name = kmalloc(len+1, GFP_NOFS);
209 	if (!fi->last_name)
210 		return -ENOMEM;
211 	memcpy(fi->last_name, name, len);
212 	fi->last_name[len] = 0;
213 	dout("note_last_dentry '%s'\n", fi->last_name);
214 	return 0;
215 }
216 
217 static int ceph_readdir(struct file *filp, void *dirent, filldir_t filldir)
218 {
219 	struct ceph_file_info *fi = filp->private_data;
220 	struct inode *inode = filp->f_dentry->d_inode;
221 	struct ceph_inode_info *ci = ceph_inode(inode);
222 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
223 	struct ceph_mds_client *mdsc = fsc->mdsc;
224 	unsigned frag = fpos_frag(filp->f_pos);
225 	int off = fpos_off(filp->f_pos);
226 	int err;
227 	u32 ftype;
228 	struct ceph_mds_reply_info_parsed *rinfo;
229 	const int max_entries = fsc->mount_options->max_readdir;
230 	const int max_bytes = fsc->mount_options->max_readdir_bytes;
231 
232 	dout("readdir %p filp %p frag %u off %u\n", inode, filp, frag, off);
233 	if (fi->at_end)
234 		return 0;
235 
236 	/* always start with . and .. */
237 	if (filp->f_pos == 0) {
238 		/* note dir version at start of readdir so we can tell
239 		 * if any dentries get dropped */
240 		fi->dir_release_count = ci->i_release_count;
241 
242 		dout("readdir off 0 -> '.'\n");
243 		if (filldir(dirent, ".", 1, ceph_make_fpos(0, 0),
244 			    inode->i_ino, inode->i_mode >> 12) < 0)
245 			return 0;
246 		filp->f_pos = 1;
247 		off = 1;
248 	}
249 	if (filp->f_pos == 1) {
250 		dout("readdir off 1 -> '..'\n");
251 		if (filldir(dirent, "..", 2, ceph_make_fpos(0, 1),
252 			    filp->f_dentry->d_parent->d_inode->i_ino,
253 			    inode->i_mode >> 12) < 0)
254 			return 0;
255 		filp->f_pos = 2;
256 		off = 2;
257 	}
258 
259 	/* can we use the dcache? */
260 	spin_lock(&inode->i_lock);
261 	if ((filp->f_pos == 2 || fi->dentry) &&
262 	    !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
263 	    ceph_snap(inode) != CEPH_SNAPDIR &&
264 	    (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
265 	    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
266 		spin_unlock(&inode->i_lock);
267 		err = __dcache_readdir(filp, dirent, filldir);
268 		if (err != -EAGAIN)
269 			return err;
270 	} else {
271 		spin_unlock(&inode->i_lock);
272 	}
273 	if (fi->dentry) {
274 		err = note_last_dentry(fi, fi->dentry->d_name.name,
275 				       fi->dentry->d_name.len);
276 		if (err)
277 			return err;
278 		dput(fi->dentry);
279 		fi->dentry = NULL;
280 	}
281 
282 	/* proceed with a normal readdir */
283 
284 more:
285 	/* do we have the correct frag content buffered? */
286 	if (fi->frag != frag || fi->last_readdir == NULL) {
287 		struct ceph_mds_request *req;
288 		int op = ceph_snap(inode) == CEPH_SNAPDIR ?
289 			CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
290 
291 		/* discard old result, if any */
292 		if (fi->last_readdir) {
293 			ceph_mdsc_put_request(fi->last_readdir);
294 			fi->last_readdir = NULL;
295 		}
296 
297 		/* requery frag tree, as the frag topology may have changed */
298 		frag = ceph_choose_frag(ceph_inode(inode), frag, NULL, NULL);
299 
300 		dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
301 		     ceph_vinop(inode), frag, fi->last_name);
302 		req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
303 		if (IS_ERR(req))
304 			return PTR_ERR(req);
305 		req->r_inode = igrab(inode);
306 		req->r_dentry = dget(filp->f_dentry);
307 		/* hints to request -> mds selection code */
308 		req->r_direct_mode = USE_AUTH_MDS;
309 		req->r_direct_hash = ceph_frag_value(frag);
310 		req->r_direct_is_hash = true;
311 		req->r_path2 = kstrdup(fi->last_name, GFP_NOFS);
312 		req->r_readdir_offset = fi->next_offset;
313 		req->r_args.readdir.frag = cpu_to_le32(frag);
314 		req->r_args.readdir.max_entries = cpu_to_le32(max_entries);
315 		req->r_args.readdir.max_bytes = cpu_to_le32(max_bytes);
316 		req->r_num_caps = max_entries + 1;
317 		err = ceph_mdsc_do_request(mdsc, NULL, req);
318 		if (err < 0) {
319 			ceph_mdsc_put_request(req);
320 			return err;
321 		}
322 		dout("readdir got and parsed readdir result=%d"
323 		     " on frag %x, end=%d, complete=%d\n", err, frag,
324 		     (int)req->r_reply_info.dir_end,
325 		     (int)req->r_reply_info.dir_complete);
326 
327 		if (!req->r_did_prepopulate) {
328 			dout("readdir !did_prepopulate");
329 			fi->dir_release_count--;    /* preclude I_COMPLETE */
330 		}
331 
332 		/* note next offset and last dentry name */
333 		fi->offset = fi->next_offset;
334 		fi->last_readdir = req;
335 
336 		if (req->r_reply_info.dir_end) {
337 			kfree(fi->last_name);
338 			fi->last_name = NULL;
339 			if (ceph_frag_is_rightmost(frag))
340 				fi->next_offset = 2;
341 			else
342 				fi->next_offset = 0;
343 		} else {
344 			rinfo = &req->r_reply_info;
345 			err = note_last_dentry(fi,
346 				       rinfo->dir_dname[rinfo->dir_nr-1],
347 				       rinfo->dir_dname_len[rinfo->dir_nr-1]);
348 			if (err)
349 				return err;
350 			fi->next_offset += rinfo->dir_nr;
351 		}
352 	}
353 
354 	rinfo = &fi->last_readdir->r_reply_info;
355 	dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
356 	     rinfo->dir_nr, off, fi->offset);
357 	while (off - fi->offset >= 0 && off - fi->offset < rinfo->dir_nr) {
358 		u64 pos = ceph_make_fpos(frag, off);
359 		struct ceph_mds_reply_inode *in =
360 			rinfo->dir_in[off - fi->offset].in;
361 		struct ceph_vino vino;
362 		ino_t ino;
363 
364 		dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
365 		     off, off - fi->offset, rinfo->dir_nr, pos,
366 		     rinfo->dir_dname_len[off - fi->offset],
367 		     rinfo->dir_dname[off - fi->offset], in);
368 		BUG_ON(!in);
369 		ftype = le32_to_cpu(in->mode) >> 12;
370 		vino.ino = le64_to_cpu(in->ino);
371 		vino.snap = le64_to_cpu(in->snapid);
372 		ino = ceph_vino_to_ino(vino);
373 		if (filldir(dirent,
374 			    rinfo->dir_dname[off - fi->offset],
375 			    rinfo->dir_dname_len[off - fi->offset],
376 			    pos, ino, ftype) < 0) {
377 			dout("filldir stopping us...\n");
378 			return 0;
379 		}
380 		off++;
381 		filp->f_pos = pos + 1;
382 	}
383 
384 	if (fi->last_name) {
385 		ceph_mdsc_put_request(fi->last_readdir);
386 		fi->last_readdir = NULL;
387 		goto more;
388 	}
389 
390 	/* more frags? */
391 	if (!ceph_frag_is_rightmost(frag)) {
392 		frag = ceph_frag_next(frag);
393 		off = 0;
394 		filp->f_pos = ceph_make_fpos(frag, off);
395 		dout("readdir next frag is %x\n", frag);
396 		goto more;
397 	}
398 	fi->at_end = 1;
399 
400 	/*
401 	 * if dir_release_count still matches the dir, no dentries
402 	 * were released during the whole readdir, and we should have
403 	 * the complete dir contents in our cache.
404 	 */
405 	spin_lock(&inode->i_lock);
406 	if (ci->i_release_count == fi->dir_release_count) {
407 		dout(" marking %p complete\n", inode);
408 		ci->i_ceph_flags |= CEPH_I_COMPLETE;
409 		ci->i_max_offset = filp->f_pos;
410 	}
411 	spin_unlock(&inode->i_lock);
412 
413 	dout("readdir %p filp %p done.\n", inode, filp);
414 	return 0;
415 }
416 
417 static void reset_readdir(struct ceph_file_info *fi)
418 {
419 	if (fi->last_readdir) {
420 		ceph_mdsc_put_request(fi->last_readdir);
421 		fi->last_readdir = NULL;
422 	}
423 	kfree(fi->last_name);
424 	fi->last_name = NULL;
425 	fi->next_offset = 2;  /* compensate for . and .. */
426 	if (fi->dentry) {
427 		dput(fi->dentry);
428 		fi->dentry = NULL;
429 	}
430 	fi->at_end = 0;
431 }
432 
433 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int origin)
434 {
435 	struct ceph_file_info *fi = file->private_data;
436 	struct inode *inode = file->f_mapping->host;
437 	loff_t old_offset = offset;
438 	loff_t retval;
439 
440 	mutex_lock(&inode->i_mutex);
441 	switch (origin) {
442 	case SEEK_END:
443 		offset += inode->i_size + 2;   /* FIXME */
444 		break;
445 	case SEEK_CUR:
446 		offset += file->f_pos;
447 	}
448 	retval = -EINVAL;
449 	if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) {
450 		if (offset != file->f_pos) {
451 			file->f_pos = offset;
452 			file->f_version = 0;
453 			fi->at_end = 0;
454 		}
455 		retval = offset;
456 
457 		/*
458 		 * discard buffered readdir content on seekdir(0), or
459 		 * seek to new frag, or seek prior to current chunk.
460 		 */
461 		if (offset == 0 ||
462 		    fpos_frag(offset) != fpos_frag(old_offset) ||
463 		    fpos_off(offset) < fi->offset) {
464 			dout("dir_llseek dropping %p content\n", file);
465 			reset_readdir(fi);
466 		}
467 
468 		/* bump dir_release_count if we did a forward seek */
469 		if (offset > old_offset)
470 			fi->dir_release_count--;
471 	}
472 	mutex_unlock(&inode->i_mutex);
473 	return retval;
474 }
475 
476 /*
477  * Process result of a lookup/open request.
478  *
479  * Mainly, make sure we return the final req->r_dentry (if it already
480  * existed) in place of the original VFS-provided dentry when they
481  * differ.
482  *
483  * Gracefully handle the case where the MDS replies with -ENOENT and
484  * no trace (which it may do, at its discretion, e.g., if it doesn't
485  * care to issue a lease on the negative dentry).
486  */
487 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
488 				  struct dentry *dentry, int err)
489 {
490 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
491 	struct inode *parent = dentry->d_parent->d_inode;
492 
493 	/* .snap dir? */
494 	if (err == -ENOENT &&
495 	    strcmp(dentry->d_name.name,
496 		   fsc->mount_options->snapdir_name) == 0) {
497 		struct inode *inode = ceph_get_snapdir(parent);
498 		dout("ENOENT on snapdir %p '%.*s', linking to snapdir %p\n",
499 		     dentry, dentry->d_name.len, dentry->d_name.name, inode);
500 		BUG_ON(!d_unhashed(dentry));
501 		d_add(dentry, inode);
502 		err = 0;
503 	}
504 
505 	if (err == -ENOENT) {
506 		/* no trace? */
507 		err = 0;
508 		if (!req->r_reply_info.head->is_dentry) {
509 			dout("ENOENT and no trace, dentry %p inode %p\n",
510 			     dentry, dentry->d_inode);
511 			if (dentry->d_inode) {
512 				d_drop(dentry);
513 				err = -ENOENT;
514 			} else {
515 				d_add(dentry, NULL);
516 			}
517 		}
518 	}
519 	if (err)
520 		dentry = ERR_PTR(err);
521 	else if (dentry != req->r_dentry)
522 		dentry = dget(req->r_dentry);   /* we got spliced */
523 	else
524 		dentry = NULL;
525 	return dentry;
526 }
527 
528 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
529 {
530 	return ceph_ino(inode) == CEPH_INO_ROOT &&
531 		strncmp(dentry->d_name.name, ".ceph", 5) == 0;
532 }
533 
534 /*
535  * Look up a single dir entry.  If there is a lookup intent, inform
536  * the MDS so that it gets our 'caps wanted' value in a single op.
537  */
538 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
539 				  struct nameidata *nd)
540 {
541 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
542 	struct ceph_mds_client *mdsc = fsc->mdsc;
543 	struct ceph_mds_request *req;
544 	int op;
545 	int err;
546 
547 	dout("lookup %p dentry %p '%.*s'\n",
548 	     dir, dentry, dentry->d_name.len, dentry->d_name.name);
549 
550 	if (dentry->d_name.len > NAME_MAX)
551 		return ERR_PTR(-ENAMETOOLONG);
552 
553 	err = ceph_init_dentry(dentry);
554 	if (err < 0)
555 		return ERR_PTR(err);
556 
557 	/* open (but not create!) intent? */
558 	if (nd &&
559 	    (nd->flags & LOOKUP_OPEN) &&
560 	    (nd->flags & LOOKUP_CONTINUE) == 0 && /* only open last component */
561 	    !(nd->intent.open.flags & O_CREAT)) {
562 		int mode = nd->intent.open.create_mode & ~current->fs->umask;
563 		return ceph_lookup_open(dir, dentry, nd, mode, 1);
564 	}
565 
566 	/* can we conclude ENOENT locally? */
567 	if (dentry->d_inode == NULL) {
568 		struct ceph_inode_info *ci = ceph_inode(dir);
569 		struct ceph_dentry_info *di = ceph_dentry(dentry);
570 
571 		spin_lock(&dir->i_lock);
572 		dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
573 		if (strncmp(dentry->d_name.name,
574 			    fsc->mount_options->snapdir_name,
575 			    dentry->d_name.len) &&
576 		    !is_root_ceph_dentry(dir, dentry) &&
577 		    (ci->i_ceph_flags & CEPH_I_COMPLETE) &&
578 		    (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
579 			spin_unlock(&dir->i_lock);
580 			dout(" dir %p complete, -ENOENT\n", dir);
581 			d_add(dentry, NULL);
582 			di->lease_shared_gen = ci->i_shared_gen;
583 			return NULL;
584 		}
585 		spin_unlock(&dir->i_lock);
586 	}
587 
588 	op = ceph_snap(dir) == CEPH_SNAPDIR ?
589 		CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
590 	req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
591 	if (IS_ERR(req))
592 		return ERR_CAST(req);
593 	req->r_dentry = dget(dentry);
594 	req->r_num_caps = 2;
595 	/* we only need inode linkage */
596 	req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INODE);
597 	req->r_locked_dir = dir;
598 	err = ceph_mdsc_do_request(mdsc, NULL, req);
599 	dentry = ceph_finish_lookup(req, dentry, err);
600 	ceph_mdsc_put_request(req);  /* will dput(dentry) */
601 	dout("lookup result=%p\n", dentry);
602 	return dentry;
603 }
604 
605 /*
606  * If we do a create but get no trace back from the MDS, follow up with
607  * a lookup (the VFS expects us to link up the provided dentry).
608  */
609 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
610 {
611 	struct dentry *result = ceph_lookup(dir, dentry, NULL);
612 
613 	if (result && !IS_ERR(result)) {
614 		/*
615 		 * We created the item, then did a lookup, and found
616 		 * it was already linked to another inode we already
617 		 * had in our cache (and thus got spliced).  Link our
618 		 * dentry to that inode, but don't hash it, just in
619 		 * case the VFS wants to dereference it.
620 		 */
621 		BUG_ON(!result->d_inode);
622 		d_instantiate(dentry, result->d_inode);
623 		return 0;
624 	}
625 	return PTR_ERR(result);
626 }
627 
628 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
629 		      int mode, dev_t rdev)
630 {
631 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
632 	struct ceph_mds_client *mdsc = fsc->mdsc;
633 	struct ceph_mds_request *req;
634 	int err;
635 
636 	if (ceph_snap(dir) != CEPH_NOSNAP)
637 		return -EROFS;
638 
639 	dout("mknod in dir %p dentry %p mode 0%o rdev %d\n",
640 	     dir, dentry, mode, rdev);
641 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
642 	if (IS_ERR(req)) {
643 		d_drop(dentry);
644 		return PTR_ERR(req);
645 	}
646 	req->r_dentry = dget(dentry);
647 	req->r_num_caps = 2;
648 	req->r_locked_dir = dir;
649 	req->r_args.mknod.mode = cpu_to_le32(mode);
650 	req->r_args.mknod.rdev = cpu_to_le32(rdev);
651 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
652 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
653 	err = ceph_mdsc_do_request(mdsc, dir, req);
654 	if (!err && !req->r_reply_info.head->is_dentry)
655 		err = ceph_handle_notrace_create(dir, dentry);
656 	ceph_mdsc_put_request(req);
657 	if (err)
658 		d_drop(dentry);
659 	return err;
660 }
661 
662 static int ceph_create(struct inode *dir, struct dentry *dentry, int mode,
663 		       struct nameidata *nd)
664 {
665 	dout("create in dir %p dentry %p name '%.*s'\n",
666 	     dir, dentry, dentry->d_name.len, dentry->d_name.name);
667 
668 	if (ceph_snap(dir) != CEPH_NOSNAP)
669 		return -EROFS;
670 
671 	if (nd) {
672 		BUG_ON((nd->flags & LOOKUP_OPEN) == 0);
673 		dentry = ceph_lookup_open(dir, dentry, nd, mode, 0);
674 		/* hrm, what should i do here if we get aliased? */
675 		if (IS_ERR(dentry))
676 			return PTR_ERR(dentry);
677 		return 0;
678 	}
679 
680 	/* fall back to mknod */
681 	return ceph_mknod(dir, dentry, (mode & ~S_IFMT) | S_IFREG, 0);
682 }
683 
684 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
685 			    const char *dest)
686 {
687 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
688 	struct ceph_mds_client *mdsc = fsc->mdsc;
689 	struct ceph_mds_request *req;
690 	int err;
691 
692 	if (ceph_snap(dir) != CEPH_NOSNAP)
693 		return -EROFS;
694 
695 	dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
696 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
697 	if (IS_ERR(req)) {
698 		d_drop(dentry);
699 		return PTR_ERR(req);
700 	}
701 	req->r_dentry = dget(dentry);
702 	req->r_num_caps = 2;
703 	req->r_path2 = kstrdup(dest, GFP_NOFS);
704 	req->r_locked_dir = dir;
705 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
706 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
707 	err = ceph_mdsc_do_request(mdsc, dir, req);
708 	if (!err && !req->r_reply_info.head->is_dentry)
709 		err = ceph_handle_notrace_create(dir, dentry);
710 	ceph_mdsc_put_request(req);
711 	if (err)
712 		d_drop(dentry);
713 	return err;
714 }
715 
716 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, int mode)
717 {
718 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
719 	struct ceph_mds_client *mdsc = fsc->mdsc;
720 	struct ceph_mds_request *req;
721 	int err = -EROFS;
722 	int op;
723 
724 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
725 		/* mkdir .snap/foo is a MKSNAP */
726 		op = CEPH_MDS_OP_MKSNAP;
727 		dout("mksnap dir %p snap '%.*s' dn %p\n", dir,
728 		     dentry->d_name.len, dentry->d_name.name, dentry);
729 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
730 		dout("mkdir dir %p dn %p mode 0%o\n", dir, dentry, mode);
731 		op = CEPH_MDS_OP_MKDIR;
732 	} else {
733 		goto out;
734 	}
735 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
736 	if (IS_ERR(req)) {
737 		err = PTR_ERR(req);
738 		goto out;
739 	}
740 
741 	req->r_dentry = dget(dentry);
742 	req->r_num_caps = 2;
743 	req->r_locked_dir = dir;
744 	req->r_args.mkdir.mode = cpu_to_le32(mode);
745 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
746 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
747 	err = ceph_mdsc_do_request(mdsc, dir, req);
748 	if (!err && !req->r_reply_info.head->is_dentry)
749 		err = ceph_handle_notrace_create(dir, dentry);
750 	ceph_mdsc_put_request(req);
751 out:
752 	if (err < 0)
753 		d_drop(dentry);
754 	return err;
755 }
756 
757 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
758 		     struct dentry *dentry)
759 {
760 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
761 	struct ceph_mds_client *mdsc = fsc->mdsc;
762 	struct ceph_mds_request *req;
763 	int err;
764 
765 	if (ceph_snap(dir) != CEPH_NOSNAP)
766 		return -EROFS;
767 
768 	dout("link in dir %p old_dentry %p dentry %p\n", dir,
769 	     old_dentry, dentry);
770 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
771 	if (IS_ERR(req)) {
772 		d_drop(dentry);
773 		return PTR_ERR(req);
774 	}
775 	req->r_dentry = dget(dentry);
776 	req->r_num_caps = 2;
777 	req->r_old_dentry = dget(old_dentry); /* or inode? hrm. */
778 	req->r_locked_dir = dir;
779 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
780 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
781 	err = ceph_mdsc_do_request(mdsc, dir, req);
782 	if (err)
783 		d_drop(dentry);
784 	else if (!req->r_reply_info.head->is_dentry)
785 		d_instantiate(dentry, igrab(old_dentry->d_inode));
786 	ceph_mdsc_put_request(req);
787 	return err;
788 }
789 
790 /*
791  * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps.  If it
792  * looks like the link count will hit 0, drop any other caps (other
793  * than PIN) we don't specifically want (due to the file still being
794  * open).
795  */
796 static int drop_caps_for_unlink(struct inode *inode)
797 {
798 	struct ceph_inode_info *ci = ceph_inode(inode);
799 	int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
800 
801 	spin_lock(&inode->i_lock);
802 	if (inode->i_nlink == 1) {
803 		drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
804 		ci->i_ceph_flags |= CEPH_I_NODELAY;
805 	}
806 	spin_unlock(&inode->i_lock);
807 	return drop;
808 }
809 
810 /*
811  * rmdir and unlink are differ only by the metadata op code
812  */
813 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
814 {
815 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
816 	struct ceph_mds_client *mdsc = fsc->mdsc;
817 	struct inode *inode = dentry->d_inode;
818 	struct ceph_mds_request *req;
819 	int err = -EROFS;
820 	int op;
821 
822 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
823 		/* rmdir .snap/foo is RMSNAP */
824 		dout("rmsnap dir %p '%.*s' dn %p\n", dir, dentry->d_name.len,
825 		     dentry->d_name.name, dentry);
826 		op = CEPH_MDS_OP_RMSNAP;
827 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
828 		dout("unlink/rmdir dir %p dn %p inode %p\n",
829 		     dir, dentry, inode);
830 		op = ((dentry->d_inode->i_mode & S_IFMT) == S_IFDIR) ?
831 			CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
832 	} else
833 		goto out;
834 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
835 	if (IS_ERR(req)) {
836 		err = PTR_ERR(req);
837 		goto out;
838 	}
839 	req->r_dentry = dget(dentry);
840 	req->r_num_caps = 2;
841 	req->r_locked_dir = dir;
842 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
843 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
844 	req->r_inode_drop = drop_caps_for_unlink(inode);
845 	err = ceph_mdsc_do_request(mdsc, dir, req);
846 	if (!err && !req->r_reply_info.head->is_dentry)
847 		d_delete(dentry);
848 	ceph_mdsc_put_request(req);
849 out:
850 	return err;
851 }
852 
853 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
854 		       struct inode *new_dir, struct dentry *new_dentry)
855 {
856 	struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
857 	struct ceph_mds_client *mdsc = fsc->mdsc;
858 	struct ceph_mds_request *req;
859 	int err;
860 
861 	if (ceph_snap(old_dir) != ceph_snap(new_dir))
862 		return -EXDEV;
863 	if (ceph_snap(old_dir) != CEPH_NOSNAP ||
864 	    ceph_snap(new_dir) != CEPH_NOSNAP)
865 		return -EROFS;
866 	dout("rename dir %p dentry %p to dir %p dentry %p\n",
867 	     old_dir, old_dentry, new_dir, new_dentry);
868 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_RENAME, USE_AUTH_MDS);
869 	if (IS_ERR(req))
870 		return PTR_ERR(req);
871 	req->r_dentry = dget(new_dentry);
872 	req->r_num_caps = 2;
873 	req->r_old_dentry = dget(old_dentry);
874 	req->r_locked_dir = new_dir;
875 	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
876 	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
877 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
878 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
879 	/* release LINK_RDCACHE on source inode (mds will lock it) */
880 	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
881 	if (new_dentry->d_inode)
882 		req->r_inode_drop = drop_caps_for_unlink(new_dentry->d_inode);
883 	err = ceph_mdsc_do_request(mdsc, old_dir, req);
884 	if (!err && !req->r_reply_info.head->is_dentry) {
885 		/*
886 		 * Normally d_move() is done by fill_trace (called by
887 		 * do_request, above).  If there is no trace, we need
888 		 * to do it here.
889 		 */
890 
891 		/* d_move screws up d_subdirs order */
892 		ceph_i_clear(new_dir, CEPH_I_COMPLETE);
893 
894 		d_move(old_dentry, new_dentry);
895 
896 		/* ensure target dentry is invalidated, despite
897 		   rehashing bug in vfs_rename_dir */
898 		ceph_invalidate_dentry_lease(new_dentry);
899 	}
900 	ceph_mdsc_put_request(req);
901 	return err;
902 }
903 
904 /*
905  * Ensure a dentry lease will no longer revalidate.
906  */
907 void ceph_invalidate_dentry_lease(struct dentry *dentry)
908 {
909 	spin_lock(&dentry->d_lock);
910 	dentry->d_time = jiffies;
911 	ceph_dentry(dentry)->lease_shared_gen = 0;
912 	spin_unlock(&dentry->d_lock);
913 }
914 
915 /*
916  * Check if dentry lease is valid.  If not, delete the lease.  Try to
917  * renew if the least is more than half up.
918  */
919 static int dentry_lease_is_valid(struct dentry *dentry)
920 {
921 	struct ceph_dentry_info *di;
922 	struct ceph_mds_session *s;
923 	int valid = 0;
924 	u32 gen;
925 	unsigned long ttl;
926 	struct ceph_mds_session *session = NULL;
927 	struct inode *dir = NULL;
928 	u32 seq = 0;
929 
930 	spin_lock(&dentry->d_lock);
931 	di = ceph_dentry(dentry);
932 	if (di && di->lease_session) {
933 		s = di->lease_session;
934 		spin_lock(&s->s_cap_lock);
935 		gen = s->s_cap_gen;
936 		ttl = s->s_cap_ttl;
937 		spin_unlock(&s->s_cap_lock);
938 
939 		if (di->lease_gen == gen &&
940 		    time_before(jiffies, dentry->d_time) &&
941 		    time_before(jiffies, ttl)) {
942 			valid = 1;
943 			if (di->lease_renew_after &&
944 			    time_after(jiffies, di->lease_renew_after)) {
945 				/* we should renew */
946 				dir = dentry->d_parent->d_inode;
947 				session = ceph_get_mds_session(s);
948 				seq = di->lease_seq;
949 				di->lease_renew_after = 0;
950 				di->lease_renew_from = jiffies;
951 			}
952 		}
953 	}
954 	spin_unlock(&dentry->d_lock);
955 
956 	if (session) {
957 		ceph_mdsc_lease_send_msg(session, dir, dentry,
958 					 CEPH_MDS_LEASE_RENEW, seq);
959 		ceph_put_mds_session(session);
960 	}
961 	dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
962 	return valid;
963 }
964 
965 /*
966  * Check if directory-wide content lease/cap is valid.
967  */
968 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
969 {
970 	struct ceph_inode_info *ci = ceph_inode(dir);
971 	struct ceph_dentry_info *di = ceph_dentry(dentry);
972 	int valid = 0;
973 
974 	spin_lock(&dir->i_lock);
975 	if (ci->i_shared_gen == di->lease_shared_gen)
976 		valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
977 	spin_unlock(&dir->i_lock);
978 	dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
979 	     dir, (unsigned)ci->i_shared_gen, dentry,
980 	     (unsigned)di->lease_shared_gen, valid);
981 	return valid;
982 }
983 
984 /*
985  * Check if cached dentry can be trusted.
986  */
987 static int ceph_d_revalidate(struct dentry *dentry, struct nameidata *nd)
988 {
989 	struct inode *dir = dentry->d_parent->d_inode;
990 
991 	dout("d_revalidate %p '%.*s' inode %p offset %lld\n", dentry,
992 	     dentry->d_name.len, dentry->d_name.name, dentry->d_inode,
993 	     ceph_dentry(dentry)->offset);
994 
995 	/* always trust cached snapped dentries, snapdir dentry */
996 	if (ceph_snap(dir) != CEPH_NOSNAP) {
997 		dout("d_revalidate %p '%.*s' inode %p is SNAPPED\n", dentry,
998 		     dentry->d_name.len, dentry->d_name.name, dentry->d_inode);
999 		goto out_touch;
1000 	}
1001 	if (dentry->d_inode && ceph_snap(dentry->d_inode) == CEPH_SNAPDIR)
1002 		goto out_touch;
1003 
1004 	if (dentry_lease_is_valid(dentry) ||
1005 	    dir_lease_is_valid(dir, dentry))
1006 		goto out_touch;
1007 
1008 	dout("d_revalidate %p invalid\n", dentry);
1009 	d_drop(dentry);
1010 	return 0;
1011 out_touch:
1012 	ceph_dentry_lru_touch(dentry);
1013 	return 1;
1014 }
1015 
1016 /*
1017  * When a dentry is released, clear the dir I_COMPLETE if it was part
1018  * of the current dir gen or if this is in the snapshot namespace.
1019  */
1020 static void ceph_dentry_release(struct dentry *dentry)
1021 {
1022 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1023 	struct inode *parent_inode = NULL;
1024 	u64 snapid = CEPH_NOSNAP;
1025 
1026 	if (!IS_ROOT(dentry)) {
1027 		parent_inode = dentry->d_parent->d_inode;
1028 		if (parent_inode)
1029 			snapid = ceph_snap(parent_inode);
1030 	}
1031 	dout("dentry_release %p parent %p\n", dentry, parent_inode);
1032 	if (parent_inode && snapid != CEPH_SNAPDIR) {
1033 		struct ceph_inode_info *ci = ceph_inode(parent_inode);
1034 
1035 		spin_lock(&parent_inode->i_lock);
1036 		if (ci->i_shared_gen == di->lease_shared_gen ||
1037 		    snapid <= CEPH_MAXSNAP) {
1038 			dout(" clearing %p complete (d_release)\n",
1039 			     parent_inode);
1040 			ci->i_ceph_flags &= ~CEPH_I_COMPLETE;
1041 			ci->i_release_count++;
1042 		}
1043 		spin_unlock(&parent_inode->i_lock);
1044 	}
1045 	if (di) {
1046 		ceph_dentry_lru_del(dentry);
1047 		if (di->lease_session)
1048 			ceph_put_mds_session(di->lease_session);
1049 		kmem_cache_free(ceph_dentry_cachep, di);
1050 		dentry->d_fsdata = NULL;
1051 	}
1052 }
1053 
1054 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1055 					  struct nameidata *nd)
1056 {
1057 	/*
1058 	 * Eventually, we'll want to revalidate snapped metadata
1059 	 * too... probably...
1060 	 */
1061 	return 1;
1062 }
1063 
1064 
1065 
1066 /*
1067  * read() on a dir.  This weird interface hack only works if mounted
1068  * with '-o dirstat'.
1069  */
1070 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1071 			     loff_t *ppos)
1072 {
1073 	struct ceph_file_info *cf = file->private_data;
1074 	struct inode *inode = file->f_dentry->d_inode;
1075 	struct ceph_inode_info *ci = ceph_inode(inode);
1076 	int left;
1077 
1078 	if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1079 		return -EISDIR;
1080 
1081 	if (!cf->dir_info) {
1082 		cf->dir_info = kmalloc(1024, GFP_NOFS);
1083 		if (!cf->dir_info)
1084 			return -ENOMEM;
1085 		cf->dir_info_len =
1086 			sprintf(cf->dir_info,
1087 				"entries:   %20lld\n"
1088 				" files:    %20lld\n"
1089 				" subdirs:  %20lld\n"
1090 				"rentries:  %20lld\n"
1091 				" rfiles:   %20lld\n"
1092 				" rsubdirs: %20lld\n"
1093 				"rbytes:    %20lld\n"
1094 				"rctime:    %10ld.%09ld\n",
1095 				ci->i_files + ci->i_subdirs,
1096 				ci->i_files,
1097 				ci->i_subdirs,
1098 				ci->i_rfiles + ci->i_rsubdirs,
1099 				ci->i_rfiles,
1100 				ci->i_rsubdirs,
1101 				ci->i_rbytes,
1102 				(long)ci->i_rctime.tv_sec,
1103 				(long)ci->i_rctime.tv_nsec);
1104 	}
1105 
1106 	if (*ppos >= cf->dir_info_len)
1107 		return 0;
1108 	size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1109 	left = copy_to_user(buf, cf->dir_info + *ppos, size);
1110 	if (left == size)
1111 		return -EFAULT;
1112 	*ppos += (size - left);
1113 	return size - left;
1114 }
1115 
1116 /*
1117  * an fsync() on a dir will wait for any uncommitted directory
1118  * operations to commit.
1119  */
1120 static int ceph_dir_fsync(struct file *file, int datasync)
1121 {
1122 	struct inode *inode = file->f_path.dentry->d_inode;
1123 	struct ceph_inode_info *ci = ceph_inode(inode);
1124 	struct list_head *head = &ci->i_unsafe_dirops;
1125 	struct ceph_mds_request *req;
1126 	u64 last_tid;
1127 	int ret = 0;
1128 
1129 	dout("dir_fsync %p\n", inode);
1130 	spin_lock(&ci->i_unsafe_lock);
1131 	if (list_empty(head))
1132 		goto out;
1133 
1134 	req = list_entry(head->prev,
1135 			 struct ceph_mds_request, r_unsafe_dir_item);
1136 	last_tid = req->r_tid;
1137 
1138 	do {
1139 		ceph_mdsc_get_request(req);
1140 		spin_unlock(&ci->i_unsafe_lock);
1141 		dout("dir_fsync %p wait on tid %llu (until %llu)\n",
1142 		     inode, req->r_tid, last_tid);
1143 		if (req->r_timeout) {
1144 			ret = wait_for_completion_timeout(
1145 				&req->r_safe_completion, req->r_timeout);
1146 			if (ret > 0)
1147 				ret = 0;
1148 			else if (ret == 0)
1149 				ret = -EIO;  /* timed out */
1150 		} else {
1151 			wait_for_completion(&req->r_safe_completion);
1152 		}
1153 		spin_lock(&ci->i_unsafe_lock);
1154 		ceph_mdsc_put_request(req);
1155 
1156 		if (ret || list_empty(head))
1157 			break;
1158 		req = list_entry(head->next,
1159 				 struct ceph_mds_request, r_unsafe_dir_item);
1160 	} while (req->r_tid < last_tid);
1161 out:
1162 	spin_unlock(&ci->i_unsafe_lock);
1163 	return ret;
1164 }
1165 
1166 /*
1167  * We maintain a private dentry LRU.
1168  *
1169  * FIXME: this needs to be changed to a per-mds lru to be useful.
1170  */
1171 void ceph_dentry_lru_add(struct dentry *dn)
1172 {
1173 	struct ceph_dentry_info *di = ceph_dentry(dn);
1174 	struct ceph_mds_client *mdsc;
1175 
1176 	dout("dentry_lru_add %p %p '%.*s'\n", di, dn,
1177 	     dn->d_name.len, dn->d_name.name);
1178 	if (di) {
1179 		mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1180 		spin_lock(&mdsc->dentry_lru_lock);
1181 		list_add_tail(&di->lru, &mdsc->dentry_lru);
1182 		mdsc->num_dentry++;
1183 		spin_unlock(&mdsc->dentry_lru_lock);
1184 	}
1185 }
1186 
1187 void ceph_dentry_lru_touch(struct dentry *dn)
1188 {
1189 	struct ceph_dentry_info *di = ceph_dentry(dn);
1190 	struct ceph_mds_client *mdsc;
1191 
1192 	dout("dentry_lru_touch %p %p '%.*s' (offset %lld)\n", di, dn,
1193 	     dn->d_name.len, dn->d_name.name, di->offset);
1194 	if (di) {
1195 		mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1196 		spin_lock(&mdsc->dentry_lru_lock);
1197 		list_move_tail(&di->lru, &mdsc->dentry_lru);
1198 		spin_unlock(&mdsc->dentry_lru_lock);
1199 	}
1200 }
1201 
1202 void ceph_dentry_lru_del(struct dentry *dn)
1203 {
1204 	struct ceph_dentry_info *di = ceph_dentry(dn);
1205 	struct ceph_mds_client *mdsc;
1206 
1207 	dout("dentry_lru_del %p %p '%.*s'\n", di, dn,
1208 	     dn->d_name.len, dn->d_name.name);
1209 	if (di) {
1210 		mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1211 		spin_lock(&mdsc->dentry_lru_lock);
1212 		list_del_init(&di->lru);
1213 		mdsc->num_dentry--;
1214 		spin_unlock(&mdsc->dentry_lru_lock);
1215 	}
1216 }
1217 
1218 const struct file_operations ceph_dir_fops = {
1219 	.read = ceph_read_dir,
1220 	.readdir = ceph_readdir,
1221 	.llseek = ceph_dir_llseek,
1222 	.open = ceph_open,
1223 	.release = ceph_release,
1224 	.unlocked_ioctl = ceph_ioctl,
1225 	.fsync = ceph_dir_fsync,
1226 };
1227 
1228 const struct inode_operations ceph_dir_iops = {
1229 	.lookup = ceph_lookup,
1230 	.permission = ceph_permission,
1231 	.getattr = ceph_getattr,
1232 	.setattr = ceph_setattr,
1233 	.setxattr = ceph_setxattr,
1234 	.getxattr = ceph_getxattr,
1235 	.listxattr = ceph_listxattr,
1236 	.removexattr = ceph_removexattr,
1237 	.mknod = ceph_mknod,
1238 	.symlink = ceph_symlink,
1239 	.mkdir = ceph_mkdir,
1240 	.link = ceph_link,
1241 	.unlink = ceph_unlink,
1242 	.rmdir = ceph_unlink,
1243 	.rename = ceph_rename,
1244 	.create = ceph_create,
1245 };
1246 
1247 const struct dentry_operations ceph_dentry_ops = {
1248 	.d_revalidate = ceph_d_revalidate,
1249 	.d_release = ceph_dentry_release,
1250 };
1251 
1252 const struct dentry_operations ceph_snapdir_dentry_ops = {
1253 	.d_revalidate = ceph_snapdir_d_revalidate,
1254 	.d_release = ceph_dentry_release,
1255 };
1256 
1257 const struct dentry_operations ceph_snap_dentry_ops = {
1258 	.d_release = ceph_dentry_release,
1259 };
1260