xref: /openbmc/linux/fs/ceph/dir.c (revision de528723)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/spinlock.h>
5 #include <linux/namei.h>
6 #include <linux/slab.h>
7 #include <linux/sched.h>
8 #include <linux/xattr.h>
9 
10 #include "super.h"
11 #include "mds_client.h"
12 
13 /*
14  * Directory operations: readdir, lookup, create, link, unlink,
15  * rename, etc.
16  */
17 
18 /*
19  * Ceph MDS operations are specified in terms of a base ino and
20  * relative path.  Thus, the client can specify an operation on a
21  * specific inode (e.g., a getattr due to fstat(2)), or as a path
22  * relative to, say, the root directory.
23  *
24  * Normally, we limit ourselves to strict inode ops (no path component)
25  * or dentry operations (a single path component relative to an ino).  The
26  * exception to this is open_root_dentry(), which will open the mount
27  * point by name.
28  */
29 
30 const struct dentry_operations ceph_dentry_ops;
31 
32 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di);
33 static int __dir_lease_try_check(const struct dentry *dentry);
34 
35 /*
36  * Initialize ceph dentry state.
37  */
38 static int ceph_d_init(struct dentry *dentry)
39 {
40 	struct ceph_dentry_info *di;
41 
42 	di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
43 	if (!di)
44 		return -ENOMEM;          /* oh well */
45 
46 	di->dentry = dentry;
47 	di->lease_session = NULL;
48 	di->time = jiffies;
49 	dentry->d_fsdata = di;
50 	INIT_LIST_HEAD(&di->lease_list);
51 	return 0;
52 }
53 
54 /*
55  * for f_pos for readdir:
56  * - hash order:
57  *	(0xff << 52) | ((24 bits hash) << 28) |
58  *	(the nth entry has hash collision);
59  * - frag+name order;
60  *	((frag value) << 28) | (the nth entry in frag);
61  */
62 #define OFFSET_BITS	28
63 #define OFFSET_MASK	((1 << OFFSET_BITS) - 1)
64 #define HASH_ORDER	(0xffull << (OFFSET_BITS + 24))
65 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
66 {
67 	loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
68 	if (hash_order)
69 		fpos |= HASH_ORDER;
70 	return fpos;
71 }
72 
73 static bool is_hash_order(loff_t p)
74 {
75 	return (p & HASH_ORDER) == HASH_ORDER;
76 }
77 
78 static unsigned fpos_frag(loff_t p)
79 {
80 	return p >> OFFSET_BITS;
81 }
82 
83 static unsigned fpos_hash(loff_t p)
84 {
85 	return ceph_frag_value(fpos_frag(p));
86 }
87 
88 static unsigned fpos_off(loff_t p)
89 {
90 	return p & OFFSET_MASK;
91 }
92 
93 static int fpos_cmp(loff_t l, loff_t r)
94 {
95 	int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
96 	if (v)
97 		return v;
98 	return (int)(fpos_off(l) - fpos_off(r));
99 }
100 
101 /*
102  * make note of the last dentry we read, so we can
103  * continue at the same lexicographical point,
104  * regardless of what dir changes take place on the
105  * server.
106  */
107 static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name,
108 		            int len, unsigned next_offset)
109 {
110 	char *buf = kmalloc(len+1, GFP_KERNEL);
111 	if (!buf)
112 		return -ENOMEM;
113 	kfree(dfi->last_name);
114 	dfi->last_name = buf;
115 	memcpy(dfi->last_name, name, len);
116 	dfi->last_name[len] = 0;
117 	dfi->next_offset = next_offset;
118 	dout("note_last_dentry '%s'\n", dfi->last_name);
119 	return 0;
120 }
121 
122 
123 static struct dentry *
124 __dcache_find_get_entry(struct dentry *parent, u64 idx,
125 			struct ceph_readdir_cache_control *cache_ctl)
126 {
127 	struct inode *dir = d_inode(parent);
128 	struct dentry *dentry;
129 	unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
130 	loff_t ptr_pos = idx * sizeof(struct dentry *);
131 	pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
132 
133 	if (ptr_pos >= i_size_read(dir))
134 		return NULL;
135 
136 	if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
137 		ceph_readdir_cache_release(cache_ctl);
138 		cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
139 		if (!cache_ctl->page) {
140 			dout(" page %lu not found\n", ptr_pgoff);
141 			return ERR_PTR(-EAGAIN);
142 		}
143 		/* reading/filling the cache are serialized by
144 		   i_mutex, no need to use page lock */
145 		unlock_page(cache_ctl->page);
146 		cache_ctl->dentries = kmap(cache_ctl->page);
147 	}
148 
149 	cache_ctl->index = idx & idx_mask;
150 
151 	rcu_read_lock();
152 	spin_lock(&parent->d_lock);
153 	/* check i_size again here, because empty directory can be
154 	 * marked as complete while not holding the i_mutex. */
155 	if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
156 		dentry = cache_ctl->dentries[cache_ctl->index];
157 	else
158 		dentry = NULL;
159 	spin_unlock(&parent->d_lock);
160 	if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
161 		dentry = NULL;
162 	rcu_read_unlock();
163 	return dentry ? : ERR_PTR(-EAGAIN);
164 }
165 
166 /*
167  * When possible, we try to satisfy a readdir by peeking at the
168  * dcache.  We make this work by carefully ordering dentries on
169  * d_child when we initially get results back from the MDS, and
170  * falling back to a "normal" sync readdir if any dentries in the dir
171  * are dropped.
172  *
173  * Complete dir indicates that we have all dentries in the dir.  It is
174  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
175  * the MDS if/when the directory is modified).
176  */
177 static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
178 			    int shared_gen)
179 {
180 	struct ceph_dir_file_info *dfi = file->private_data;
181 	struct dentry *parent = file->f_path.dentry;
182 	struct inode *dir = d_inode(parent);
183 	struct dentry *dentry, *last = NULL;
184 	struct ceph_dentry_info *di;
185 	struct ceph_readdir_cache_control cache_ctl = {};
186 	u64 idx = 0;
187 	int err = 0;
188 
189 	dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
190 
191 	/* search start position */
192 	if (ctx->pos > 2) {
193 		u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
194 		while (count > 0) {
195 			u64 step = count >> 1;
196 			dentry = __dcache_find_get_entry(parent, idx + step,
197 							 &cache_ctl);
198 			if (!dentry) {
199 				/* use linar search */
200 				idx = 0;
201 				break;
202 			}
203 			if (IS_ERR(dentry)) {
204 				err = PTR_ERR(dentry);
205 				goto out;
206 			}
207 			di = ceph_dentry(dentry);
208 			spin_lock(&dentry->d_lock);
209 			if (fpos_cmp(di->offset, ctx->pos) < 0) {
210 				idx += step + 1;
211 				count -= step + 1;
212 			} else {
213 				count = step;
214 			}
215 			spin_unlock(&dentry->d_lock);
216 			dput(dentry);
217 		}
218 
219 		dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
220 	}
221 
222 
223 	for (;;) {
224 		bool emit_dentry = false;
225 		dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
226 		if (!dentry) {
227 			dfi->file_info.flags |= CEPH_F_ATEND;
228 			err = 0;
229 			break;
230 		}
231 		if (IS_ERR(dentry)) {
232 			err = PTR_ERR(dentry);
233 			goto out;
234 		}
235 
236 		spin_lock(&dentry->d_lock);
237 		di = ceph_dentry(dentry);
238 		if (d_unhashed(dentry) ||
239 		    d_really_is_negative(dentry) ||
240 		    di->lease_shared_gen != shared_gen) {
241 			spin_unlock(&dentry->d_lock);
242 			dput(dentry);
243 			err = -EAGAIN;
244 			goto out;
245 		}
246 		if (fpos_cmp(ctx->pos, di->offset) <= 0) {
247 			__ceph_dentry_dir_lease_touch(di);
248 			emit_dentry = true;
249 		}
250 		spin_unlock(&dentry->d_lock);
251 
252 		if (emit_dentry) {
253 			dout(" %llx dentry %p %pd %p\n", di->offset,
254 			     dentry, dentry, d_inode(dentry));
255 			ctx->pos = di->offset;
256 			if (!dir_emit(ctx, dentry->d_name.name,
257 				      dentry->d_name.len,
258 				      ceph_translate_ino(dentry->d_sb,
259 							 d_inode(dentry)->i_ino),
260 				      d_inode(dentry)->i_mode >> 12)) {
261 				dput(dentry);
262 				err = 0;
263 				break;
264 			}
265 			ctx->pos++;
266 
267 			if (last)
268 				dput(last);
269 			last = dentry;
270 		} else {
271 			dput(dentry);
272 		}
273 	}
274 out:
275 	ceph_readdir_cache_release(&cache_ctl);
276 	if (last) {
277 		int ret;
278 		di = ceph_dentry(last);
279 		ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len,
280 				       fpos_off(di->offset) + 1);
281 		if (ret < 0)
282 			err = ret;
283 		dput(last);
284 		/* last_name no longer match cache index */
285 		if (dfi->readdir_cache_idx >= 0) {
286 			dfi->readdir_cache_idx = -1;
287 			dfi->dir_release_count = 0;
288 		}
289 	}
290 	return err;
291 }
292 
293 static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
294 {
295 	if (!dfi->last_readdir)
296 		return true;
297 	if (is_hash_order(pos))
298 		return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
299 	else
300 		return dfi->frag != fpos_frag(pos);
301 }
302 
303 static int ceph_readdir(struct file *file, struct dir_context *ctx)
304 {
305 	struct ceph_dir_file_info *dfi = file->private_data;
306 	struct inode *inode = file_inode(file);
307 	struct ceph_inode_info *ci = ceph_inode(inode);
308 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
309 	struct ceph_mds_client *mdsc = fsc->mdsc;
310 	int i;
311 	int err;
312 	unsigned frag = -1;
313 	struct ceph_mds_reply_info_parsed *rinfo;
314 
315 	dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
316 	if (dfi->file_info.flags & CEPH_F_ATEND)
317 		return 0;
318 
319 	/* always start with . and .. */
320 	if (ctx->pos == 0) {
321 		dout("readdir off 0 -> '.'\n");
322 		if (!dir_emit(ctx, ".", 1,
323 			    ceph_translate_ino(inode->i_sb, inode->i_ino),
324 			    inode->i_mode >> 12))
325 			return 0;
326 		ctx->pos = 1;
327 	}
328 	if (ctx->pos == 1) {
329 		ino_t ino = parent_ino(file->f_path.dentry);
330 		dout("readdir off 1 -> '..'\n");
331 		if (!dir_emit(ctx, "..", 2,
332 			    ceph_translate_ino(inode->i_sb, ino),
333 			    inode->i_mode >> 12))
334 			return 0;
335 		ctx->pos = 2;
336 	}
337 
338 	/* can we use the dcache? */
339 	spin_lock(&ci->i_ceph_lock);
340 	if (ceph_test_mount_opt(fsc, DCACHE) &&
341 	    !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
342 	    ceph_snap(inode) != CEPH_SNAPDIR &&
343 	    __ceph_dir_is_complete_ordered(ci) &&
344 	    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1)) {
345 		int shared_gen = atomic_read(&ci->i_shared_gen);
346 		spin_unlock(&ci->i_ceph_lock);
347 		err = __dcache_readdir(file, ctx, shared_gen);
348 		if (err != -EAGAIN)
349 			return err;
350 	} else {
351 		spin_unlock(&ci->i_ceph_lock);
352 	}
353 
354 	/* proceed with a normal readdir */
355 more:
356 	/* do we have the correct frag content buffered? */
357 	if (need_send_readdir(dfi, ctx->pos)) {
358 		struct ceph_mds_request *req;
359 		int op = ceph_snap(inode) == CEPH_SNAPDIR ?
360 			CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
361 
362 		/* discard old result, if any */
363 		if (dfi->last_readdir) {
364 			ceph_mdsc_put_request(dfi->last_readdir);
365 			dfi->last_readdir = NULL;
366 		}
367 
368 		if (is_hash_order(ctx->pos)) {
369 			/* fragtree isn't always accurate. choose frag
370 			 * based on previous reply when possible. */
371 			if (frag == (unsigned)-1)
372 				frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
373 							NULL, NULL);
374 		} else {
375 			frag = fpos_frag(ctx->pos);
376 		}
377 
378 		dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
379 		     ceph_vinop(inode), frag, dfi->last_name);
380 		req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
381 		if (IS_ERR(req))
382 			return PTR_ERR(req);
383 		err = ceph_alloc_readdir_reply_buffer(req, inode);
384 		if (err) {
385 			ceph_mdsc_put_request(req);
386 			return err;
387 		}
388 		/* hints to request -> mds selection code */
389 		req->r_direct_mode = USE_AUTH_MDS;
390 		if (op == CEPH_MDS_OP_READDIR) {
391 			req->r_direct_hash = ceph_frag_value(frag);
392 			__set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
393 			req->r_inode_drop = CEPH_CAP_FILE_EXCL;
394 		}
395 		if (dfi->last_name) {
396 			req->r_path2 = kstrdup(dfi->last_name, GFP_KERNEL);
397 			if (!req->r_path2) {
398 				ceph_mdsc_put_request(req);
399 				return -ENOMEM;
400 			}
401 		} else if (is_hash_order(ctx->pos)) {
402 			req->r_args.readdir.offset_hash =
403 				cpu_to_le32(fpos_hash(ctx->pos));
404 		}
405 
406 		req->r_dir_release_cnt = dfi->dir_release_count;
407 		req->r_dir_ordered_cnt = dfi->dir_ordered_count;
408 		req->r_readdir_cache_idx = dfi->readdir_cache_idx;
409 		req->r_readdir_offset = dfi->next_offset;
410 		req->r_args.readdir.frag = cpu_to_le32(frag);
411 		req->r_args.readdir.flags =
412 				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
413 
414 		req->r_inode = inode;
415 		ihold(inode);
416 		req->r_dentry = dget(file->f_path.dentry);
417 		err = ceph_mdsc_do_request(mdsc, NULL, req);
418 		if (err < 0) {
419 			ceph_mdsc_put_request(req);
420 			return err;
421 		}
422 		dout("readdir got and parsed readdir result=%d on "
423 		     "frag %x, end=%d, complete=%d, hash_order=%d\n",
424 		     err, frag,
425 		     (int)req->r_reply_info.dir_end,
426 		     (int)req->r_reply_info.dir_complete,
427 		     (int)req->r_reply_info.hash_order);
428 
429 		rinfo = &req->r_reply_info;
430 		if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
431 			frag = le32_to_cpu(rinfo->dir_dir->frag);
432 			if (!rinfo->hash_order) {
433 				dfi->next_offset = req->r_readdir_offset;
434 				/* adjust ctx->pos to beginning of frag */
435 				ctx->pos = ceph_make_fpos(frag,
436 							  dfi->next_offset,
437 							  false);
438 			}
439 		}
440 
441 		dfi->frag = frag;
442 		dfi->last_readdir = req;
443 
444 		if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
445 			dfi->readdir_cache_idx = req->r_readdir_cache_idx;
446 			if (dfi->readdir_cache_idx < 0) {
447 				/* preclude from marking dir ordered */
448 				dfi->dir_ordered_count = 0;
449 			} else if (ceph_frag_is_leftmost(frag) &&
450 				   dfi->next_offset == 2) {
451 				/* note dir version at start of readdir so
452 				 * we can tell if any dentries get dropped */
453 				dfi->dir_release_count = req->r_dir_release_cnt;
454 				dfi->dir_ordered_count = req->r_dir_ordered_cnt;
455 			}
456 		} else {
457 			dout("readdir !did_prepopulate\n");
458 			/* disable readdir cache */
459 			dfi->readdir_cache_idx = -1;
460 			/* preclude from marking dir complete */
461 			dfi->dir_release_count = 0;
462 		}
463 
464 		/* note next offset and last dentry name */
465 		if (rinfo->dir_nr > 0) {
466 			struct ceph_mds_reply_dir_entry *rde =
467 					rinfo->dir_entries + (rinfo->dir_nr-1);
468 			unsigned next_offset = req->r_reply_info.dir_end ?
469 					2 : (fpos_off(rde->offset) + 1);
470 			err = note_last_dentry(dfi, rde->name, rde->name_len,
471 					       next_offset);
472 			if (err)
473 				return err;
474 		} else if (req->r_reply_info.dir_end) {
475 			dfi->next_offset = 2;
476 			/* keep last name */
477 		}
478 	}
479 
480 	rinfo = &dfi->last_readdir->r_reply_info;
481 	dout("readdir frag %x num %d pos %llx chunk first %llx\n",
482 	     dfi->frag, rinfo->dir_nr, ctx->pos,
483 	     rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
484 
485 	i = 0;
486 	/* search start position */
487 	if (rinfo->dir_nr > 0) {
488 		int step, nr = rinfo->dir_nr;
489 		while (nr > 0) {
490 			step = nr >> 1;
491 			if (rinfo->dir_entries[i + step].offset < ctx->pos) {
492 				i +=  step + 1;
493 				nr -= step + 1;
494 			} else {
495 				nr = step;
496 			}
497 		}
498 	}
499 	for (; i < rinfo->dir_nr; i++) {
500 		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
501 		struct ceph_vino vino;
502 		ino_t ino;
503 		u32 ftype;
504 
505 		BUG_ON(rde->offset < ctx->pos);
506 
507 		ctx->pos = rde->offset;
508 		dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
509 		     i, rinfo->dir_nr, ctx->pos,
510 		     rde->name_len, rde->name, &rde->inode.in);
511 
512 		BUG_ON(!rde->inode.in);
513 		ftype = le32_to_cpu(rde->inode.in->mode) >> 12;
514 		vino.ino = le64_to_cpu(rde->inode.in->ino);
515 		vino.snap = le64_to_cpu(rde->inode.in->snapid);
516 		ino = ceph_vino_to_ino(vino);
517 
518 		if (!dir_emit(ctx, rde->name, rde->name_len,
519 			      ceph_translate_ino(inode->i_sb, ino), ftype)) {
520 			dout("filldir stopping us...\n");
521 			return 0;
522 		}
523 		ctx->pos++;
524 	}
525 
526 	ceph_mdsc_put_request(dfi->last_readdir);
527 	dfi->last_readdir = NULL;
528 
529 	if (dfi->next_offset > 2) {
530 		frag = dfi->frag;
531 		goto more;
532 	}
533 
534 	/* more frags? */
535 	if (!ceph_frag_is_rightmost(dfi->frag)) {
536 		frag = ceph_frag_next(dfi->frag);
537 		if (is_hash_order(ctx->pos)) {
538 			loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
539 							dfi->next_offset, true);
540 			if (new_pos > ctx->pos)
541 				ctx->pos = new_pos;
542 			/* keep last_name */
543 		} else {
544 			ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
545 							false);
546 			kfree(dfi->last_name);
547 			dfi->last_name = NULL;
548 		}
549 		dout("readdir next frag is %x\n", frag);
550 		goto more;
551 	}
552 	dfi->file_info.flags |= CEPH_F_ATEND;
553 
554 	/*
555 	 * if dir_release_count still matches the dir, no dentries
556 	 * were released during the whole readdir, and we should have
557 	 * the complete dir contents in our cache.
558 	 */
559 	if (atomic64_read(&ci->i_release_count) ==
560 					dfi->dir_release_count) {
561 		spin_lock(&ci->i_ceph_lock);
562 		if (dfi->dir_ordered_count ==
563 				atomic64_read(&ci->i_ordered_count)) {
564 			dout(" marking %p complete and ordered\n", inode);
565 			/* use i_size to track number of entries in
566 			 * readdir cache */
567 			BUG_ON(dfi->readdir_cache_idx < 0);
568 			i_size_write(inode, dfi->readdir_cache_idx *
569 				     sizeof(struct dentry*));
570 		} else {
571 			dout(" marking %p complete\n", inode);
572 		}
573 		__ceph_dir_set_complete(ci, dfi->dir_release_count,
574 					dfi->dir_ordered_count);
575 		spin_unlock(&ci->i_ceph_lock);
576 	}
577 
578 	dout("readdir %p file %p done.\n", inode, file);
579 	return 0;
580 }
581 
582 static void reset_readdir(struct ceph_dir_file_info *dfi)
583 {
584 	if (dfi->last_readdir) {
585 		ceph_mdsc_put_request(dfi->last_readdir);
586 		dfi->last_readdir = NULL;
587 	}
588 	kfree(dfi->last_name);
589 	dfi->last_name = NULL;
590 	dfi->dir_release_count = 0;
591 	dfi->readdir_cache_idx = -1;
592 	dfi->next_offset = 2;  /* compensate for . and .. */
593 	dfi->file_info.flags &= ~CEPH_F_ATEND;
594 }
595 
596 /*
597  * discard buffered readdir content on seekdir(0), or seek to new frag,
598  * or seek prior to current chunk
599  */
600 static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
601 {
602 	struct ceph_mds_reply_info_parsed *rinfo;
603 	loff_t chunk_offset;
604 	if (new_pos == 0)
605 		return true;
606 	if (is_hash_order(new_pos)) {
607 		/* no need to reset last_name for a forward seek when
608 		 * dentries are sotred in hash order */
609 	} else if (dfi->frag != fpos_frag(new_pos)) {
610 		return true;
611 	}
612 	rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
613 	if (!rinfo || !rinfo->dir_nr)
614 		return true;
615 	chunk_offset = rinfo->dir_entries[0].offset;
616 	return new_pos < chunk_offset ||
617 	       is_hash_order(new_pos) != is_hash_order(chunk_offset);
618 }
619 
620 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
621 {
622 	struct ceph_dir_file_info *dfi = file->private_data;
623 	struct inode *inode = file->f_mapping->host;
624 	loff_t retval;
625 
626 	inode_lock(inode);
627 	retval = -EINVAL;
628 	switch (whence) {
629 	case SEEK_CUR:
630 		offset += file->f_pos;
631 	case SEEK_SET:
632 		break;
633 	case SEEK_END:
634 		retval = -EOPNOTSUPP;
635 	default:
636 		goto out;
637 	}
638 
639 	if (offset >= 0) {
640 		if (need_reset_readdir(dfi, offset)) {
641 			dout("dir_llseek dropping %p content\n", file);
642 			reset_readdir(dfi);
643 		} else if (is_hash_order(offset) && offset > file->f_pos) {
644 			/* for hash offset, we don't know if a forward seek
645 			 * is within same frag */
646 			dfi->dir_release_count = 0;
647 			dfi->readdir_cache_idx = -1;
648 		}
649 
650 		if (offset != file->f_pos) {
651 			file->f_pos = offset;
652 			file->f_version = 0;
653 			dfi->file_info.flags &= ~CEPH_F_ATEND;
654 		}
655 		retval = offset;
656 	}
657 out:
658 	inode_unlock(inode);
659 	return retval;
660 }
661 
662 /*
663  * Handle lookups for the hidden .snap directory.
664  */
665 int ceph_handle_snapdir(struct ceph_mds_request *req,
666 			struct dentry *dentry, int err)
667 {
668 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
669 	struct inode *parent = d_inode(dentry->d_parent); /* we hold i_mutex */
670 
671 	/* .snap dir? */
672 	if (err == -ENOENT &&
673 	    ceph_snap(parent) == CEPH_NOSNAP &&
674 	    strcmp(dentry->d_name.name,
675 		   fsc->mount_options->snapdir_name) == 0) {
676 		struct inode *inode = ceph_get_snapdir(parent);
677 		dout("ENOENT on snapdir %p '%pd', linking to snapdir %p\n",
678 		     dentry, dentry, inode);
679 		BUG_ON(!d_unhashed(dentry));
680 		d_add(dentry, inode);
681 		err = 0;
682 	}
683 	return err;
684 }
685 
686 /*
687  * Figure out final result of a lookup/open request.
688  *
689  * Mainly, make sure we return the final req->r_dentry (if it already
690  * existed) in place of the original VFS-provided dentry when they
691  * differ.
692  *
693  * Gracefully handle the case where the MDS replies with -ENOENT and
694  * no trace (which it may do, at its discretion, e.g., if it doesn't
695  * care to issue a lease on the negative dentry).
696  */
697 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
698 				  struct dentry *dentry, int err)
699 {
700 	if (err == -ENOENT) {
701 		/* no trace? */
702 		err = 0;
703 		if (!req->r_reply_info.head->is_dentry) {
704 			dout("ENOENT and no trace, dentry %p inode %p\n",
705 			     dentry, d_inode(dentry));
706 			if (d_really_is_positive(dentry)) {
707 				d_drop(dentry);
708 				err = -ENOENT;
709 			} else {
710 				d_add(dentry, NULL);
711 			}
712 		}
713 	}
714 	if (err)
715 		dentry = ERR_PTR(err);
716 	else if (dentry != req->r_dentry)
717 		dentry = dget(req->r_dentry);   /* we got spliced */
718 	else
719 		dentry = NULL;
720 	return dentry;
721 }
722 
723 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
724 {
725 	return ceph_ino(inode) == CEPH_INO_ROOT &&
726 		strncmp(dentry->d_name.name, ".ceph", 5) == 0;
727 }
728 
729 /*
730  * Look up a single dir entry.  If there is a lookup intent, inform
731  * the MDS so that it gets our 'caps wanted' value in a single op.
732  */
733 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
734 				  unsigned int flags)
735 {
736 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
737 	struct ceph_mds_client *mdsc = fsc->mdsc;
738 	struct ceph_mds_request *req;
739 	int op;
740 	int mask;
741 	int err;
742 
743 	dout("lookup %p dentry %p '%pd'\n",
744 	     dir, dentry, dentry);
745 
746 	if (dentry->d_name.len > NAME_MAX)
747 		return ERR_PTR(-ENAMETOOLONG);
748 
749 	/* can we conclude ENOENT locally? */
750 	if (d_really_is_negative(dentry)) {
751 		struct ceph_inode_info *ci = ceph_inode(dir);
752 		struct ceph_dentry_info *di = ceph_dentry(dentry);
753 
754 		spin_lock(&ci->i_ceph_lock);
755 		dout(" dir %p flags are %d\n", dir, ci->i_ceph_flags);
756 		if (strncmp(dentry->d_name.name,
757 			    fsc->mount_options->snapdir_name,
758 			    dentry->d_name.len) &&
759 		    !is_root_ceph_dentry(dir, dentry) &&
760 		    ceph_test_mount_opt(fsc, DCACHE) &&
761 		    __ceph_dir_is_complete(ci) &&
762 		    (__ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1))) {
763 			spin_unlock(&ci->i_ceph_lock);
764 			dout(" dir %p complete, -ENOENT\n", dir);
765 			d_add(dentry, NULL);
766 			di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
767 			return NULL;
768 		}
769 		spin_unlock(&ci->i_ceph_lock);
770 	}
771 
772 	op = ceph_snap(dir) == CEPH_SNAPDIR ?
773 		CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
774 	req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
775 	if (IS_ERR(req))
776 		return ERR_CAST(req);
777 	req->r_dentry = dget(dentry);
778 	req->r_num_caps = 2;
779 
780 	mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
781 	if (ceph_security_xattr_wanted(dir))
782 		mask |= CEPH_CAP_XATTR_SHARED;
783 	req->r_args.getattr.mask = cpu_to_le32(mask);
784 
785 	req->r_parent = dir;
786 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
787 	err = ceph_mdsc_do_request(mdsc, NULL, req);
788 	err = ceph_handle_snapdir(req, dentry, err);
789 	dentry = ceph_finish_lookup(req, dentry, err);
790 	ceph_mdsc_put_request(req);  /* will dput(dentry) */
791 	dout("lookup result=%p\n", dentry);
792 	return dentry;
793 }
794 
795 /*
796  * If we do a create but get no trace back from the MDS, follow up with
797  * a lookup (the VFS expects us to link up the provided dentry).
798  */
799 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
800 {
801 	struct dentry *result = ceph_lookup(dir, dentry, 0);
802 
803 	if (result && !IS_ERR(result)) {
804 		/*
805 		 * We created the item, then did a lookup, and found
806 		 * it was already linked to another inode we already
807 		 * had in our cache (and thus got spliced). To not
808 		 * confuse VFS (especially when inode is a directory),
809 		 * we don't link our dentry to that inode, return an
810 		 * error instead.
811 		 *
812 		 * This event should be rare and it happens only when
813 		 * we talk to old MDS. Recent MDS does not send traceless
814 		 * reply for request that creates new inode.
815 		 */
816 		d_drop(result);
817 		return -ESTALE;
818 	}
819 	return PTR_ERR(result);
820 }
821 
822 static int ceph_mknod(struct inode *dir, struct dentry *dentry,
823 		      umode_t mode, dev_t rdev)
824 {
825 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
826 	struct ceph_mds_client *mdsc = fsc->mdsc;
827 	struct ceph_mds_request *req;
828 	struct ceph_acl_sec_ctx as_ctx = {};
829 	int err;
830 
831 	if (ceph_snap(dir) != CEPH_NOSNAP)
832 		return -EROFS;
833 
834 	if (ceph_quota_is_max_files_exceeded(dir)) {
835 		err = -EDQUOT;
836 		goto out;
837 	}
838 
839 	err = ceph_pre_init_acls(dir, &mode, &as_ctx);
840 	if (err < 0)
841 		goto out;
842 	err = ceph_security_init_secctx(dentry, mode, &as_ctx);
843 	if (err < 0)
844 		goto out;
845 
846 	dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
847 	     dir, dentry, mode, rdev);
848 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
849 	if (IS_ERR(req)) {
850 		err = PTR_ERR(req);
851 		goto out;
852 	}
853 	req->r_dentry = dget(dentry);
854 	req->r_num_caps = 2;
855 	req->r_parent = dir;
856 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
857 	req->r_args.mknod.mode = cpu_to_le32(mode);
858 	req->r_args.mknod.rdev = cpu_to_le32(rdev);
859 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
860 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
861 	if (as_ctx.pagelist) {
862 		req->r_pagelist = as_ctx.pagelist;
863 		as_ctx.pagelist = NULL;
864 	}
865 	err = ceph_mdsc_do_request(mdsc, dir, req);
866 	if (!err && !req->r_reply_info.head->is_dentry)
867 		err = ceph_handle_notrace_create(dir, dentry);
868 	ceph_mdsc_put_request(req);
869 out:
870 	if (!err)
871 		ceph_init_inode_acls(d_inode(dentry), &as_ctx);
872 	else
873 		d_drop(dentry);
874 	ceph_release_acl_sec_ctx(&as_ctx);
875 	return err;
876 }
877 
878 static int ceph_create(struct inode *dir, struct dentry *dentry, umode_t mode,
879 		       bool excl)
880 {
881 	return ceph_mknod(dir, dentry, mode, 0);
882 }
883 
884 static int ceph_symlink(struct inode *dir, struct dentry *dentry,
885 			    const char *dest)
886 {
887 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
888 	struct ceph_mds_client *mdsc = fsc->mdsc;
889 	struct ceph_mds_request *req;
890 	struct ceph_acl_sec_ctx as_ctx = {};
891 	int err;
892 
893 	if (ceph_snap(dir) != CEPH_NOSNAP)
894 		return -EROFS;
895 
896 	if (ceph_quota_is_max_files_exceeded(dir)) {
897 		err = -EDQUOT;
898 		goto out;
899 	}
900 
901 	err = ceph_security_init_secctx(dentry, S_IFLNK | 0777, &as_ctx);
902 	if (err < 0)
903 		goto out;
904 
905 	dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
906 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
907 	if (IS_ERR(req)) {
908 		err = PTR_ERR(req);
909 		goto out;
910 	}
911 	req->r_path2 = kstrdup(dest, GFP_KERNEL);
912 	if (!req->r_path2) {
913 		err = -ENOMEM;
914 		ceph_mdsc_put_request(req);
915 		goto out;
916 	}
917 	req->r_parent = dir;
918 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
919 	req->r_dentry = dget(dentry);
920 	req->r_num_caps = 2;
921 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
922 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
923 	err = ceph_mdsc_do_request(mdsc, dir, req);
924 	if (!err && !req->r_reply_info.head->is_dentry)
925 		err = ceph_handle_notrace_create(dir, dentry);
926 	ceph_mdsc_put_request(req);
927 out:
928 	if (err)
929 		d_drop(dentry);
930 	ceph_release_acl_sec_ctx(&as_ctx);
931 	return err;
932 }
933 
934 static int ceph_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
935 {
936 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
937 	struct ceph_mds_client *mdsc = fsc->mdsc;
938 	struct ceph_mds_request *req;
939 	struct ceph_acl_sec_ctx as_ctx = {};
940 	int err = -EROFS;
941 	int op;
942 
943 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
944 		/* mkdir .snap/foo is a MKSNAP */
945 		op = CEPH_MDS_OP_MKSNAP;
946 		dout("mksnap dir %p snap '%pd' dn %p\n", dir,
947 		     dentry, dentry);
948 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
949 		dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
950 		op = CEPH_MDS_OP_MKDIR;
951 	} else {
952 		goto out;
953 	}
954 
955 	if (op == CEPH_MDS_OP_MKDIR &&
956 	    ceph_quota_is_max_files_exceeded(dir)) {
957 		err = -EDQUOT;
958 		goto out;
959 	}
960 
961 	mode |= S_IFDIR;
962 	err = ceph_pre_init_acls(dir, &mode, &as_ctx);
963 	if (err < 0)
964 		goto out;
965 	err = ceph_security_init_secctx(dentry, mode, &as_ctx);
966 	if (err < 0)
967 		goto out;
968 
969 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
970 	if (IS_ERR(req)) {
971 		err = PTR_ERR(req);
972 		goto out;
973 	}
974 
975 	req->r_dentry = dget(dentry);
976 	req->r_num_caps = 2;
977 	req->r_parent = dir;
978 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
979 	req->r_args.mkdir.mode = cpu_to_le32(mode);
980 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
981 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
982 	if (as_ctx.pagelist) {
983 		req->r_pagelist = as_ctx.pagelist;
984 		as_ctx.pagelist = NULL;
985 	}
986 	err = ceph_mdsc_do_request(mdsc, dir, req);
987 	if (!err &&
988 	    !req->r_reply_info.head->is_target &&
989 	    !req->r_reply_info.head->is_dentry)
990 		err = ceph_handle_notrace_create(dir, dentry);
991 	ceph_mdsc_put_request(req);
992 out:
993 	if (!err)
994 		ceph_init_inode_acls(d_inode(dentry), &as_ctx);
995 	else
996 		d_drop(dentry);
997 	ceph_release_acl_sec_ctx(&as_ctx);
998 	return err;
999 }
1000 
1001 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
1002 		     struct dentry *dentry)
1003 {
1004 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1005 	struct ceph_mds_client *mdsc = fsc->mdsc;
1006 	struct ceph_mds_request *req;
1007 	int err;
1008 
1009 	if (ceph_snap(dir) != CEPH_NOSNAP)
1010 		return -EROFS;
1011 
1012 	dout("link in dir %p old_dentry %p dentry %p\n", dir,
1013 	     old_dentry, dentry);
1014 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
1015 	if (IS_ERR(req)) {
1016 		d_drop(dentry);
1017 		return PTR_ERR(req);
1018 	}
1019 	req->r_dentry = dget(dentry);
1020 	req->r_num_caps = 2;
1021 	req->r_old_dentry = dget(old_dentry);
1022 	req->r_parent = dir;
1023 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1024 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1025 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1026 	/* release LINK_SHARED on source inode (mds will lock it) */
1027 	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1028 	err = ceph_mdsc_do_request(mdsc, dir, req);
1029 	if (err) {
1030 		d_drop(dentry);
1031 	} else if (!req->r_reply_info.head->is_dentry) {
1032 		ihold(d_inode(old_dentry));
1033 		d_instantiate(dentry, d_inode(old_dentry));
1034 	}
1035 	ceph_mdsc_put_request(req);
1036 	return err;
1037 }
1038 
1039 /*
1040  * rmdir and unlink are differ only by the metadata op code
1041  */
1042 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1043 {
1044 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1045 	struct ceph_mds_client *mdsc = fsc->mdsc;
1046 	struct inode *inode = d_inode(dentry);
1047 	struct ceph_mds_request *req;
1048 	int err = -EROFS;
1049 	int op;
1050 
1051 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
1052 		/* rmdir .snap/foo is RMSNAP */
1053 		dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1054 		op = CEPH_MDS_OP_RMSNAP;
1055 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
1056 		dout("unlink/rmdir dir %p dn %p inode %p\n",
1057 		     dir, dentry, inode);
1058 		op = d_is_dir(dentry) ?
1059 			CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1060 	} else
1061 		goto out;
1062 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1063 	if (IS_ERR(req)) {
1064 		err = PTR_ERR(req);
1065 		goto out;
1066 	}
1067 	req->r_dentry = dget(dentry);
1068 	req->r_num_caps = 2;
1069 	req->r_parent = dir;
1070 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1071 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1072 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1073 	req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1074 	err = ceph_mdsc_do_request(mdsc, dir, req);
1075 	if (!err && !req->r_reply_info.head->is_dentry)
1076 		d_delete(dentry);
1077 	ceph_mdsc_put_request(req);
1078 out:
1079 	return err;
1080 }
1081 
1082 static int ceph_rename(struct inode *old_dir, struct dentry *old_dentry,
1083 		       struct inode *new_dir, struct dentry *new_dentry,
1084 		       unsigned int flags)
1085 {
1086 	struct ceph_fs_client *fsc = ceph_sb_to_client(old_dir->i_sb);
1087 	struct ceph_mds_client *mdsc = fsc->mdsc;
1088 	struct ceph_mds_request *req;
1089 	int op = CEPH_MDS_OP_RENAME;
1090 	int err;
1091 
1092 	if (flags)
1093 		return -EINVAL;
1094 
1095 	if (ceph_snap(old_dir) != ceph_snap(new_dir))
1096 		return -EXDEV;
1097 	if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1098 		if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1099 			op = CEPH_MDS_OP_RENAMESNAP;
1100 		else
1101 			return -EROFS;
1102 	}
1103 	/* don't allow cross-quota renames */
1104 	if ((old_dir != new_dir) &&
1105 	    (!ceph_quota_is_same_realm(old_dir, new_dir)))
1106 		return -EXDEV;
1107 
1108 	dout("rename dir %p dentry %p to dir %p dentry %p\n",
1109 	     old_dir, old_dentry, new_dir, new_dentry);
1110 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1111 	if (IS_ERR(req))
1112 		return PTR_ERR(req);
1113 	ihold(old_dir);
1114 	req->r_dentry = dget(new_dentry);
1115 	req->r_num_caps = 2;
1116 	req->r_old_dentry = dget(old_dentry);
1117 	req->r_old_dentry_dir = old_dir;
1118 	req->r_parent = new_dir;
1119 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1120 	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1121 	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1122 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1123 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1124 	/* release LINK_RDCACHE on source inode (mds will lock it) */
1125 	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1126 	if (d_really_is_positive(new_dentry)) {
1127 		req->r_inode_drop =
1128 			ceph_drop_caps_for_unlink(d_inode(new_dentry));
1129 	}
1130 	err = ceph_mdsc_do_request(mdsc, old_dir, req);
1131 	if (!err && !req->r_reply_info.head->is_dentry) {
1132 		/*
1133 		 * Normally d_move() is done by fill_trace (called by
1134 		 * do_request, above).  If there is no trace, we need
1135 		 * to do it here.
1136 		 */
1137 		d_move(old_dentry, new_dentry);
1138 	}
1139 	ceph_mdsc_put_request(req);
1140 	return err;
1141 }
1142 
1143 /*
1144  * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
1145  * Leases at front of the list will expire first. (Assume all leases have
1146  * similar duration)
1147  *
1148  * Called under dentry->d_lock.
1149  */
1150 void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
1151 {
1152 	struct dentry *dn = di->dentry;
1153 	struct ceph_mds_client *mdsc;
1154 
1155 	dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn);
1156 
1157 	di->flags |= CEPH_DENTRY_LEASE_LIST;
1158 	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1159 		di->flags |= CEPH_DENTRY_REFERENCED;
1160 		return;
1161 	}
1162 
1163 	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1164 	spin_lock(&mdsc->dentry_list_lock);
1165 	list_move_tail(&di->lease_list, &mdsc->dentry_leases);
1166 	spin_unlock(&mdsc->dentry_list_lock);
1167 }
1168 
1169 static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
1170 				     struct ceph_dentry_info *di)
1171 {
1172 	di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
1173 	di->lease_gen = 0;
1174 	di->time = jiffies;
1175 	list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
1176 }
1177 
1178 /*
1179  * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
1180  * list if it's not in the list, otherwise set 'referenced' flag.
1181  *
1182  * Called under dentry->d_lock.
1183  */
1184 void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
1185 {
1186 	struct dentry *dn = di->dentry;
1187 	struct ceph_mds_client *mdsc;
1188 
1189 	dout("dentry_dir_lease_touch %p %p '%pd' (offset %lld)\n",
1190 	     di, dn, dn, di->offset);
1191 
1192 	if (!list_empty(&di->lease_list)) {
1193 		if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1194 			/* don't remove dentry from dentry lease list
1195 			 * if its lease is valid */
1196 			if (__dentry_lease_is_valid(di))
1197 				return;
1198 		} else {
1199 			di->flags |= CEPH_DENTRY_REFERENCED;
1200 			return;
1201 		}
1202 	}
1203 
1204 	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1205 		di->flags |= CEPH_DENTRY_REFERENCED;
1206 		di->flags &= ~CEPH_DENTRY_LEASE_LIST;
1207 		return;
1208 	}
1209 
1210 	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1211 	spin_lock(&mdsc->dentry_list_lock);
1212 	__dentry_dir_lease_touch(mdsc, di),
1213 	spin_unlock(&mdsc->dentry_list_lock);
1214 }
1215 
1216 static void __dentry_lease_unlist(struct ceph_dentry_info *di)
1217 {
1218 	struct ceph_mds_client *mdsc;
1219 	if (di->flags & CEPH_DENTRY_SHRINK_LIST)
1220 		return;
1221 	if (list_empty(&di->lease_list))
1222 		return;
1223 
1224 	mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc;
1225 	spin_lock(&mdsc->dentry_list_lock);
1226 	list_del_init(&di->lease_list);
1227 	spin_unlock(&mdsc->dentry_list_lock);
1228 }
1229 
1230 enum {
1231 	KEEP	= 0,
1232 	DELETE	= 1,
1233 	TOUCH	= 2,
1234 	STOP	= 4,
1235 };
1236 
1237 struct ceph_lease_walk_control {
1238 	bool dir_lease;
1239 	bool expire_dir_lease;
1240 	unsigned long nr_to_scan;
1241 	unsigned long dir_lease_ttl;
1242 };
1243 
1244 static unsigned long
1245 __dentry_leases_walk(struct ceph_mds_client *mdsc,
1246 		     struct ceph_lease_walk_control *lwc,
1247 		     int (*check)(struct dentry*, void*))
1248 {
1249 	struct ceph_dentry_info *di, *tmp;
1250 	struct dentry *dentry, *last = NULL;
1251 	struct list_head* list;
1252         LIST_HEAD(dispose);
1253 	unsigned long freed = 0;
1254 	int ret = 0;
1255 
1256 	list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
1257 	spin_lock(&mdsc->dentry_list_lock);
1258 	list_for_each_entry_safe(di, tmp, list, lease_list) {
1259 		if (!lwc->nr_to_scan)
1260 			break;
1261 		--lwc->nr_to_scan;
1262 
1263 		dentry = di->dentry;
1264 		if (last == dentry)
1265 			break;
1266 
1267 		if (!spin_trylock(&dentry->d_lock))
1268 			continue;
1269 
1270 		if (__lockref_is_dead(&dentry->d_lockref)) {
1271 			list_del_init(&di->lease_list);
1272 			goto next;
1273 		}
1274 
1275 		ret = check(dentry, lwc);
1276 		if (ret & TOUCH) {
1277 			/* move it into tail of dir lease list */
1278 			__dentry_dir_lease_touch(mdsc, di);
1279 			if (!last)
1280 				last = dentry;
1281 		}
1282 		if (ret & DELETE) {
1283 			/* stale lease */
1284 			di->flags &= ~CEPH_DENTRY_REFERENCED;
1285 			if (dentry->d_lockref.count > 0) {
1286 				/* update_dentry_lease() will re-add
1287 				 * it to lease list, or
1288 				 * ceph_d_delete() will return 1 when
1289 				 * last reference is dropped */
1290 				list_del_init(&di->lease_list);
1291 			} else {
1292 				di->flags |= CEPH_DENTRY_SHRINK_LIST;
1293 				list_move_tail(&di->lease_list, &dispose);
1294 				dget_dlock(dentry);
1295 			}
1296 		}
1297 next:
1298 		spin_unlock(&dentry->d_lock);
1299 		if (ret & STOP)
1300 			break;
1301 	}
1302 	spin_unlock(&mdsc->dentry_list_lock);
1303 
1304 	while (!list_empty(&dispose)) {
1305 		di = list_first_entry(&dispose, struct ceph_dentry_info,
1306 				      lease_list);
1307 		dentry = di->dentry;
1308 		spin_lock(&dentry->d_lock);
1309 
1310 		list_del_init(&di->lease_list);
1311 		di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
1312 		if (di->flags & CEPH_DENTRY_REFERENCED) {
1313 			spin_lock(&mdsc->dentry_list_lock);
1314 			if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1315 				list_add_tail(&di->lease_list,
1316 					      &mdsc->dentry_leases);
1317 			} else {
1318 				__dentry_dir_lease_touch(mdsc, di);
1319 			}
1320 			spin_unlock(&mdsc->dentry_list_lock);
1321 		} else {
1322 			freed++;
1323 		}
1324 
1325 		spin_unlock(&dentry->d_lock);
1326 		/* ceph_d_delete() does the trick */
1327 		dput(dentry);
1328 	}
1329 	return freed;
1330 }
1331 
1332 static int __dentry_lease_check(struct dentry *dentry, void *arg)
1333 {
1334 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1335 	int ret;
1336 
1337 	if (__dentry_lease_is_valid(di))
1338 		return STOP;
1339 	ret = __dir_lease_try_check(dentry);
1340 	if (ret == -EBUSY)
1341 		return KEEP;
1342 	if (ret > 0)
1343 		return TOUCH;
1344 	return DELETE;
1345 }
1346 
1347 static int __dir_lease_check(struct dentry *dentry, void *arg)
1348 {
1349 	struct ceph_lease_walk_control *lwc = arg;
1350 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1351 
1352 	int ret = __dir_lease_try_check(dentry);
1353 	if (ret == -EBUSY)
1354 		return KEEP;
1355 	if (ret > 0) {
1356 		if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
1357 			return STOP;
1358 		/* Move dentry to tail of dir lease list if we don't want
1359 		 * to delete it. So dentries in the list are checked in a
1360 		 * round robin manner */
1361 		if (!lwc->expire_dir_lease)
1362 			return TOUCH;
1363 		if (dentry->d_lockref.count > 0 ||
1364 		    (di->flags & CEPH_DENTRY_REFERENCED))
1365 			return TOUCH;
1366 		/* invalidate dir lease */
1367 		di->lease_shared_gen = 0;
1368 	}
1369 	return DELETE;
1370 }
1371 
1372 int ceph_trim_dentries(struct ceph_mds_client *mdsc)
1373 {
1374 	struct ceph_lease_walk_control lwc;
1375 	unsigned long count;
1376 	unsigned long freed;
1377 
1378 	spin_lock(&mdsc->caps_list_lock);
1379         if (mdsc->caps_use_max > 0 &&
1380             mdsc->caps_use_count > mdsc->caps_use_max)
1381 		count = mdsc->caps_use_count - mdsc->caps_use_max;
1382 	else
1383 		count = 0;
1384         spin_unlock(&mdsc->caps_list_lock);
1385 
1386 	lwc.dir_lease = false;
1387 	lwc.nr_to_scan  = CEPH_CAPS_PER_RELEASE * 2;
1388 	freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check);
1389 	if (!lwc.nr_to_scan) /* more invalid leases */
1390 		return -EAGAIN;
1391 
1392 	if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
1393 		lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
1394 
1395 	lwc.dir_lease = true;
1396 	lwc.expire_dir_lease = freed < count;
1397 	lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ;
1398 	freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check);
1399 	if (!lwc.nr_to_scan) /* more to check */
1400 		return -EAGAIN;
1401 
1402 	return freed > 0 ? 1 : 0;
1403 }
1404 
1405 /*
1406  * Ensure a dentry lease will no longer revalidate.
1407  */
1408 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1409 {
1410 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1411 	spin_lock(&dentry->d_lock);
1412 	di->time = jiffies;
1413 	di->lease_shared_gen = 0;
1414 	__dentry_lease_unlist(di);
1415 	spin_unlock(&dentry->d_lock);
1416 }
1417 
1418 /*
1419  * Check if dentry lease is valid.  If not, delete the lease.  Try to
1420  * renew if the least is more than half up.
1421  */
1422 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di)
1423 {
1424 	struct ceph_mds_session *session;
1425 
1426 	if (!di->lease_gen)
1427 		return false;
1428 
1429 	session = di->lease_session;
1430 	if (session) {
1431 		u32 gen;
1432 		unsigned long ttl;
1433 
1434 		spin_lock(&session->s_gen_ttl_lock);
1435 		gen = session->s_cap_gen;
1436 		ttl = session->s_cap_ttl;
1437 		spin_unlock(&session->s_gen_ttl_lock);
1438 
1439 		if (di->lease_gen == gen &&
1440 		    time_before(jiffies, ttl) &&
1441 		    time_before(jiffies, di->time))
1442 			return true;
1443 	}
1444 	di->lease_gen = 0;
1445 	return false;
1446 }
1447 
1448 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags)
1449 {
1450 	struct ceph_dentry_info *di;
1451 	struct ceph_mds_session *session = NULL;
1452 	u32 seq = 0;
1453 	int valid = 0;
1454 
1455 	spin_lock(&dentry->d_lock);
1456 	di = ceph_dentry(dentry);
1457 	if (di && __dentry_lease_is_valid(di)) {
1458 		valid = 1;
1459 
1460 		if (di->lease_renew_after &&
1461 		    time_after(jiffies, di->lease_renew_after)) {
1462 			/*
1463 			 * We should renew. If we're in RCU walk mode
1464 			 * though, we can't do that so just return
1465 			 * -ECHILD.
1466 			 */
1467 			if (flags & LOOKUP_RCU) {
1468 				valid = -ECHILD;
1469 			} else {
1470 				session = ceph_get_mds_session(di->lease_session);
1471 				seq = di->lease_seq;
1472 				di->lease_renew_after = 0;
1473 				di->lease_renew_from = jiffies;
1474 			}
1475 		}
1476 	}
1477 	spin_unlock(&dentry->d_lock);
1478 
1479 	if (session) {
1480 		ceph_mdsc_lease_send_msg(session, dentry,
1481 					 CEPH_MDS_LEASE_RENEW, seq);
1482 		ceph_put_mds_session(session);
1483 	}
1484 	dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1485 	return valid;
1486 }
1487 
1488 /*
1489  * Called under dentry->d_lock.
1490  */
1491 static int __dir_lease_try_check(const struct dentry *dentry)
1492 {
1493 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1494 	struct inode *dir;
1495 	struct ceph_inode_info *ci;
1496 	int valid = 0;
1497 
1498 	if (!di->lease_shared_gen)
1499 		return 0;
1500 	if (IS_ROOT(dentry))
1501 		return 0;
1502 
1503 	dir = d_inode(dentry->d_parent);
1504 	ci = ceph_inode(dir);
1505 
1506 	if (spin_trylock(&ci->i_ceph_lock)) {
1507 		if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen &&
1508 		    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0))
1509 			valid = 1;
1510 		spin_unlock(&ci->i_ceph_lock);
1511 	} else {
1512 		valid = -EBUSY;
1513 	}
1514 
1515 	if (!valid)
1516 		di->lease_shared_gen = 0;
1517 	return valid;
1518 }
1519 
1520 /*
1521  * Check if directory-wide content lease/cap is valid.
1522  */
1523 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry)
1524 {
1525 	struct ceph_inode_info *ci = ceph_inode(dir);
1526 	int valid;
1527 	int shared_gen;
1528 
1529 	spin_lock(&ci->i_ceph_lock);
1530 	valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1531 	shared_gen = atomic_read(&ci->i_shared_gen);
1532 	spin_unlock(&ci->i_ceph_lock);
1533 	if (valid) {
1534 		struct ceph_dentry_info *di;
1535 		spin_lock(&dentry->d_lock);
1536 		di = ceph_dentry(dentry);
1537 		if (dir == d_inode(dentry->d_parent) &&
1538 		    di && di->lease_shared_gen == shared_gen)
1539 			__ceph_dentry_dir_lease_touch(di);
1540 		else
1541 			valid = 0;
1542 		spin_unlock(&dentry->d_lock);
1543 	}
1544 	dout("dir_lease_is_valid dir %p v%u dentry %p = %d\n",
1545 	     dir, (unsigned)atomic_read(&ci->i_shared_gen), dentry, valid);
1546 	return valid;
1547 }
1548 
1549 /*
1550  * Check if cached dentry can be trusted.
1551  */
1552 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1553 {
1554 	int valid = 0;
1555 	struct dentry *parent;
1556 	struct inode *dir;
1557 
1558 	if (flags & LOOKUP_RCU) {
1559 		parent = READ_ONCE(dentry->d_parent);
1560 		dir = d_inode_rcu(parent);
1561 		if (!dir)
1562 			return -ECHILD;
1563 	} else {
1564 		parent = dget_parent(dentry);
1565 		dir = d_inode(parent);
1566 	}
1567 
1568 	dout("d_revalidate %p '%pd' inode %p offset %lld\n", dentry,
1569 	     dentry, d_inode(dentry), ceph_dentry(dentry)->offset);
1570 
1571 	/* always trust cached snapped dentries, snapdir dentry */
1572 	if (ceph_snap(dir) != CEPH_NOSNAP) {
1573 		dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1574 		     dentry, d_inode(dentry));
1575 		valid = 1;
1576 	} else if (d_really_is_positive(dentry) &&
1577 		   ceph_snap(d_inode(dentry)) == CEPH_SNAPDIR) {
1578 		valid = 1;
1579 	} else {
1580 		valid = dentry_lease_is_valid(dentry, flags);
1581 		if (valid == -ECHILD)
1582 			return valid;
1583 		if (valid || dir_lease_is_valid(dir, dentry)) {
1584 			if (d_really_is_positive(dentry))
1585 				valid = ceph_is_any_caps(d_inode(dentry));
1586 			else
1587 				valid = 1;
1588 		}
1589 	}
1590 
1591 	if (!valid) {
1592 		struct ceph_mds_client *mdsc =
1593 			ceph_sb_to_client(dir->i_sb)->mdsc;
1594 		struct ceph_mds_request *req;
1595 		int op, err;
1596 		u32 mask;
1597 
1598 		if (flags & LOOKUP_RCU)
1599 			return -ECHILD;
1600 
1601 		op = ceph_snap(dir) == CEPH_SNAPDIR ?
1602 			CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1603 		req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1604 		if (!IS_ERR(req)) {
1605 			req->r_dentry = dget(dentry);
1606 			req->r_num_caps = 2;
1607 			req->r_parent = dir;
1608 
1609 			mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1610 			if (ceph_security_xattr_wanted(dir))
1611 				mask |= CEPH_CAP_XATTR_SHARED;
1612 			req->r_args.getattr.mask = cpu_to_le32(mask);
1613 
1614 			err = ceph_mdsc_do_request(mdsc, NULL, req);
1615 			switch (err) {
1616 			case 0:
1617 				if (d_really_is_positive(dentry) &&
1618 				    d_inode(dentry) == req->r_target_inode)
1619 					valid = 1;
1620 				break;
1621 			case -ENOENT:
1622 				if (d_really_is_negative(dentry))
1623 					valid = 1;
1624 				/* Fallthrough */
1625 			default:
1626 				break;
1627 			}
1628 			ceph_mdsc_put_request(req);
1629 			dout("d_revalidate %p lookup result=%d\n",
1630 			     dentry, err);
1631 		}
1632 	}
1633 
1634 	dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1635 	if (!valid)
1636 		ceph_dir_clear_complete(dir);
1637 
1638 	if (!(flags & LOOKUP_RCU))
1639 		dput(parent);
1640 	return valid;
1641 }
1642 
1643 /*
1644  * Delete unused dentry that doesn't have valid lease
1645  *
1646  * Called under dentry->d_lock.
1647  */
1648 static int ceph_d_delete(const struct dentry *dentry)
1649 {
1650 	struct ceph_dentry_info *di;
1651 
1652 	/* won't release caps */
1653 	if (d_really_is_negative(dentry))
1654 		return 0;
1655 	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1656 		return 0;
1657 	/* vaild lease? */
1658 	di = ceph_dentry(dentry);
1659 	if (di) {
1660 		if (__dentry_lease_is_valid(di))
1661 			return 0;
1662 		if (__dir_lease_try_check(dentry))
1663 			return 0;
1664 	}
1665 	return 1;
1666 }
1667 
1668 /*
1669  * Release our ceph_dentry_info.
1670  */
1671 static void ceph_d_release(struct dentry *dentry)
1672 {
1673 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1674 
1675 	dout("d_release %p\n", dentry);
1676 
1677 	spin_lock(&dentry->d_lock);
1678 	__dentry_lease_unlist(di);
1679 	dentry->d_fsdata = NULL;
1680 	spin_unlock(&dentry->d_lock);
1681 
1682 	if (di->lease_session)
1683 		ceph_put_mds_session(di->lease_session);
1684 	kmem_cache_free(ceph_dentry_cachep, di);
1685 }
1686 
1687 /*
1688  * When the VFS prunes a dentry from the cache, we need to clear the
1689  * complete flag on the parent directory.
1690  *
1691  * Called under dentry->d_lock.
1692  */
1693 static void ceph_d_prune(struct dentry *dentry)
1694 {
1695 	struct ceph_inode_info *dir_ci;
1696 	struct ceph_dentry_info *di;
1697 
1698 	dout("ceph_d_prune %pd %p\n", dentry, dentry);
1699 
1700 	/* do we have a valid parent? */
1701 	if (IS_ROOT(dentry))
1702 		return;
1703 
1704 	/* we hold d_lock, so d_parent is stable */
1705 	dir_ci = ceph_inode(d_inode(dentry->d_parent));
1706 	if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
1707 		return;
1708 
1709 	/* who calls d_delete() should also disable dcache readdir */
1710 	if (d_really_is_negative(dentry))
1711 		return;
1712 
1713 	/* d_fsdata does not get cleared until d_release */
1714 	if (!d_unhashed(dentry)) {
1715 		__ceph_dir_clear_complete(dir_ci);
1716 		return;
1717 	}
1718 
1719 	/* Disable dcache readdir just in case that someone called d_drop()
1720 	 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1721 	 * properly (dcache readdir is still enabled) */
1722 	di = ceph_dentry(dentry);
1723 	if (di->offset > 0 &&
1724 	    di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
1725 		__ceph_dir_clear_ordered(dir_ci);
1726 }
1727 
1728 /*
1729  * read() on a dir.  This weird interface hack only works if mounted
1730  * with '-o dirstat'.
1731  */
1732 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1733 			     loff_t *ppos)
1734 {
1735 	struct ceph_dir_file_info *dfi = file->private_data;
1736 	struct inode *inode = file_inode(file);
1737 	struct ceph_inode_info *ci = ceph_inode(inode);
1738 	int left;
1739 	const int bufsize = 1024;
1740 
1741 	if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1742 		return -EISDIR;
1743 
1744 	if (!dfi->dir_info) {
1745 		dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
1746 		if (!dfi->dir_info)
1747 			return -ENOMEM;
1748 		dfi->dir_info_len =
1749 			snprintf(dfi->dir_info, bufsize,
1750 				"entries:   %20lld\n"
1751 				" files:    %20lld\n"
1752 				" subdirs:  %20lld\n"
1753 				"rentries:  %20lld\n"
1754 				" rfiles:   %20lld\n"
1755 				" rsubdirs: %20lld\n"
1756 				"rbytes:    %20lld\n"
1757 				"rctime:    %10lld.%09ld\n",
1758 				ci->i_files + ci->i_subdirs,
1759 				ci->i_files,
1760 				ci->i_subdirs,
1761 				ci->i_rfiles + ci->i_rsubdirs,
1762 				ci->i_rfiles,
1763 				ci->i_rsubdirs,
1764 				ci->i_rbytes,
1765 				ci->i_rctime.tv_sec,
1766 				ci->i_rctime.tv_nsec);
1767 	}
1768 
1769 	if (*ppos >= dfi->dir_info_len)
1770 		return 0;
1771 	size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
1772 	left = copy_to_user(buf, dfi->dir_info + *ppos, size);
1773 	if (left == size)
1774 		return -EFAULT;
1775 	*ppos += (size - left);
1776 	return size - left;
1777 }
1778 
1779 
1780 
1781 /*
1782  * Return name hash for a given dentry.  This is dependent on
1783  * the parent directory's hash function.
1784  */
1785 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1786 {
1787 	struct ceph_inode_info *dci = ceph_inode(dir);
1788 	unsigned hash;
1789 
1790 	switch (dci->i_dir_layout.dl_dir_hash) {
1791 	case 0:	/* for backward compat */
1792 	case CEPH_STR_HASH_LINUX:
1793 		return dn->d_name.hash;
1794 
1795 	default:
1796 		spin_lock(&dn->d_lock);
1797 		hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
1798 				     dn->d_name.name, dn->d_name.len);
1799 		spin_unlock(&dn->d_lock);
1800 		return hash;
1801 	}
1802 }
1803 
1804 const struct file_operations ceph_dir_fops = {
1805 	.read = ceph_read_dir,
1806 	.iterate = ceph_readdir,
1807 	.llseek = ceph_dir_llseek,
1808 	.open = ceph_open,
1809 	.release = ceph_release,
1810 	.unlocked_ioctl = ceph_ioctl,
1811 	.fsync = ceph_fsync,
1812 	.lock = ceph_lock,
1813 	.flock = ceph_flock,
1814 };
1815 
1816 const struct file_operations ceph_snapdir_fops = {
1817 	.iterate = ceph_readdir,
1818 	.llseek = ceph_dir_llseek,
1819 	.open = ceph_open,
1820 	.release = ceph_release,
1821 };
1822 
1823 const struct inode_operations ceph_dir_iops = {
1824 	.lookup = ceph_lookup,
1825 	.permission = ceph_permission,
1826 	.getattr = ceph_getattr,
1827 	.setattr = ceph_setattr,
1828 	.listxattr = ceph_listxattr,
1829 	.get_acl = ceph_get_acl,
1830 	.set_acl = ceph_set_acl,
1831 	.mknod = ceph_mknod,
1832 	.symlink = ceph_symlink,
1833 	.mkdir = ceph_mkdir,
1834 	.link = ceph_link,
1835 	.unlink = ceph_unlink,
1836 	.rmdir = ceph_unlink,
1837 	.rename = ceph_rename,
1838 	.create = ceph_create,
1839 	.atomic_open = ceph_atomic_open,
1840 };
1841 
1842 const struct inode_operations ceph_snapdir_iops = {
1843 	.lookup = ceph_lookup,
1844 	.permission = ceph_permission,
1845 	.getattr = ceph_getattr,
1846 	.mkdir = ceph_mkdir,
1847 	.rmdir = ceph_unlink,
1848 	.rename = ceph_rename,
1849 };
1850 
1851 const struct dentry_operations ceph_dentry_ops = {
1852 	.d_revalidate = ceph_d_revalidate,
1853 	.d_delete = ceph_d_delete,
1854 	.d_release = ceph_d_release,
1855 	.d_prune = ceph_d_prune,
1856 	.d_init = ceph_d_init,
1857 };
1858