xref: /openbmc/linux/fs/ceph/dir.c (revision 2a5beea3)
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 dentry_operations ceph_dentry_ops;
30 
31 /*
32  * Initialize ceph dentry state.
33  */
34 int ceph_init_dentry(struct dentry *dentry)
35 {
36 	struct ceph_dentry_info *di;
37 
38 	if (dentry->d_fsdata)
39 		return 0;
40 
41 	di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
42 	if (!di)
43 		return -ENOMEM;          /* oh well */
44 
45 	spin_lock(&dentry->d_lock);
46 	if (dentry->d_fsdata) {
47 		/* lost a race */
48 		kmem_cache_free(ceph_dentry_cachep, di);
49 		goto out_unlock;
50 	}
51 
52 	if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_NOSNAP)
53 		d_set_d_op(dentry, &ceph_dentry_ops);
54 	else if (ceph_snap(d_inode(dentry->d_parent)) == CEPH_SNAPDIR)
55 		d_set_d_op(dentry, &ceph_snapdir_dentry_ops);
56 	else
57 		d_set_d_op(dentry, &ceph_snap_dentry_ops);
58 
59 	di->dentry = dentry;
60 	di->lease_session = NULL;
61 	dentry->d_time = jiffies;
62 	/* avoid reordering d_fsdata setup so that the check above is safe */
63 	smp_mb();
64 	dentry->d_fsdata = di;
65 	ceph_dentry_lru_add(dentry);
66 out_unlock:
67 	spin_unlock(&dentry->d_lock);
68 	return 0;
69 }
70 
71 /*
72  * for readdir, we encode the directory frag and offset within that
73  * frag into f_pos.
74  */
75 static unsigned fpos_frag(loff_t p)
76 {
77 	return p >> 32;
78 }
79 static unsigned fpos_off(loff_t p)
80 {
81 	return p & 0xffffffff;
82 }
83 
84 static int fpos_cmp(loff_t l, loff_t r)
85 {
86 	int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
87 	if (v)
88 		return v;
89 	return (int)(fpos_off(l) - fpos_off(r));
90 }
91 
92 /*
93  * make note of the last dentry we read, so we can
94  * continue at the same lexicographical point,
95  * regardless of what dir changes take place on the
96  * server.
97  */
98 static int note_last_dentry(struct ceph_file_info *fi, const char *name,
99 		            int len, unsigned next_offset)
100 {
101 	char *buf = kmalloc(len+1, GFP_KERNEL);
102 	if (!buf)
103 		return -ENOMEM;
104 	kfree(fi->last_name);
105 	fi->last_name = buf;
106 	memcpy(fi->last_name, name, len);
107 	fi->last_name[len] = 0;
108 	fi->next_offset = next_offset;
109 	dout("note_last_dentry '%s'\n", fi->last_name);
110 	return 0;
111 }
112 
113 
114 static struct dentry *
115 __dcache_find_get_entry(struct dentry *parent, u64 idx,
116 			struct ceph_readdir_cache_control *cache_ctl)
117 {
118 	struct inode *dir = d_inode(parent);
119 	struct dentry *dentry;
120 	unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
121 	loff_t ptr_pos = idx * sizeof(struct dentry *);
122 	pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
123 
124 	if (ptr_pos >= i_size_read(dir))
125 		return NULL;
126 
127 	if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
128 		ceph_readdir_cache_release(cache_ctl);
129 		cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
130 		if (!cache_ctl->page) {
131 			dout(" page %lu not found\n", ptr_pgoff);
132 			return ERR_PTR(-EAGAIN);
133 		}
134 		/* reading/filling the cache are serialized by
135 		   i_mutex, no need to use page lock */
136 		unlock_page(cache_ctl->page);
137 		cache_ctl->dentries = kmap(cache_ctl->page);
138 	}
139 
140 	cache_ctl->index = idx & idx_mask;
141 
142 	rcu_read_lock();
143 	spin_lock(&parent->d_lock);
144 	/* check i_size again here, because empty directory can be
145 	 * marked as complete while not holding the i_mutex. */
146 	if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
147 		dentry = cache_ctl->dentries[cache_ctl->index];
148 	else
149 		dentry = NULL;
150 	spin_unlock(&parent->d_lock);
151 	if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
152 		dentry = NULL;
153 	rcu_read_unlock();
154 	return dentry ? : ERR_PTR(-EAGAIN);
155 }
156 
157 /*
158  * When possible, we try to satisfy a readdir by peeking at the
159  * dcache.  We make this work by carefully ordering dentries on
160  * d_child when we initially get results back from the MDS, and
161  * falling back to a "normal" sync readdir if any dentries in the dir
162  * are dropped.
163  *
164  * Complete dir indicates that we have all dentries in the dir.  It is
165  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
166  * the MDS if/when the directory is modified).
167  */
168 static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
169 			    u32 shared_gen)
170 {
171 	struct ceph_file_info *fi = file->private_data;
172 	struct dentry *parent = file->f_path.dentry;
173 	struct inode *dir = d_inode(parent);
174 	struct dentry *dentry, *last = NULL;
175 	struct ceph_dentry_info *di;
176 	struct ceph_readdir_cache_control cache_ctl = {};
177 	u64 idx = 0;
178 	int err = 0;
179 
180 	dout("__dcache_readdir %p v%u at %llu\n", dir, shared_gen, ctx->pos);
181 
182 	/* search start position */
183 	if (ctx->pos > 2) {
184 		u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
185 		while (count > 0) {
186 			u64 step = count >> 1;
187 			dentry = __dcache_find_get_entry(parent, idx + step,
188 							 &cache_ctl);
189 			if (!dentry) {
190 				/* use linar search */
191 				idx = 0;
192 				break;
193 			}
194 			if (IS_ERR(dentry)) {
195 				err = PTR_ERR(dentry);
196 				goto out;
197 			}
198 			di = ceph_dentry(dentry);
199 			spin_lock(&dentry->d_lock);
200 			if (fpos_cmp(di->offset, ctx->pos) < 0) {
201 				idx += step + 1;
202 				count -= step + 1;
203 			} else {
204 				count = step;
205 			}
206 			spin_unlock(&dentry->d_lock);
207 			dput(dentry);
208 		}
209 
210 		dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
211 	}
212 
213 
214 	for (;;) {
215 		bool emit_dentry = false;
216 		dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
217 		if (!dentry) {
218 			fi->flags |= CEPH_F_ATEND;
219 			err = 0;
220 			break;
221 		}
222 		if (IS_ERR(dentry)) {
223 			err = PTR_ERR(dentry);
224 			goto out;
225 		}
226 
227 		di = ceph_dentry(dentry);
228 		spin_lock(&dentry->d_lock);
229 		if (di->lease_shared_gen == shared_gen &&
230 		    d_really_is_positive(dentry) &&
231 		    fpos_cmp(ctx->pos, di->offset) <= 0) {
232 			emit_dentry = true;
233 		}
234 		spin_unlock(&dentry->d_lock);
235 
236 		if (emit_dentry) {
237 			dout(" %llu (%llu) dentry %p %pd %p\n", di->offset, ctx->pos,
238 			     dentry, dentry, d_inode(dentry));
239 			ctx->pos = di->offset;
240 			if (!dir_emit(ctx, dentry->d_name.name,
241 				      dentry->d_name.len,
242 				      ceph_translate_ino(dentry->d_sb,
243 							 d_inode(dentry)->i_ino),
244 				      d_inode(dentry)->i_mode >> 12)) {
245 				dput(dentry);
246 				err = 0;
247 				break;
248 			}
249 			ctx->pos++;
250 
251 			if (last)
252 				dput(last);
253 			last = dentry;
254 		} else {
255 			dput(dentry);
256 		}
257 	}
258 out:
259 	ceph_readdir_cache_release(&cache_ctl);
260 	if (last) {
261 		int ret;
262 		di = ceph_dentry(last);
263 		ret = note_last_dentry(fi, last->d_name.name, last->d_name.len,
264 				       fpos_off(di->offset) + 1);
265 		if (ret < 0)
266 			err = ret;
267 		dput(last);
268 	}
269 	return err;
270 }
271 
272 static int ceph_readdir(struct file *file, struct dir_context *ctx)
273 {
274 	struct ceph_file_info *fi = file->private_data;
275 	struct inode *inode = file_inode(file);
276 	struct ceph_inode_info *ci = ceph_inode(inode);
277 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
278 	struct ceph_mds_client *mdsc = fsc->mdsc;
279 	unsigned frag = fpos_frag(ctx->pos);
280 	int off = fpos_off(ctx->pos);
281 	int err;
282 	u32 ftype;
283 	struct ceph_mds_reply_info_parsed *rinfo;
284 
285 	dout("readdir %p file %p frag %u off %u\n", inode, file, frag, off);
286 	if (fi->flags & CEPH_F_ATEND)
287 		return 0;
288 
289 	/* always start with . and .. */
290 	if (ctx->pos == 0) {
291 		dout("readdir off 0 -> '.'\n");
292 		if (!dir_emit(ctx, ".", 1,
293 			    ceph_translate_ino(inode->i_sb, inode->i_ino),
294 			    inode->i_mode >> 12))
295 			return 0;
296 		ctx->pos = 1;
297 		off = 1;
298 	}
299 	if (ctx->pos == 1) {
300 		ino_t ino = parent_ino(file->f_path.dentry);
301 		dout("readdir off 1 -> '..'\n");
302 		if (!dir_emit(ctx, "..", 2,
303 			    ceph_translate_ino(inode->i_sb, ino),
304 			    inode->i_mode >> 12))
305 			return 0;
306 		ctx->pos = 2;
307 		off = 2;
308 	}
309 
310 	/* can we use the dcache? */
311 	spin_lock(&ci->i_ceph_lock);
312 	if (ceph_test_mount_opt(fsc, DCACHE) &&
313 	    !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
314 	    ceph_snap(inode) != CEPH_SNAPDIR &&
315 	    __ceph_dir_is_complete_ordered(ci) &&
316 	    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
317 		u32 shared_gen = ci->i_shared_gen;
318 		spin_unlock(&ci->i_ceph_lock);
319 		err = __dcache_readdir(file, ctx, shared_gen);
320 		if (err != -EAGAIN)
321 			return err;
322 		frag = fpos_frag(ctx->pos);
323 		off = fpos_off(ctx->pos);
324 	} else {
325 		spin_unlock(&ci->i_ceph_lock);
326 	}
327 
328 	/* proceed with a normal readdir */
329 more:
330 	/* do we have the correct frag content buffered? */
331 	if (fi->frag != frag || fi->last_readdir == NULL) {
332 		struct ceph_mds_request *req;
333 		int op = ceph_snap(inode) == CEPH_SNAPDIR ?
334 			CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
335 
336 		/* discard old result, if any */
337 		if (fi->last_readdir) {
338 			ceph_mdsc_put_request(fi->last_readdir);
339 			fi->last_readdir = NULL;
340 		}
341 
342 		dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
343 		     ceph_vinop(inode), frag, fi->last_name);
344 		req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
345 		if (IS_ERR(req))
346 			return PTR_ERR(req);
347 		err = ceph_alloc_readdir_reply_buffer(req, inode);
348 		if (err) {
349 			ceph_mdsc_put_request(req);
350 			return err;
351 		}
352 		/* hints to request -> mds selection code */
353 		req->r_direct_mode = USE_AUTH_MDS;
354 		req->r_direct_hash = ceph_frag_value(frag);
355 		req->r_direct_is_hash = true;
356 		if (fi->last_name) {
357 			req->r_path2 = kstrdup(fi->last_name, GFP_KERNEL);
358 			if (!req->r_path2) {
359 				ceph_mdsc_put_request(req);
360 				return -ENOMEM;
361 			}
362 		}
363 		req->r_dir_release_cnt = fi->dir_release_count;
364 		req->r_dir_ordered_cnt = fi->dir_ordered_count;
365 		req->r_readdir_cache_idx = fi->readdir_cache_idx;
366 		req->r_readdir_offset = fi->next_offset;
367 		req->r_args.readdir.frag = cpu_to_le32(frag);
368 
369 		req->r_inode = inode;
370 		ihold(inode);
371 		req->r_dentry = dget(file->f_path.dentry);
372 		err = ceph_mdsc_do_request(mdsc, NULL, req);
373 		if (err < 0) {
374 			ceph_mdsc_put_request(req);
375 			return err;
376 		}
377 		dout("readdir got and parsed readdir result=%d"
378 		     " on frag %x, end=%d, complete=%d\n", err, frag,
379 		     (int)req->r_reply_info.dir_end,
380 		     (int)req->r_reply_info.dir_complete);
381 
382 
383 		/* note next offset and last dentry name */
384 		rinfo = &req->r_reply_info;
385 		if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
386 			frag = le32_to_cpu(rinfo->dir_dir->frag);
387 			off = req->r_readdir_offset;
388 			fi->next_offset = off;
389 		}
390 
391 		fi->frag = frag;
392 		fi->offset = fi->next_offset;
393 		fi->last_readdir = req;
394 
395 		if (req->r_did_prepopulate) {
396 			fi->readdir_cache_idx = req->r_readdir_cache_idx;
397 			if (fi->readdir_cache_idx < 0) {
398 				/* preclude from marking dir ordered */
399 				fi->dir_ordered_count = 0;
400 			} else if (ceph_frag_is_leftmost(frag) && off == 2) {
401 				/* note dir version at start of readdir so
402 				 * we can tell if any dentries get dropped */
403 				fi->dir_release_count = req->r_dir_release_cnt;
404 				fi->dir_ordered_count = req->r_dir_ordered_cnt;
405 			}
406 		} else {
407 			dout("readdir !did_prepopulate");
408 			/* disable readdir cache */
409 			fi->readdir_cache_idx = -1;
410 			/* preclude from marking dir complete */
411 			fi->dir_release_count = 0;
412 		}
413 
414 		if (req->r_reply_info.dir_end) {
415 			kfree(fi->last_name);
416 			fi->last_name = NULL;
417 			fi->next_offset = 2;
418 		} else {
419 			struct ceph_mds_reply_dir_entry *rde =
420 					rinfo->dir_entries + (rinfo->dir_nr-1);
421 			err = note_last_dentry(fi, rde->name, rde->name_len,
422 				       fi->next_offset + rinfo->dir_nr);
423 			if (err)
424 				return err;
425 		}
426 	}
427 
428 	rinfo = &fi->last_readdir->r_reply_info;
429 	dout("readdir frag %x num %d off %d chunkoff %d\n", frag,
430 	     rinfo->dir_nr, off, fi->offset);
431 
432 	ctx->pos = ceph_make_fpos(frag, off);
433 	while (off >= fi->offset && off - fi->offset < rinfo->dir_nr) {
434 		struct ceph_mds_reply_dir_entry *rde =
435 			rinfo->dir_entries + (off - fi->offset);
436 		struct ceph_vino vino;
437 		ino_t ino;
438 
439 		dout("readdir off %d (%d/%d) -> %lld '%.*s' %p\n",
440 		     off, off - fi->offset, rinfo->dir_nr, ctx->pos,
441 		     rde->name_len, rde->name, &rde->inode.in);
442 		BUG_ON(!rde->inode.in);
443 		ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
444 		vino.ino = le64_to_cpu(rde->inode.in->ino);
445 		vino.snap = le64_to_cpu(rde->inode.in->snapid);
446 		ino = ceph_vino_to_ino(vino);
447 		if (!dir_emit(ctx, rde->name, rde->name_len,
448 			      ceph_translate_ino(inode->i_sb, ino), ftype)) {
449 			dout("filldir stopping us...\n");
450 			return 0;
451 		}
452 		off++;
453 		ctx->pos++;
454 	}
455 
456 	if (fi->last_name) {
457 		ceph_mdsc_put_request(fi->last_readdir);
458 		fi->last_readdir = NULL;
459 		goto more;
460 	}
461 
462 	/* more frags? */
463 	if (!ceph_frag_is_rightmost(frag)) {
464 		frag = ceph_frag_next(frag);
465 		off = 2;
466 		ctx->pos = ceph_make_fpos(frag, off);
467 		dout("readdir next frag is %x\n", frag);
468 		goto more;
469 	}
470 	fi->flags |= CEPH_F_ATEND;
471 
472 	/*
473 	 * if dir_release_count still matches the dir, no dentries
474 	 * were released during the whole readdir, and we should have
475 	 * the complete dir contents in our cache.
476 	 */
477 	if (atomic64_read(&ci->i_release_count) == fi->dir_release_count) {
478 		spin_lock(&ci->i_ceph_lock);
479 		if (fi->dir_ordered_count == atomic64_read(&ci->i_ordered_count)) {
480 			dout(" marking %p complete and ordered\n", inode);
481 			/* use i_size to track number of entries in
482 			 * readdir cache */
483 			BUG_ON(fi->readdir_cache_idx < 0);
484 			i_size_write(inode, fi->readdir_cache_idx *
485 				     sizeof(struct dentry*));
486 		} else {
487 			dout(" marking %p complete\n", inode);
488 		}
489 		__ceph_dir_set_complete(ci, fi->dir_release_count,
490 					fi->dir_ordered_count);
491 		spin_unlock(&ci->i_ceph_lock);
492 	}
493 
494 	dout("readdir %p file %p done.\n", inode, file);
495 	return 0;
496 }
497 
498 static void reset_readdir(struct ceph_file_info *fi, unsigned frag)
499 {
500 	if (fi->last_readdir) {
501 		ceph_mdsc_put_request(fi->last_readdir);
502 		fi->last_readdir = NULL;
503 	}
504 	kfree(fi->last_name);
505 	fi->last_name = NULL;
506 	fi->dir_release_count = 0;
507 	fi->readdir_cache_idx = -1;
508 	fi->next_offset = 2;  /* compensate for . and .. */
509 	fi->flags &= ~CEPH_F_ATEND;
510 }
511 
512 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
513 {
514 	struct ceph_file_info *fi = file->private_data;
515 	struct inode *inode = file->f_mapping->host;
516 	loff_t old_offset = ceph_make_fpos(fi->frag, fi->next_offset);
517 	loff_t retval;
518 
519 	inode_lock(inode);
520 	retval = -EINVAL;
521 	switch (whence) {
522 	case SEEK_CUR:
523 		offset += file->f_pos;
524 	case SEEK_SET:
525 		break;
526 	case SEEK_END:
527 		retval = -EOPNOTSUPP;
528 	default:
529 		goto out;
530 	}
531 
532 	if (offset >= 0) {
533 		if (offset != file->f_pos) {
534 			file->f_pos = offset;
535 			file->f_version = 0;
536 			fi->flags &= ~CEPH_F_ATEND;
537 		}
538 		retval = offset;
539 
540 		if (offset == 0 ||
541 		    fpos_frag(offset) != fi->frag ||
542 		    fpos_off(offset) < fi->offset) {
543 			/* discard buffered readdir content on seekdir(0), or
544 			 * seek to new frag, or seek prior to current chunk */
545 			dout("dir_llseek dropping %p content\n", file);
546 			reset_readdir(fi, fpos_frag(offset));
547 		} else if (fpos_cmp(offset, old_offset) > 0) {
548 			/* reset dir_release_count if we did a forward seek */
549 			fi->dir_release_count = 0;
550 			fi->readdir_cache_idx = -1;
551 		}
552 	}
553 out:
554 	inode_unlock(inode);
555 	return retval;
556 }
557 
558 /*
559  * Handle lookups for the hidden .snap directory.
560  */
561 int ceph_handle_snapdir(struct ceph_mds_request *req,
562 			struct dentry *dentry, int err)
563 {
564 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
565 	struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
566 
567 	/* .snap dir? */
568 	if (err == -ENOENT &&
569 	    ceph_snap(parent) == CEPH_NOSNAP &&
570 	    strcmp(dentry->d_name.name,
571 		   fsc->mount_options->snapdir_name) == 0) {
572 		struct inode *inode = ceph_get_snapdir(parent);
573 		dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
574 		     dentry, dentry, inode);
575 		BUG_ON(!d_unhashed(dentry));
576 		d_add(dentry, inode);
577 		err = 0;
578 	}
579 	return err;
580 }
581 
582 /*
583  * Figure out final result of a lookup/open request.
584  *
585  * Mainly, make sure we return the final req->r_dentry (if it already
586  * existed) in place of the original VFS-provided dentry when they
587  * differ.
588  *
589  * Gracefully handle the case where the MDS replies with -ENOENT and
590  * no trace (which it may do, at its discretion, e.g., if it doesn't
591  * care to issue a lease on the negative dentry).
592  */
593 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
594 				  struct dentry *dentry, int err)
595 {
596 	if (err == -ENOENT) {
597 		/* no trace? */
598 		err = 0;
599 		if (!req->r_reply_info.head->is_dentry) {
600 			dout("ENOENT and no trace, dentry %p inode %p\n",
601 			     dentry, d_inode(dentry));
602 			if (d_really_is_positive(dentry)) {
603 				d_drop(dentry);
604 				err = -ENOENT;
605 			} else {
606 				d_add(dentry, NULL);
607 			}
608 		}
609 	}
610 	if (err)
611 		dentry = ERR_PTR(err);
612 	else if (dentry != req->r_dentry)
613 		dentry = dget(req->r_dentry);   /* we got spliced */
614 	else
615 		dentry = NULL;
616 	return dentry;
617 }
618 
619 static int is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
620 {
621 	return ceph_ino(inode) == CEPH_INO_ROOT &&
622 		strncmp(dentry->d_name.name, ".ceph", 5) == 0;
623 }
624 
625 /*
626  * Look up a single dir entry.  If there is a lookup intent, inform
627  * the MDS so that it gets our 'caps wanted' value in a single op.
628  */
629 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
630 				  unsigned int flags)
631 {
632 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
633 	struct ceph_mds_client *mdsc = fsc->mdsc;
634 	struct ceph_mds_request *req;
635 	int op;
636 	int mask;
637 	int err;
638 
639 	dout("lookup %p dentry %p '%pd'\n",
640 	     dir, dentry, dentry);
641 
642 	if (dentry->d_name.len > NAME_MAX)
643 		return ERR_PTR(-ENAMETOOLONG);
644 
645 	err = ceph_init_dentry(dentry);
646 	if (err < 0)
647 		return ERR_PTR(err);
648 
649 	/* can we conclude ENOENT locally? */
650 	if (d_really_is_negative(dentry)) {
651 		struct ceph_inode_info *ci = ceph_inode(dir);
652 		struct ceph_dentry_info *di = ceph_dentry(dentry);
653 
654 		spin_lock(&ci->i_ceph_lock);
655 		dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
656 		if (strncmp(dentry->d_name.name,
657 			    fsc->mount_options->snapdir_name,
658 			    dentry->d_name.len) &&
659 		    !is_root_ceph_dentry(dir, dentry) &&
660 		    ceph_test_mount_opt(fsc, DCACHE) &&
661 		    __ceph_dir_is_complete(ci) &&
662 		    (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
663 			spin_unlock(&ci->i_ceph_lock);
664 			dout(" dir %p complete, -ENOENT\n", dir);
665 			d_add(dentry, NULL);
666 			di->lease_shared_gen = ci->i_shared_gen;
667 			return NULL;
668 		}
669 		spin_unlock(&ci->i_ceph_lock);
670 	}
671 
672 	op = ceph_snap(dir) == CEPH_SNAPDIR ?
673 		CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
674 	req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
675 	if (IS_ERR(req))
676 		return ERR_CAST(req);
677 	req->r_dentry = dget(dentry);
678 	req->r_num_caps = 2;
679 
680 	mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
681 	if (ceph_security_xattr_wanted(dir))
682 		mask |= CEPH_CAP_XATTR_SHARED;
683 	req->r_args.getattr.mask = cpu_to_le32(mask);
684 
685 	req->r_locked_dir = dir;
686 	err = ceph_mdsc_do_request(mdsc, NULL, req);
687 	err = ceph_handle_snapdir(req, dentry, err);
688 	dentry = ceph_finish_lookup(req, dentry, err);
689 	ceph_mdsc_put_request(req);  /* will dput(dentry) */
690 	dout("lookup result=%p\n", dentry);
691 	return dentry;
692 }
693 
694 /*
695  * If we do a create but get no trace back from the MDS, follow up with
696  * a lookup (the VFS expects us to link up the provided dentry).
697  */
698 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
699 {
700 	struct dentry *result = ceph_lookup(dir, dentry, 0);
701 
702 	if (result && !IS_ERR(result)) {
703 		/*
704 		 * We created the item, then did a lookup, and found
705 		 * it was already linked to another inode we already
706 		 * had in our cache (and thus got spliced). To not
707 		 * confuse VFS (especially when inode is a directory),
708 		 * we don't link our dentry to that inode, return an
709 		 * error instead.
710 		 *
711 		 * This event should be rare and it happens only when
712 		 * we talk to old MDS. Recent MDS does not send traceless
713 		 * reply for request that creates new inode.
714 		 */
715 		d_drop(result);
716 		return -ESTALE;
717 	}
718 	return PTR_ERR(result);
719 }
720 
721 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
722 		      umode_t mode, dev_t rdev)
723 {
724 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
725 	struct ceph_mds_client *mdsc = fsc->mdsc;
726 	struct ceph_mds_request *req;
727 	struct ceph_acls_info acls = {};
728 	int err;
729 
730 	if (ceph_snap(dir) != CEPH_NOSNAP)
731 		return -EROFS;
732 
733 	err = ceph_pre_init_acls(dir, &mode, &acls);
734 	if (err < 0)
735 		return err;
736 
737 	dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
738 	     dir, dentry, mode, rdev);
739 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
740 	if (IS_ERR(req)) {
741 		err = PTR_ERR(req);
742 		goto out;
743 	}
744 	req->r_dentry = dget(dentry);
745 	req->r_num_caps = 2;
746 	req->r_locked_dir = dir;
747 	req->r_args.mknod.mode = cpu_to_le32(mode);
748 	req->r_args.mknod.rdev = cpu_to_le32(rdev);
749 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
750 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
751 	if (acls.pagelist) {
752 		req->r_pagelist = acls.pagelist;
753 		acls.pagelist = NULL;
754 	}
755 	err = ceph_mdsc_do_request(mdsc, dir, req);
756 	if (!err && !req->r_reply_info.head->is_dentry)
757 		err = ceph_handle_notrace_create(dir, dentry);
758 	ceph_mdsc_put_request(req);
759 out:
760 	if (!err)
761 		ceph_init_inode_acls(d_inode(dentry), &acls);
762 	else
763 		d_drop(dentry);
764 	ceph_release_acls_info(&acls);
765 	return err;
766 }
767 
768 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
769 		       bool excl)
770 {
771 	return ceph_mknod(dir, dentry, mode, 0);
772 }
773 
774 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
775 			    const char *dest)
776 {
777 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
778 	struct ceph_mds_client *mdsc = fsc->mdsc;
779 	struct ceph_mds_request *req;
780 	int err;
781 
782 	if (ceph_snap(dir) != CEPH_NOSNAP)
783 		return -EROFS;
784 
785 	dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
786 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
787 	if (IS_ERR(req)) {
788 		err = PTR_ERR(req);
789 		goto out;
790 	}
791 	req->r_path2 = kstrdup(dest, GFP_KERNEL);
792 	if (!req->r_path2) {
793 		err = -ENOMEM;
794 		ceph_mdsc_put_request(req);
795 		goto out;
796 	}
797 	req->r_locked_dir = dir;
798 	req->r_dentry = dget(dentry);
799 	req->r_num_caps = 2;
800 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
801 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
802 	err = ceph_mdsc_do_request(mdsc, dir, req);
803 	if (!err && !req->r_reply_info.head->is_dentry)
804 		err = ceph_handle_notrace_create(dir, dentry);
805 	ceph_mdsc_put_request(req);
806 out:
807 	if (err)
808 		d_drop(dentry);
809 	return err;
810 }
811 
812 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
813 {
814 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
815 	struct ceph_mds_client *mdsc = fsc->mdsc;
816 	struct ceph_mds_request *req;
817 	struct ceph_acls_info acls = {};
818 	int err = -EROFS;
819 	int op;
820 
821 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
822 		/* mkdir .snap/foo is a MKSNAP */
823 		op = CEPH_MDS_OP_MKSNAP;
824 		dout("mksnap dir %p snap '%pd' dn %p\n", dir,
825 		     dentry, dentry);
826 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
827 		dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
828 		op = CEPH_MDS_OP_MKDIR;
829 	} else {
830 		goto out;
831 	}
832 
833 	mode |= S_IFDIR;
834 	err = ceph_pre_init_acls(dir, &mode, &acls);
835 	if (err < 0)
836 		goto out;
837 
838 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
839 	if (IS_ERR(req)) {
840 		err = PTR_ERR(req);
841 		goto out;
842 	}
843 
844 	req->r_dentry = dget(dentry);
845 	req->r_num_caps = 2;
846 	req->r_locked_dir = dir;
847 	req->r_args.mkdir.mode = cpu_to_le32(mode);
848 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
849 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
850 	if (acls.pagelist) {
851 		req->r_pagelist = acls.pagelist;
852 		acls.pagelist = NULL;
853 	}
854 	err = ceph_mdsc_do_request(mdsc, dir, req);
855 	if (!err &&
856 	    !req->r_reply_info.head->is_target &&
857 	    !req->r_reply_info.head->is_dentry)
858 		err = ceph_handle_notrace_create(dir, dentry);
859 	ceph_mdsc_put_request(req);
860 out:
861 	if (!err)
862 		ceph_init_inode_acls(d_inode(dentry), &acls);
863 	else
864 		d_drop(dentry);
865 	ceph_release_acls_info(&acls);
866 	return err;
867 }
868 
869 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
870 		     struct dentry *dentry)
871 {
872 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
873 	struct ceph_mds_client *mdsc = fsc->mdsc;
874 	struct ceph_mds_request *req;
875 	int err;
876 
877 	if (ceph_snap(dir) != CEPH_NOSNAP)
878 		return -EROFS;
879 
880 	dout("link in dir %p old_dentry %p dentry %p\n", dir,
881 	     old_dentry, dentry);
882 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
883 	if (IS_ERR(req)) {
884 		d_drop(dentry);
885 		return PTR_ERR(req);
886 	}
887 	req->r_dentry = dget(dentry);
888 	req->r_num_caps = 2;
889 	req->r_old_dentry = dget(old_dentry);
890 	req->r_locked_dir = dir;
891 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
892 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
893 	/* release LINK_SHARED on source inode (mds will lock it) */
894 	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
895 	err = ceph_mdsc_do_request(mdsc, dir, req);
896 	if (err) {
897 		d_drop(dentry);
898 	} else if (!req->r_reply_info.head->is_dentry) {
899 		ihold(d_inode(old_dentry));
900 		d_instantiate(dentry, d_inode(old_dentry));
901 	}
902 	ceph_mdsc_put_request(req);
903 	return err;
904 }
905 
906 /*
907  * For a soon-to-be unlinked file, drop the AUTH_RDCACHE caps.  If it
908  * looks like the link count will hit 0, drop any other caps (other
909  * than PIN) we don't specifically want (due to the file still being
910  * open).
911  */
912 static int drop_caps_for_unlink(struct inode *inode)
913 {
914 	struct ceph_inode_info *ci = ceph_inode(inode);
915 	int drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
916 
917 	spin_lock(&ci->i_ceph_lock);
918 	if (inode->i_nlink == 1) {
919 		drop |= ~(__ceph_caps_wanted(ci) | CEPH_CAP_PIN);
920 		ci->i_ceph_flags |= CEPH_I_NODELAY;
921 	}
922 	spin_unlock(&ci->i_ceph_lock);
923 	return drop;
924 }
925 
926 /*
927  * rmdir and unlink are differ only by the metadata op code
928  */
929 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
930 {
931 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
932 	struct ceph_mds_client *mdsc = fsc->mdsc;
933 	struct inode *inode = d_inode(dentry);
934 	struct ceph_mds_request *req;
935 	int err = -EROFS;
936 	int op;
937 
938 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
939 		/* rmdir .snap/foo is RMSNAP */
940 		dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
941 		op = CEPH_MDS_OP_RMSNAP;
942 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
943 		dout("unlink/rmdir dir %p dn %p inode %p\n",
944 		     dir, dentry, inode);
945 		op = d_is_dir(dentry) ?
946 			CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
947 	} else
948 		goto out;
949 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
950 	if (IS_ERR(req)) {
951 		err = PTR_ERR(req);
952 		goto out;
953 	}
954 	req->r_dentry = dget(dentry);
955 	req->r_num_caps = 2;
956 	req->r_locked_dir = dir;
957 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
958 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
959 	req->r_inode_drop = drop_caps_for_unlink(inode);
960 	err = ceph_mdsc_do_request(mdsc, dir, req);
961 	if (!err && !req->r_reply_info.head->is_dentry)
962 		d_delete(dentry);
963 	ceph_mdsc_put_request(req);
964 out:
965 	return err;
966 }
967 
968 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
969 		       struct inode *new_dir, struct dentry *new_dentry)
970 {
971 	struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
972 	struct ceph_mds_client *mdsc = fsc->mdsc;
973 	struct ceph_mds_request *req;
974 	int op = CEPH_MDS_OP_RENAME;
975 	int err;
976 
977 	if (ceph_snap(old_dir) != ceph_snap(new_dir))
978 		return -EXDEV;
979 	if (ceph_snap(old_dir) != CEPH_NOSNAP) {
980 		if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
981 			op = CEPH_MDS_OP_RENAMESNAP;
982 		else
983 			return -EROFS;
984 	}
985 	dout("rename dir %p dentry %p to dir %p dentry %p\n",
986 	     old_dir, old_dentry, new_dir, new_dentry);
987 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
988 	if (IS_ERR(req))
989 		return PTR_ERR(req);
990 	ihold(old_dir);
991 	req->r_dentry = dget(new_dentry);
992 	req->r_num_caps = 2;
993 	req->r_old_dentry = dget(old_dentry);
994 	req->r_old_dentry_dir = old_dir;
995 	req->r_locked_dir = new_dir;
996 	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
997 	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
998 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
999 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1000 	/* release LINK_RDCACHE on source inode (mds will lock it) */
1001 	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED;
1002 	if (d_really_is_positive(new_dentry))
1003 		req->r_inode_drop = drop_caps_for_unlink(d_inode(new_dentry));
1004 	err = ceph_mdsc_do_request(mdsc, old_dir, req);
1005 	if (!err && !req->r_reply_info.head->is_dentry) {
1006 		/*
1007 		 * Normally d_move() is done by fill_trace (called by
1008 		 * do_request, above).  If there is no trace, we need
1009 		 * to do it here.
1010 		 */
1011 
1012 		/* d_move screws up sibling dentries' offsets */
1013 		ceph_dir_clear_complete(old_dir);
1014 		ceph_dir_clear_complete(new_dir);
1015 
1016 		d_move(old_dentry, new_dentry);
1017 
1018 		/* ensure target dentry is invalidated, despite
1019 		   rehashing bug in vfs_rename_dir */
1020 		ceph_invalidate_dentry_lease(new_dentry);
1021 	}
1022 	ceph_mdsc_put_request(req);
1023 	return err;
1024 }
1025 
1026 /*
1027  * Ensure a dentry lease will no longer revalidate.
1028  */
1029 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1030 {
1031 	spin_lock(&dentry->d_lock);
1032 	dentry->d_time = jiffies;
1033 	ceph_dentry(dentry)->lease_shared_gen = 0;
1034 	spin_unlock(&dentry->d_lock);
1035 }
1036 
1037 /*
1038  * Check if dentry lease is valid.  If not, delete the lease.  Try to
1039  * renew if the least is more than half up.
1040  */
1041 static int dentry_lease_is_valid(struct dentry *dentry)
1042 {
1043 	struct ceph_dentry_info *di;
1044 	struct ceph_mds_session *s;
1045 	int valid = 0;
1046 	u32 gen;
1047 	unsigned long ttl;
1048 	struct ceph_mds_session *session = NULL;
1049 	struct inode *dir = NULL;
1050 	u32 seq = 0;
1051 
1052 	spin_lock(&dentry->d_lock);
1053 	di = ceph_dentry(dentry);
1054 	if (di->lease_session) {
1055 		s = di->lease_session;
1056 		spin_lock(&s->s_gen_ttl_lock);
1057 		gen = s->s_cap_gen;
1058 		ttl = s->s_cap_ttl;
1059 		spin_unlock(&s->s_gen_ttl_lock);
1060 
1061 		if (di->lease_gen == gen &&
1062 		    time_before(jiffies, dentry->d_time) &&
1063 		    time_before(jiffies, ttl)) {
1064 			valid = 1;
1065 			if (di->lease_renew_after &&
1066 			    time_after(jiffies, di->lease_renew_after)) {
1067 				/* we should renew */
1068 				dir = d_inode(dentry->d_parent);
1069 				session = ceph_get_mds_session(s);
1070 				seq = di->lease_seq;
1071 				di->lease_renew_after = 0;
1072 				di->lease_renew_from = jiffies;
1073 			}
1074 		}
1075 	}
1076 	spin_unlock(&dentry->d_lock);
1077 
1078 	if (session) {
1079 		ceph_mdsc_lease_send_msg(session, dir, dentry,
1080 					 CEPH_MDS_LEASE_RENEW, seq);
1081 		ceph_put_mds_session(session);
1082 	}
1083 	dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1084 	return valid;
1085 }
1086 
1087 /*
1088  * Check if directory-wide content lease/cap is valid.
1089  */
1090 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1091 {
1092 	struct ceph_inode_info *ci = ceph_inode(dir);
1093 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1094 	int valid = 0;
1095 
1096 	spin_lock(&ci->i_ceph_lock);
1097 	if (ci->i_shared_gen == di->lease_shared_gen)
1098 		valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1099 	spin_unlock(&ci->i_ceph_lock);
1100 	dout("dir_lease_is_valid dir %p v%u dentry %p v%u = %d\n",
1101 	     dir, (unsigned)ci->i_shared_gen, dentry,
1102 	     (unsigned)di->lease_shared_gen, valid);
1103 	return valid;
1104 }
1105 
1106 /*
1107  * Check if cached dentry can be trusted.
1108  */
1109 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1110 {
1111 	int valid = 0;
1112 	struct dentry *parent;
1113 	struct inode *dir;
1114 
1115 	if (flags & LOOKUP_RCU)
1116 		return -ECHILD;
1117 
1118 	dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1119 	     dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1120 
1121 	parent = dget_parent(dentry);
1122 	dir = d_inode(parent);
1123 
1124 	/* always trust cached snapped dentries, snapdir dentry */
1125 	if (ceph_snap(dir) != CEPH_NOSNAP) {
1126 		dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1127 		     dentry, d_inode(dentry));
1128 		valid = 1;
1129 	} else if (d_really_is_positive(dentry) &&
1130 		   ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1131 		valid = 1;
1132 	} else if (dentry_lease_is_valid(dentry) ||
1133 		   dir_lease_is_valid(dir, dentry)) {
1134 		if (d_really_is_positive(dentry))
1135 			valid = ceph_is_any_caps(d_inode(dentry));
1136 		else
1137 			valid = 1;
1138 	}
1139 
1140 	if (!valid) {
1141 		struct ceph_mds_client *mdsc =
1142 			ceph_sb_to_client(dir->i_sb)->mdsc;
1143 		struct ceph_mds_request *req;
1144 		int op, mask, err;
1145 
1146 		op = ceph_snap(dir) == CEPH_SNAPDIR ?
1147 			CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1148 		req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1149 		if (!IS_ERR(req)) {
1150 			req->r_dentry = dget(dentry);
1151 			req->r_num_caps = 2;
1152 
1153 			mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1154 			if (ceph_security_xattr_wanted(dir))
1155 				mask |= CEPH_CAP_XATTR_SHARED;
1156 			req->r_args.getattr.mask = mask;
1157 
1158 			req->r_locked_dir = dir;
1159 			err = ceph_mdsc_do_request(mdsc, NULL, req);
1160 			if (err == 0 || err == -ENOENT) {
1161 				if (dentry == req->r_dentry) {
1162 					valid = !d_unhashed(dentry);
1163 				} else {
1164 					d_invalidate(req->r_dentry);
1165 					err = -EAGAIN;
1166 				}
1167 			}
1168 			ceph_mdsc_put_request(req);
1169 			dout("d_revalidate %p lookup result=%d\n",
1170 			     dentry, err);
1171 		}
1172 	}
1173 
1174 	dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1175 	if (valid) {
1176 		ceph_dentry_lru_touch(dentry);
1177 	} else {
1178 		ceph_dir_clear_complete(dir);
1179 	}
1180 
1181 	dput(parent);
1182 	return valid;
1183 }
1184 
1185 /*
1186  * Release our ceph_dentry_info.
1187  */
1188 static void ceph_d_release(struct dentry *dentry)
1189 {
1190 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1191 
1192 	dout("d_release %p\n", dentry);
1193 	ceph_dentry_lru_del(dentry);
1194 	if (di->lease_session)
1195 		ceph_put_mds_session(di->lease_session);
1196 	kmem_cache_free(ceph_dentry_cachep, di);
1197 	dentry->d_fsdata = NULL;
1198 }
1199 
1200 static int ceph_snapdir_d_revalidate(struct dentry *dentry,
1201 					  unsigned int flags)
1202 {
1203 	/*
1204 	 * Eventually, we'll want to revalidate snapped metadata
1205 	 * too... probably...
1206 	 */
1207 	return 1;
1208 }
1209 
1210 /*
1211  * When the VFS prunes a dentry from the cache, we need to clear the
1212  * complete flag on the parent directory.
1213  *
1214  * Called under dentry->d_lock.
1215  */
1216 static void ceph_d_prune(struct dentry *dentry)
1217 {
1218 	dout("ceph_d_prune %p\n", dentry);
1219 
1220 	/* do we have a valid parent? */
1221 	if (IS_ROOT(dentry))
1222 		return;
1223 
1224 	/* if we are not hashed, we don't affect dir's completeness */
1225 	if (d_unhashed(dentry))
1226 		return;
1227 
1228 	/*
1229 	 * we hold d_lock, so d_parent is stable, and d_fsdata is never
1230 	 * cleared until d_release
1231 	 */
1232 	ceph_dir_clear_complete(d_inode(dentry->d_parent));
1233 }
1234 
1235 /*
1236  * read() on a dir.  This weird interface hack only works if mounted
1237  * with '-o dirstat'.
1238  */
1239 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1240 			     loff_t *ppos)
1241 {
1242 	struct ceph_file_info *cf = file->private_data;
1243 	struct inode *inode = file_inode(file);
1244 	struct ceph_inode_info *ci = ceph_inode(inode);
1245 	int left;
1246 	const int bufsize = 1024;
1247 
1248 	if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1249 		return -EISDIR;
1250 
1251 	if (!cf->dir_info) {
1252 		cf->dir_info = kmalloc(bufsize, GFP_KERNEL);
1253 		if (!cf->dir_info)
1254 			return -ENOMEM;
1255 		cf->dir_info_len =
1256 			snprintf(cf->dir_info, bufsize,
1257 				"entries:   %20lld\n"
1258 				" files:    %20lld\n"
1259 				" subdirs:  %20lld\n"
1260 				"rentries:  %20lld\n"
1261 				" rfiles:   %20lld\n"
1262 				" rsubdirs: %20lld\n"
1263 				"rbytes:    %20lld\n"
1264 				"rctime:    %10ld.%09ld\n",
1265 				ci->i_files + ci->i_subdirs,
1266 				ci->i_files,
1267 				ci->i_subdirs,
1268 				ci->i_rfiles + ci->i_rsubdirs,
1269 				ci->i_rfiles,
1270 				ci->i_rsubdirs,
1271 				ci->i_rbytes,
1272 				(long)ci->i_rctime.tv_sec,
1273 				(long)ci->i_rctime.tv_nsec);
1274 	}
1275 
1276 	if (*ppos >= cf->dir_info_len)
1277 		return 0;
1278 	size = min_t(unsigned, size, cf->dir_info_len-*ppos);
1279 	left = copy_to_user(buf, cf->dir_info + *ppos, size);
1280 	if (left == size)
1281 		return -EFAULT;
1282 	*ppos += (size - left);
1283 	return size - left;
1284 }
1285 
1286 /*
1287  * We maintain a private dentry LRU.
1288  *
1289  * FIXME: this needs to be changed to a per-mds lru to be useful.
1290  */
1291 void ceph_dentry_lru_add(struct dentry *dn)
1292 {
1293 	struct ceph_dentry_info *di = ceph_dentry(dn);
1294 	struct ceph_mds_client *mdsc;
1295 
1296 	dout("dentry_lru_add %p %p '%pd'\n", di, dn, dn);
1297 	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1298 	spin_lock(&mdsc->dentry_lru_lock);
1299 	list_add_tail(&di->lru, &mdsc->dentry_lru);
1300 	mdsc->num_dentry++;
1301 	spin_unlock(&mdsc->dentry_lru_lock);
1302 }
1303 
1304 void ceph_dentry_lru_touch(struct dentry *dn)
1305 {
1306 	struct ceph_dentry_info *di = ceph_dentry(dn);
1307 	struct ceph_mds_client *mdsc;
1308 
1309 	dout("dentry_lru_touch %p %p '%pd' (offset %lld)\n", di, dn, dn,
1310 	     di->offset);
1311 	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1312 	spin_lock(&mdsc->dentry_lru_lock);
1313 	list_move_tail(&di->lru, &mdsc->dentry_lru);
1314 	spin_unlock(&mdsc->dentry_lru_lock);
1315 }
1316 
1317 void ceph_dentry_lru_del(struct dentry *dn)
1318 {
1319 	struct ceph_dentry_info *di = ceph_dentry(dn);
1320 	struct ceph_mds_client *mdsc;
1321 
1322 	dout("dentry_lru_del %p %p '%pd'\n", di, dn, dn);
1323 	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1324 	spin_lock(&mdsc->dentry_lru_lock);
1325 	list_del_init(&di->lru);
1326 	mdsc->num_dentry--;
1327 	spin_unlock(&mdsc->dentry_lru_lock);
1328 }
1329 
1330 /*
1331  * Return name hash for a given dentry.  This is dependent on
1332  * the parent directory's hash function.
1333  */
1334 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1335 {
1336 	struct ceph_inode_info *dci = ceph_inode(dir);
1337 
1338 	switch (dci->i_dir_layout.dl_dir_hash) {
1339 	case 0:	/* for backward compat */
1340 	case CEPH_STR_HASH_LINUX:
1341 		return dn->d_name.hash;
1342 
1343 	default:
1344 		return ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1345 				     dn->d_name.name, dn->d_name.len);
1346 	}
1347 }
1348 
1349 const struct file_operations ceph_dir_fops = {
1350 	.read = ceph_read_dir,
1351 	.iterate = ceph_readdir,
1352 	.llseek = ceph_dir_llseek,
1353 	.open = ceph_open,
1354 	.release = ceph_release,
1355 	.unlocked_ioctl = ceph_ioctl,
1356 	.fsync = ceph_fsync,
1357 };
1358 
1359 const struct file_operations ceph_snapdir_fops = {
1360 	.iterate = ceph_readdir,
1361 	.llseek = ceph_dir_llseek,
1362 	.open = ceph_open,
1363 	.release = ceph_release,
1364 };
1365 
1366 const struct inode_operations ceph_dir_iops = {
1367 	.lookup = ceph_lookup,
1368 	.permission = ceph_permission,
1369 	.getattr = ceph_getattr,
1370 	.setattr = ceph_setattr,
1371 	.setxattr = ceph_setxattr,
1372 	.getxattr = ceph_getxattr,
1373 	.listxattr = ceph_listxattr,
1374 	.removexattr = ceph_removexattr,
1375 	.get_acl = ceph_get_acl,
1376 	.set_acl = ceph_set_acl,
1377 	.mknod = ceph_mknod,
1378 	.symlink = ceph_symlink,
1379 	.mkdir = ceph_mkdir,
1380 	.link = ceph_link,
1381 	.unlink = ceph_unlink,
1382 	.rmdir = ceph_unlink,
1383 	.rename = ceph_rename,
1384 	.create = ceph_create,
1385 	.atomic_open = ceph_atomic_open,
1386 };
1387 
1388 const struct inode_operations ceph_snapdir_iops = {
1389 	.lookup = ceph_lookup,
1390 	.permission = ceph_permission,
1391 	.getattr = ceph_getattr,
1392 	.mkdir = ceph_mkdir,
1393 	.rmdir = ceph_unlink,
1394 	.rename = ceph_rename,
1395 };
1396 
1397 const struct dentry_operations ceph_dentry_ops = {
1398 	.d_revalidate = ceph_d_revalidate,
1399 	.d_release = ceph_d_release,
1400 	.d_prune = ceph_d_prune,
1401 };
1402 
1403 const struct dentry_operations ceph_snapdir_dentry_ops = {
1404 	.d_revalidate = ceph_snapdir_d_revalidate,
1405 	.d_release = ceph_d_release,
1406 };
1407 
1408 const struct dentry_operations ceph_snap_dentry_ops = {
1409 	.d_release = ceph_d_release,
1410 	.d_prune = ceph_d_prune,
1411 };
1412