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  */
ceph_d_init(struct dentry * dentry)38 static int ceph_d_init(struct dentry *dentry)
39 {
40 	struct ceph_dentry_info *di;
41 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dentry->d_sb);
42 
43 	di = kmem_cache_zalloc(ceph_dentry_cachep, GFP_KERNEL);
44 	if (!di)
45 		return -ENOMEM;          /* oh well */
46 
47 	di->dentry = dentry;
48 	di->lease_session = NULL;
49 	di->time = jiffies;
50 	dentry->d_fsdata = di;
51 	INIT_LIST_HEAD(&di->lease_list);
52 
53 	atomic64_inc(&mdsc->metric.total_dentries);
54 
55 	return 0;
56 }
57 
58 /*
59  * for f_pos for readdir:
60  * - hash order:
61  *	(0xff << 52) | ((24 bits hash) << 28) |
62  *	(the nth entry has hash collision);
63  * - frag+name order;
64  *	((frag value) << 28) | (the nth entry in frag);
65  */
66 #define OFFSET_BITS	28
67 #define OFFSET_MASK	((1 << OFFSET_BITS) - 1)
68 #define HASH_ORDER	(0xffull << (OFFSET_BITS + 24))
ceph_make_fpos(unsigned high,unsigned off,bool hash_order)69 loff_t ceph_make_fpos(unsigned high, unsigned off, bool hash_order)
70 {
71 	loff_t fpos = ((loff_t)high << 28) | (loff_t)off;
72 	if (hash_order)
73 		fpos |= HASH_ORDER;
74 	return fpos;
75 }
76 
is_hash_order(loff_t p)77 static bool is_hash_order(loff_t p)
78 {
79 	return (p & HASH_ORDER) == HASH_ORDER;
80 }
81 
fpos_frag(loff_t p)82 static unsigned fpos_frag(loff_t p)
83 {
84 	return p >> OFFSET_BITS;
85 }
86 
fpos_hash(loff_t p)87 static unsigned fpos_hash(loff_t p)
88 {
89 	return ceph_frag_value(fpos_frag(p));
90 }
91 
fpos_off(loff_t p)92 static unsigned fpos_off(loff_t p)
93 {
94 	return p & OFFSET_MASK;
95 }
96 
fpos_cmp(loff_t l,loff_t r)97 static int fpos_cmp(loff_t l, loff_t r)
98 {
99 	int v = ceph_frag_compare(fpos_frag(l), fpos_frag(r));
100 	if (v)
101 		return v;
102 	return (int)(fpos_off(l) - fpos_off(r));
103 }
104 
105 /*
106  * make note of the last dentry we read, so we can
107  * continue at the same lexicographical point,
108  * regardless of what dir changes take place on the
109  * server.
110  */
note_last_dentry(struct ceph_dir_file_info * dfi,const char * name,int len,unsigned next_offset)111 static int note_last_dentry(struct ceph_dir_file_info *dfi, const char *name,
112 		            int len, unsigned next_offset)
113 {
114 	char *buf = kmalloc(len+1, GFP_KERNEL);
115 	if (!buf)
116 		return -ENOMEM;
117 	kfree(dfi->last_name);
118 	dfi->last_name = buf;
119 	memcpy(dfi->last_name, name, len);
120 	dfi->last_name[len] = 0;
121 	dfi->next_offset = next_offset;
122 	dout("note_last_dentry '%s'\n", dfi->last_name);
123 	return 0;
124 }
125 
126 
127 static struct dentry *
__dcache_find_get_entry(struct dentry * parent,u64 idx,struct ceph_readdir_cache_control * cache_ctl)128 __dcache_find_get_entry(struct dentry *parent, u64 idx,
129 			struct ceph_readdir_cache_control *cache_ctl)
130 {
131 	struct inode *dir = d_inode(parent);
132 	struct dentry *dentry;
133 	unsigned idx_mask = (PAGE_SIZE / sizeof(struct dentry *)) - 1;
134 	loff_t ptr_pos = idx * sizeof(struct dentry *);
135 	pgoff_t ptr_pgoff = ptr_pos >> PAGE_SHIFT;
136 
137 	if (ptr_pos >= i_size_read(dir))
138 		return NULL;
139 
140 	if (!cache_ctl->page || ptr_pgoff != page_index(cache_ctl->page)) {
141 		ceph_readdir_cache_release(cache_ctl);
142 		cache_ctl->page = find_lock_page(&dir->i_data, ptr_pgoff);
143 		if (!cache_ctl->page) {
144 			dout(" page %lu not found\n", ptr_pgoff);
145 			return ERR_PTR(-EAGAIN);
146 		}
147 		/* reading/filling the cache are serialized by
148 		   i_rwsem, no need to use page lock */
149 		unlock_page(cache_ctl->page);
150 		cache_ctl->dentries = kmap(cache_ctl->page);
151 	}
152 
153 	cache_ctl->index = idx & idx_mask;
154 
155 	rcu_read_lock();
156 	spin_lock(&parent->d_lock);
157 	/* check i_size again here, because empty directory can be
158 	 * marked as complete while not holding the i_rwsem. */
159 	if (ceph_dir_is_complete_ordered(dir) && ptr_pos < i_size_read(dir))
160 		dentry = cache_ctl->dentries[cache_ctl->index];
161 	else
162 		dentry = NULL;
163 	spin_unlock(&parent->d_lock);
164 	if (dentry && !lockref_get_not_dead(&dentry->d_lockref))
165 		dentry = NULL;
166 	rcu_read_unlock();
167 	return dentry ? : ERR_PTR(-EAGAIN);
168 }
169 
170 /*
171  * When possible, we try to satisfy a readdir by peeking at the
172  * dcache.  We make this work by carefully ordering dentries on
173  * d_child when we initially get results back from the MDS, and
174  * falling back to a "normal" sync readdir if any dentries in the dir
175  * are dropped.
176  *
177  * Complete dir indicates that we have all dentries in the dir.  It is
178  * defined IFF we hold CEPH_CAP_FILE_SHARED (which will be revoked by
179  * the MDS if/when the directory is modified).
180  */
__dcache_readdir(struct file * file,struct dir_context * ctx,int shared_gen)181 static int __dcache_readdir(struct file *file,  struct dir_context *ctx,
182 			    int shared_gen)
183 {
184 	struct ceph_dir_file_info *dfi = file->private_data;
185 	struct dentry *parent = file->f_path.dentry;
186 	struct inode *dir = d_inode(parent);
187 	struct dentry *dentry, *last = NULL;
188 	struct ceph_dentry_info *di;
189 	struct ceph_readdir_cache_control cache_ctl = {};
190 	u64 idx = 0;
191 	int err = 0;
192 
193 	dout("__dcache_readdir %p v%u at %llx\n", dir, (unsigned)shared_gen, ctx->pos);
194 
195 	/* search start position */
196 	if (ctx->pos > 2) {
197 		u64 count = div_u64(i_size_read(dir), sizeof(struct dentry *));
198 		while (count > 0) {
199 			u64 step = count >> 1;
200 			dentry = __dcache_find_get_entry(parent, idx + step,
201 							 &cache_ctl);
202 			if (!dentry) {
203 				/* use linar search */
204 				idx = 0;
205 				break;
206 			}
207 			if (IS_ERR(dentry)) {
208 				err = PTR_ERR(dentry);
209 				goto out;
210 			}
211 			di = ceph_dentry(dentry);
212 			spin_lock(&dentry->d_lock);
213 			if (fpos_cmp(di->offset, ctx->pos) < 0) {
214 				idx += step + 1;
215 				count -= step + 1;
216 			} else {
217 				count = step;
218 			}
219 			spin_unlock(&dentry->d_lock);
220 			dput(dentry);
221 		}
222 
223 		dout("__dcache_readdir %p cache idx %llu\n", dir, idx);
224 	}
225 
226 
227 	for (;;) {
228 		bool emit_dentry = false;
229 		dentry = __dcache_find_get_entry(parent, idx++, &cache_ctl);
230 		if (!dentry) {
231 			dfi->file_info.flags |= CEPH_F_ATEND;
232 			err = 0;
233 			break;
234 		}
235 		if (IS_ERR(dentry)) {
236 			err = PTR_ERR(dentry);
237 			goto out;
238 		}
239 
240 		spin_lock(&dentry->d_lock);
241 		di = ceph_dentry(dentry);
242 		if (d_unhashed(dentry) ||
243 		    d_really_is_negative(dentry) ||
244 		    di->lease_shared_gen != shared_gen) {
245 			spin_unlock(&dentry->d_lock);
246 			dput(dentry);
247 			err = -EAGAIN;
248 			goto out;
249 		}
250 		if (fpos_cmp(ctx->pos, di->offset) <= 0) {
251 			__ceph_dentry_dir_lease_touch(di);
252 			emit_dentry = true;
253 		}
254 		spin_unlock(&dentry->d_lock);
255 
256 		if (emit_dentry) {
257 			dout(" %llx dentry %p %pd %p\n", di->offset,
258 			     dentry, dentry, d_inode(dentry));
259 			ctx->pos = di->offset;
260 			if (!dir_emit(ctx, dentry->d_name.name,
261 				      dentry->d_name.len, ceph_present_inode(d_inode(dentry)),
262 				      d_inode(dentry)->i_mode >> 12)) {
263 				dput(dentry);
264 				err = 0;
265 				break;
266 			}
267 			ctx->pos++;
268 
269 			if (last)
270 				dput(last);
271 			last = dentry;
272 		} else {
273 			dput(dentry);
274 		}
275 	}
276 out:
277 	ceph_readdir_cache_release(&cache_ctl);
278 	if (last) {
279 		int ret;
280 		di = ceph_dentry(last);
281 		ret = note_last_dentry(dfi, last->d_name.name, last->d_name.len,
282 				       fpos_off(di->offset) + 1);
283 		if (ret < 0)
284 			err = ret;
285 		dput(last);
286 		/* last_name no longer match cache index */
287 		if (dfi->readdir_cache_idx >= 0) {
288 			dfi->readdir_cache_idx = -1;
289 			dfi->dir_release_count = 0;
290 		}
291 	}
292 	return err;
293 }
294 
need_send_readdir(struct ceph_dir_file_info * dfi,loff_t pos)295 static bool need_send_readdir(struct ceph_dir_file_info *dfi, loff_t pos)
296 {
297 	if (!dfi->last_readdir)
298 		return true;
299 	if (is_hash_order(pos))
300 		return !ceph_frag_contains_value(dfi->frag, fpos_hash(pos));
301 	else
302 		return dfi->frag != fpos_frag(pos);
303 }
304 
ceph_readdir(struct file * file,struct dir_context * ctx)305 static int ceph_readdir(struct file *file, struct dir_context *ctx)
306 {
307 	struct ceph_dir_file_info *dfi = file->private_data;
308 	struct inode *inode = file_inode(file);
309 	struct ceph_inode_info *ci = ceph_inode(inode);
310 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
311 	struct ceph_mds_client *mdsc = fsc->mdsc;
312 	int i;
313 	int err;
314 	unsigned frag = -1;
315 	struct ceph_mds_reply_info_parsed *rinfo;
316 
317 	dout("readdir %p file %p pos %llx\n", inode, file, ctx->pos);
318 	if (dfi->file_info.flags & CEPH_F_ATEND)
319 		return 0;
320 
321 	/* always start with . and .. */
322 	if (ctx->pos == 0) {
323 		dout("readdir off 0 -> '.'\n");
324 		if (!dir_emit(ctx, ".", 1, ceph_present_inode(inode),
325 			    inode->i_mode >> 12))
326 			return 0;
327 		ctx->pos = 1;
328 	}
329 	if (ctx->pos == 1) {
330 		u64 ino;
331 		struct dentry *dentry = file->f_path.dentry;
332 
333 		spin_lock(&dentry->d_lock);
334 		ino = ceph_present_inode(dentry->d_parent->d_inode);
335 		spin_unlock(&dentry->d_lock);
336 
337 		dout("readdir off 1 -> '..'\n");
338 		if (!dir_emit(ctx, "..", 2, ino, inode->i_mode >> 12))
339 			return 0;
340 		ctx->pos = 2;
341 	}
342 
343 	spin_lock(&ci->i_ceph_lock);
344 	/* request Fx cap. if have Fx, we don't need to release Fs cap
345 	 * for later create/unlink. */
346 	__ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_WR);
347 	/* can we use the dcache? */
348 	if (ceph_test_mount_opt(fsc, DCACHE) &&
349 	    !ceph_test_mount_opt(fsc, NOASYNCREADDIR) &&
350 	    ceph_snap(inode) != CEPH_SNAPDIR &&
351 	    __ceph_dir_is_complete_ordered(ci) &&
352 	    __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
353 		int shared_gen = atomic_read(&ci->i_shared_gen);
354 
355 		spin_unlock(&ci->i_ceph_lock);
356 		err = __dcache_readdir(file, ctx, shared_gen);
357 		if (err != -EAGAIN)
358 			return err;
359 	} else {
360 		spin_unlock(&ci->i_ceph_lock);
361 	}
362 
363 	/* proceed with a normal readdir */
364 more:
365 	/* do we have the correct frag content buffered? */
366 	if (need_send_readdir(dfi, ctx->pos)) {
367 		struct ceph_mds_request *req;
368 		int op = ceph_snap(inode) == CEPH_SNAPDIR ?
369 			CEPH_MDS_OP_LSSNAP : CEPH_MDS_OP_READDIR;
370 
371 		/* discard old result, if any */
372 		if (dfi->last_readdir) {
373 			ceph_mdsc_put_request(dfi->last_readdir);
374 			dfi->last_readdir = NULL;
375 		}
376 
377 		if (is_hash_order(ctx->pos)) {
378 			/* fragtree isn't always accurate. choose frag
379 			 * based on previous reply when possible. */
380 			if (frag == (unsigned)-1)
381 				frag = ceph_choose_frag(ci, fpos_hash(ctx->pos),
382 							NULL, NULL);
383 		} else {
384 			frag = fpos_frag(ctx->pos);
385 		}
386 
387 		dout("readdir fetching %llx.%llx frag %x offset '%s'\n",
388 		     ceph_vinop(inode), frag, dfi->last_name);
389 		req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
390 		if (IS_ERR(req))
391 			return PTR_ERR(req);
392 		err = ceph_alloc_readdir_reply_buffer(req, inode);
393 		if (err) {
394 			ceph_mdsc_put_request(req);
395 			return err;
396 		}
397 		/* hints to request -> mds selection code */
398 		req->r_direct_mode = USE_AUTH_MDS;
399 		if (op == CEPH_MDS_OP_READDIR) {
400 			req->r_direct_hash = ceph_frag_value(frag);
401 			__set_bit(CEPH_MDS_R_DIRECT_IS_HASH, &req->r_req_flags);
402 			req->r_inode_drop = CEPH_CAP_FILE_EXCL;
403 		}
404 		if (dfi->last_name) {
405 			req->r_path2 = kstrdup(dfi->last_name, GFP_KERNEL);
406 			if (!req->r_path2) {
407 				ceph_mdsc_put_request(req);
408 				return -ENOMEM;
409 			}
410 		} else if (is_hash_order(ctx->pos)) {
411 			req->r_args.readdir.offset_hash =
412 				cpu_to_le32(fpos_hash(ctx->pos));
413 		}
414 
415 		req->r_dir_release_cnt = dfi->dir_release_count;
416 		req->r_dir_ordered_cnt = dfi->dir_ordered_count;
417 		req->r_readdir_cache_idx = dfi->readdir_cache_idx;
418 		req->r_readdir_offset = dfi->next_offset;
419 		req->r_args.readdir.frag = cpu_to_le32(frag);
420 		req->r_args.readdir.flags =
421 				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
422 
423 		req->r_inode = inode;
424 		ihold(inode);
425 		req->r_dentry = dget(file->f_path.dentry);
426 		err = ceph_mdsc_do_request(mdsc, NULL, req);
427 		if (err < 0) {
428 			ceph_mdsc_put_request(req);
429 			return err;
430 		}
431 		dout("readdir got and parsed readdir result=%d on "
432 		     "frag %x, end=%d, complete=%d, hash_order=%d\n",
433 		     err, frag,
434 		     (int)req->r_reply_info.dir_end,
435 		     (int)req->r_reply_info.dir_complete,
436 		     (int)req->r_reply_info.hash_order);
437 
438 		rinfo = &req->r_reply_info;
439 		if (le32_to_cpu(rinfo->dir_dir->frag) != frag) {
440 			frag = le32_to_cpu(rinfo->dir_dir->frag);
441 			if (!rinfo->hash_order) {
442 				dfi->next_offset = req->r_readdir_offset;
443 				/* adjust ctx->pos to beginning of frag */
444 				ctx->pos = ceph_make_fpos(frag,
445 							  dfi->next_offset,
446 							  false);
447 			}
448 		}
449 
450 		dfi->frag = frag;
451 		dfi->last_readdir = req;
452 
453 		if (test_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags)) {
454 			dfi->readdir_cache_idx = req->r_readdir_cache_idx;
455 			if (dfi->readdir_cache_idx < 0) {
456 				/* preclude from marking dir ordered */
457 				dfi->dir_ordered_count = 0;
458 			} else if (ceph_frag_is_leftmost(frag) &&
459 				   dfi->next_offset == 2) {
460 				/* note dir version at start of readdir so
461 				 * we can tell if any dentries get dropped */
462 				dfi->dir_release_count = req->r_dir_release_cnt;
463 				dfi->dir_ordered_count = req->r_dir_ordered_cnt;
464 			}
465 		} else {
466 			dout("readdir !did_prepopulate\n");
467 			/* disable readdir cache */
468 			dfi->readdir_cache_idx = -1;
469 			/* preclude from marking dir complete */
470 			dfi->dir_release_count = 0;
471 		}
472 
473 		/* note next offset and last dentry name */
474 		if (rinfo->dir_nr > 0) {
475 			struct ceph_mds_reply_dir_entry *rde =
476 					rinfo->dir_entries + (rinfo->dir_nr-1);
477 			unsigned next_offset = req->r_reply_info.dir_end ?
478 					2 : (fpos_off(rde->offset) + 1);
479 			err = note_last_dentry(dfi, rde->name, rde->name_len,
480 					       next_offset);
481 			if (err) {
482 				ceph_mdsc_put_request(dfi->last_readdir);
483 				dfi->last_readdir = NULL;
484 				return err;
485 			}
486 		} else if (req->r_reply_info.dir_end) {
487 			dfi->next_offset = 2;
488 			/* keep last name */
489 		}
490 	}
491 
492 	rinfo = &dfi->last_readdir->r_reply_info;
493 	dout("readdir frag %x num %d pos %llx chunk first %llx\n",
494 	     dfi->frag, rinfo->dir_nr, ctx->pos,
495 	     rinfo->dir_nr ? rinfo->dir_entries[0].offset : 0LL);
496 
497 	i = 0;
498 	/* search start position */
499 	if (rinfo->dir_nr > 0) {
500 		int step, nr = rinfo->dir_nr;
501 		while (nr > 0) {
502 			step = nr >> 1;
503 			if (rinfo->dir_entries[i + step].offset < ctx->pos) {
504 				i +=  step + 1;
505 				nr -= step + 1;
506 			} else {
507 				nr = step;
508 			}
509 		}
510 	}
511 	for (; i < rinfo->dir_nr; i++) {
512 		struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
513 
514 		BUG_ON(rde->offset < ctx->pos);
515 
516 		ctx->pos = rde->offset;
517 		dout("readdir (%d/%d) -> %llx '%.*s' %p\n",
518 		     i, rinfo->dir_nr, ctx->pos,
519 		     rde->name_len, rde->name, &rde->inode.in);
520 
521 		BUG_ON(!rde->inode.in);
522 
523 		if (!dir_emit(ctx, rde->name, rde->name_len,
524 			      ceph_present_ino(inode->i_sb, le64_to_cpu(rde->inode.in->ino)),
525 			      le32_to_cpu(rde->inode.in->mode) >> 12)) {
526 			/*
527 			 * NOTE: Here no need to put the 'dfi->last_readdir',
528 			 * because when dir_emit stops us it's most likely
529 			 * doesn't have enough memory, etc. So for next readdir
530 			 * it will continue.
531 			 */
532 			dout("filldir stopping us...\n");
533 			return 0;
534 		}
535 		ctx->pos++;
536 	}
537 
538 	ceph_mdsc_put_request(dfi->last_readdir);
539 	dfi->last_readdir = NULL;
540 
541 	if (dfi->next_offset > 2) {
542 		frag = dfi->frag;
543 		goto more;
544 	}
545 
546 	/* more frags? */
547 	if (!ceph_frag_is_rightmost(dfi->frag)) {
548 		frag = ceph_frag_next(dfi->frag);
549 		if (is_hash_order(ctx->pos)) {
550 			loff_t new_pos = ceph_make_fpos(ceph_frag_value(frag),
551 							dfi->next_offset, true);
552 			if (new_pos > ctx->pos)
553 				ctx->pos = new_pos;
554 			/* keep last_name */
555 		} else {
556 			ctx->pos = ceph_make_fpos(frag, dfi->next_offset,
557 							false);
558 			kfree(dfi->last_name);
559 			dfi->last_name = NULL;
560 		}
561 		dout("readdir next frag is %x\n", frag);
562 		goto more;
563 	}
564 	dfi->file_info.flags |= CEPH_F_ATEND;
565 
566 	/*
567 	 * if dir_release_count still matches the dir, no dentries
568 	 * were released during the whole readdir, and we should have
569 	 * the complete dir contents in our cache.
570 	 */
571 	if (atomic64_read(&ci->i_release_count) ==
572 					dfi->dir_release_count) {
573 		spin_lock(&ci->i_ceph_lock);
574 		if (dfi->dir_ordered_count ==
575 				atomic64_read(&ci->i_ordered_count)) {
576 			dout(" marking %p complete and ordered\n", inode);
577 			/* use i_size to track number of entries in
578 			 * readdir cache */
579 			BUG_ON(dfi->readdir_cache_idx < 0);
580 			i_size_write(inode, dfi->readdir_cache_idx *
581 				     sizeof(struct dentry*));
582 		} else {
583 			dout(" marking %p complete\n", inode);
584 		}
585 		__ceph_dir_set_complete(ci, dfi->dir_release_count,
586 					dfi->dir_ordered_count);
587 		spin_unlock(&ci->i_ceph_lock);
588 	}
589 
590 	dout("readdir %p file %p done.\n", inode, file);
591 	return 0;
592 }
593 
reset_readdir(struct ceph_dir_file_info * dfi)594 static void reset_readdir(struct ceph_dir_file_info *dfi)
595 {
596 	if (dfi->last_readdir) {
597 		ceph_mdsc_put_request(dfi->last_readdir);
598 		dfi->last_readdir = NULL;
599 	}
600 	kfree(dfi->last_name);
601 	dfi->last_name = NULL;
602 	dfi->dir_release_count = 0;
603 	dfi->readdir_cache_idx = -1;
604 	dfi->next_offset = 2;  /* compensate for . and .. */
605 	dfi->file_info.flags &= ~CEPH_F_ATEND;
606 }
607 
608 /*
609  * discard buffered readdir content on seekdir(0), or seek to new frag,
610  * or seek prior to current chunk
611  */
need_reset_readdir(struct ceph_dir_file_info * dfi,loff_t new_pos)612 static bool need_reset_readdir(struct ceph_dir_file_info *dfi, loff_t new_pos)
613 {
614 	struct ceph_mds_reply_info_parsed *rinfo;
615 	loff_t chunk_offset;
616 	if (new_pos == 0)
617 		return true;
618 	if (is_hash_order(new_pos)) {
619 		/* no need to reset last_name for a forward seek when
620 		 * dentries are sotred in hash order */
621 	} else if (dfi->frag != fpos_frag(new_pos)) {
622 		return true;
623 	}
624 	rinfo = dfi->last_readdir ? &dfi->last_readdir->r_reply_info : NULL;
625 	if (!rinfo || !rinfo->dir_nr)
626 		return true;
627 	chunk_offset = rinfo->dir_entries[0].offset;
628 	return new_pos < chunk_offset ||
629 	       is_hash_order(new_pos) != is_hash_order(chunk_offset);
630 }
631 
ceph_dir_llseek(struct file * file,loff_t offset,int whence)632 static loff_t ceph_dir_llseek(struct file *file, loff_t offset, int whence)
633 {
634 	struct ceph_dir_file_info *dfi = file->private_data;
635 	struct inode *inode = file->f_mapping->host;
636 	loff_t retval;
637 
638 	inode_lock(inode);
639 	retval = -EINVAL;
640 	switch (whence) {
641 	case SEEK_CUR:
642 		offset += file->f_pos;
643 		break;
644 	case SEEK_SET:
645 		break;
646 	case SEEK_END:
647 		retval = -EOPNOTSUPP;
648 		goto out;
649 	default:
650 		goto out;
651 	}
652 
653 	if (offset >= 0) {
654 		if (need_reset_readdir(dfi, offset)) {
655 			dout("dir_llseek dropping %p content\n", file);
656 			reset_readdir(dfi);
657 		} else if (is_hash_order(offset) && offset > file->f_pos) {
658 			/* for hash offset, we don't know if a forward seek
659 			 * is within same frag */
660 			dfi->dir_release_count = 0;
661 			dfi->readdir_cache_idx = -1;
662 		}
663 
664 		if (offset != file->f_pos) {
665 			file->f_pos = offset;
666 			file->f_version = 0;
667 			dfi->file_info.flags &= ~CEPH_F_ATEND;
668 		}
669 		retval = offset;
670 	}
671 out:
672 	inode_unlock(inode);
673 	return retval;
674 }
675 
676 /*
677  * Handle lookups for the hidden .snap directory.
678  */
ceph_handle_snapdir(struct ceph_mds_request * req,struct dentry * dentry)679 struct dentry *ceph_handle_snapdir(struct ceph_mds_request *req,
680 				   struct dentry *dentry)
681 {
682 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
683 	struct inode *parent = d_inode(dentry->d_parent); /* we hold i_rwsem */
684 
685 	/* .snap dir? */
686 	if (ceph_snap(parent) == CEPH_NOSNAP &&
687 	    strcmp(dentry->d_name.name, fsc->mount_options->snapdir_name) == 0) {
688 		struct dentry *res;
689 		struct inode *inode = ceph_get_snapdir(parent);
690 
691 		res = d_splice_alias(inode, dentry);
692 		dout("ENOENT on snapdir %p '%pd', linking to snapdir %p. Spliced dentry %p\n",
693 		     dentry, dentry, inode, res);
694 		if (res)
695 			dentry = res;
696 	}
697 	return dentry;
698 }
699 
700 /*
701  * Figure out final result of a lookup/open request.
702  *
703  * Mainly, make sure we return the final req->r_dentry (if it already
704  * existed) in place of the original VFS-provided dentry when they
705  * differ.
706  *
707  * Gracefully handle the case where the MDS replies with -ENOENT and
708  * no trace (which it may do, at its discretion, e.g., if it doesn't
709  * care to issue a lease on the negative dentry).
710  */
ceph_finish_lookup(struct ceph_mds_request * req,struct dentry * dentry,int err)711 struct dentry *ceph_finish_lookup(struct ceph_mds_request *req,
712 				  struct dentry *dentry, int err)
713 {
714 	if (err == -ENOENT) {
715 		/* no trace? */
716 		err = 0;
717 		if (!req->r_reply_info.head->is_dentry) {
718 			dout("ENOENT and no trace, dentry %p inode %p\n",
719 			     dentry, d_inode(dentry));
720 			if (d_really_is_positive(dentry)) {
721 				d_drop(dentry);
722 				err = -ENOENT;
723 			} else {
724 				d_add(dentry, NULL);
725 			}
726 		}
727 	}
728 	if (err)
729 		dentry = ERR_PTR(err);
730 	else if (dentry != req->r_dentry)
731 		dentry = dget(req->r_dentry);   /* we got spliced */
732 	else
733 		dentry = NULL;
734 	return dentry;
735 }
736 
is_root_ceph_dentry(struct inode * inode,struct dentry * dentry)737 static bool is_root_ceph_dentry(struct inode *inode, struct dentry *dentry)
738 {
739 	return ceph_ino(inode) == CEPH_INO_ROOT &&
740 		strncmp(dentry->d_name.name, ".ceph", 5) == 0;
741 }
742 
743 /*
744  * Look up a single dir entry.  If there is a lookup intent, inform
745  * the MDS so that it gets our 'caps wanted' value in a single op.
746  */
ceph_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)747 static struct dentry *ceph_lookup(struct inode *dir, struct dentry *dentry,
748 				  unsigned int flags)
749 {
750 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
751 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
752 	struct ceph_mds_request *req;
753 	int op;
754 	int mask;
755 	int err;
756 
757 	dout("lookup %p dentry %p '%pd'\n",
758 	     dir, dentry, dentry);
759 
760 	if (dentry->d_name.len > NAME_MAX)
761 		return ERR_PTR(-ENAMETOOLONG);
762 
763 	/* can we conclude ENOENT locally? */
764 	if (d_really_is_negative(dentry)) {
765 		struct ceph_inode_info *ci = ceph_inode(dir);
766 		struct ceph_dentry_info *di = ceph_dentry(dentry);
767 
768 		spin_lock(&ci->i_ceph_lock);
769 		dout(" dir %p flags are 0x%lx\n", dir, ci->i_ceph_flags);
770 		if (strncmp(dentry->d_name.name,
771 			    fsc->mount_options->snapdir_name,
772 			    dentry->d_name.len) &&
773 		    !is_root_ceph_dentry(dir, dentry) &&
774 		    ceph_test_mount_opt(fsc, DCACHE) &&
775 		    __ceph_dir_is_complete(ci) &&
776 		    __ceph_caps_issued_mask_metric(ci, CEPH_CAP_FILE_SHARED, 1)) {
777 			__ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
778 			spin_unlock(&ci->i_ceph_lock);
779 			dout(" dir %p complete, -ENOENT\n", dir);
780 			d_add(dentry, NULL);
781 			di->lease_shared_gen = atomic_read(&ci->i_shared_gen);
782 			return NULL;
783 		}
784 		spin_unlock(&ci->i_ceph_lock);
785 	}
786 
787 	op = ceph_snap(dir) == CEPH_SNAPDIR ?
788 		CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
789 	req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
790 	if (IS_ERR(req))
791 		return ERR_CAST(req);
792 	req->r_dentry = dget(dentry);
793 	req->r_num_caps = 2;
794 
795 	mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
796 	if (ceph_security_xattr_wanted(dir))
797 		mask |= CEPH_CAP_XATTR_SHARED;
798 	req->r_args.getattr.mask = cpu_to_le32(mask);
799 
800 	ihold(dir);
801 	req->r_parent = dir;
802 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
803 	err = ceph_mdsc_do_request(mdsc, NULL, req);
804 	if (err == -ENOENT) {
805 		struct dentry *res;
806 
807 		res = ceph_handle_snapdir(req, dentry);
808 		if (IS_ERR(res)) {
809 			err = PTR_ERR(res);
810 		} else {
811 			dentry = res;
812 			err = 0;
813 		}
814 	}
815 	dentry = ceph_finish_lookup(req, dentry, err);
816 	ceph_mdsc_put_request(req);  /* will dput(dentry) */
817 	dout("lookup result=%p\n", dentry);
818 	return dentry;
819 }
820 
821 /*
822  * If we do a create but get no trace back from the MDS, follow up with
823  * a lookup (the VFS expects us to link up the provided dentry).
824  */
ceph_handle_notrace_create(struct inode * dir,struct dentry * dentry)825 int ceph_handle_notrace_create(struct inode *dir, struct dentry *dentry)
826 {
827 	struct dentry *result = ceph_lookup(dir, dentry, 0);
828 
829 	if (result && !IS_ERR(result)) {
830 		/*
831 		 * We created the item, then did a lookup, and found
832 		 * it was already linked to another inode we already
833 		 * had in our cache (and thus got spliced). To not
834 		 * confuse VFS (especially when inode is a directory),
835 		 * we don't link our dentry to that inode, return an
836 		 * error instead.
837 		 *
838 		 * This event should be rare and it happens only when
839 		 * we talk to old MDS. Recent MDS does not send traceless
840 		 * reply for request that creates new inode.
841 		 */
842 		d_drop(result);
843 		return -ESTALE;
844 	}
845 	return PTR_ERR(result);
846 }
847 
ceph_mknod(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)848 static int ceph_mknod(struct user_namespace *mnt_userns, struct inode *dir,
849 		      struct dentry *dentry, umode_t mode, dev_t rdev)
850 {
851 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
852 	struct ceph_mds_request *req;
853 	struct ceph_acl_sec_ctx as_ctx = {};
854 	int err;
855 
856 	if (ceph_snap(dir) != CEPH_NOSNAP)
857 		return -EROFS;
858 
859 	err = ceph_wait_on_conflict_unlink(dentry);
860 	if (err)
861 		return err;
862 
863 	if (ceph_quota_is_max_files_exceeded(dir)) {
864 		err = -EDQUOT;
865 		goto out;
866 	}
867 
868 	err = ceph_pre_init_acls(dir, &mode, &as_ctx);
869 	if (err < 0)
870 		goto out;
871 	err = ceph_security_init_secctx(dentry, mode, &as_ctx);
872 	if (err < 0)
873 		goto out;
874 
875 	dout("mknod in dir %p dentry %p mode 0%ho rdev %d\n",
876 	     dir, dentry, mode, rdev);
877 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_MKNOD, USE_AUTH_MDS);
878 	if (IS_ERR(req)) {
879 		err = PTR_ERR(req);
880 		goto out;
881 	}
882 	req->r_dentry = dget(dentry);
883 	req->r_num_caps = 2;
884 	req->r_parent = dir;
885 	ihold(dir);
886 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
887 	req->r_args.mknod.mode = cpu_to_le32(mode);
888 	req->r_args.mknod.rdev = cpu_to_le32(rdev);
889 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
890 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
891 	if (as_ctx.pagelist) {
892 		req->r_pagelist = as_ctx.pagelist;
893 		as_ctx.pagelist = NULL;
894 	}
895 	err = ceph_mdsc_do_request(mdsc, dir, req);
896 	if (!err && !req->r_reply_info.head->is_dentry)
897 		err = ceph_handle_notrace_create(dir, dentry);
898 	ceph_mdsc_put_request(req);
899 out:
900 	if (!err)
901 		ceph_init_inode_acls(d_inode(dentry), &as_ctx);
902 	else
903 		d_drop(dentry);
904 	ceph_release_acl_sec_ctx(&as_ctx);
905 	return err;
906 }
907 
ceph_create(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)908 static int ceph_create(struct user_namespace *mnt_userns, struct inode *dir,
909 		       struct dentry *dentry, umode_t mode, bool excl)
910 {
911 	return ceph_mknod(mnt_userns, dir, dentry, mode, 0);
912 }
913 
ceph_symlink(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,const char * dest)914 static int ceph_symlink(struct user_namespace *mnt_userns, struct inode *dir,
915 			struct dentry *dentry, const char *dest)
916 {
917 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
918 	struct ceph_mds_request *req;
919 	struct ceph_acl_sec_ctx as_ctx = {};
920 	int err;
921 
922 	if (ceph_snap(dir) != CEPH_NOSNAP)
923 		return -EROFS;
924 
925 	err = ceph_wait_on_conflict_unlink(dentry);
926 	if (err)
927 		return err;
928 
929 	if (ceph_quota_is_max_files_exceeded(dir)) {
930 		err = -EDQUOT;
931 		goto out;
932 	}
933 
934 	err = ceph_security_init_secctx(dentry, S_IFLNK | 0777, &as_ctx);
935 	if (err < 0)
936 		goto out;
937 
938 	dout("symlink in dir %p dentry %p to '%s'\n", dir, dentry, dest);
939 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SYMLINK, USE_AUTH_MDS);
940 	if (IS_ERR(req)) {
941 		err = PTR_ERR(req);
942 		goto out;
943 	}
944 	req->r_path2 = kstrdup(dest, GFP_KERNEL);
945 	if (!req->r_path2) {
946 		err = -ENOMEM;
947 		ceph_mdsc_put_request(req);
948 		goto out;
949 	}
950 	req->r_parent = dir;
951 	ihold(dir);
952 
953 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
954 	req->r_dentry = dget(dentry);
955 	req->r_num_caps = 2;
956 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
957 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
958 	if (as_ctx.pagelist) {
959 		req->r_pagelist = as_ctx.pagelist;
960 		as_ctx.pagelist = NULL;
961 	}
962 	err = ceph_mdsc_do_request(mdsc, dir, req);
963 	if (!err && !req->r_reply_info.head->is_dentry)
964 		err = ceph_handle_notrace_create(dir, dentry);
965 	ceph_mdsc_put_request(req);
966 out:
967 	if (err)
968 		d_drop(dentry);
969 	ceph_release_acl_sec_ctx(&as_ctx);
970 	return err;
971 }
972 
ceph_mkdir(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode)973 static int ceph_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
974 		      struct dentry *dentry, umode_t mode)
975 {
976 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
977 	struct ceph_mds_request *req;
978 	struct ceph_acl_sec_ctx as_ctx = {};
979 	int err;
980 	int op;
981 
982 	err = ceph_wait_on_conflict_unlink(dentry);
983 	if (err)
984 		return err;
985 
986 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
987 		/* mkdir .snap/foo is a MKSNAP */
988 		op = CEPH_MDS_OP_MKSNAP;
989 		dout("mksnap dir %p snap '%pd' dn %p\n", dir,
990 		     dentry, dentry);
991 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
992 		dout("mkdir dir %p dn %p mode 0%ho\n", dir, dentry, mode);
993 		op = CEPH_MDS_OP_MKDIR;
994 	} else {
995 		err = -EROFS;
996 		goto out;
997 	}
998 
999 	if (op == CEPH_MDS_OP_MKDIR &&
1000 	    ceph_quota_is_max_files_exceeded(dir)) {
1001 		err = -EDQUOT;
1002 		goto out;
1003 	}
1004 
1005 	mode |= S_IFDIR;
1006 	err = ceph_pre_init_acls(dir, &mode, &as_ctx);
1007 	if (err < 0)
1008 		goto out;
1009 	err = ceph_security_init_secctx(dentry, mode, &as_ctx);
1010 	if (err < 0)
1011 		goto out;
1012 
1013 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1014 	if (IS_ERR(req)) {
1015 		err = PTR_ERR(req);
1016 		goto out;
1017 	}
1018 
1019 	req->r_dentry = dget(dentry);
1020 	req->r_num_caps = 2;
1021 	req->r_parent = dir;
1022 	ihold(dir);
1023 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1024 	req->r_args.mkdir.mode = cpu_to_le32(mode);
1025 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED | CEPH_CAP_AUTH_EXCL;
1026 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1027 	if (as_ctx.pagelist) {
1028 		req->r_pagelist = as_ctx.pagelist;
1029 		as_ctx.pagelist = NULL;
1030 	}
1031 	err = ceph_mdsc_do_request(mdsc, dir, req);
1032 	if (!err &&
1033 	    !req->r_reply_info.head->is_target &&
1034 	    !req->r_reply_info.head->is_dentry)
1035 		err = ceph_handle_notrace_create(dir, dentry);
1036 	ceph_mdsc_put_request(req);
1037 out:
1038 	if (!err)
1039 		ceph_init_inode_acls(d_inode(dentry), &as_ctx);
1040 	else
1041 		d_drop(dentry);
1042 	ceph_release_acl_sec_ctx(&as_ctx);
1043 	return err;
1044 }
1045 
ceph_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)1046 static int ceph_link(struct dentry *old_dentry, struct inode *dir,
1047 		     struct dentry *dentry)
1048 {
1049 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(dir->i_sb);
1050 	struct ceph_mds_request *req;
1051 	int err;
1052 
1053 	err = ceph_wait_on_conflict_unlink(dentry);
1054 	if (err)
1055 		return err;
1056 
1057 	if (ceph_snap(dir) != CEPH_NOSNAP)
1058 		return -EROFS;
1059 
1060 	dout("link in dir %p old_dentry %p dentry %p\n", dir,
1061 	     old_dentry, dentry);
1062 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LINK, USE_AUTH_MDS);
1063 	if (IS_ERR(req)) {
1064 		d_drop(dentry);
1065 		return PTR_ERR(req);
1066 	}
1067 	req->r_dentry = dget(dentry);
1068 	req->r_num_caps = 2;
1069 	req->r_old_dentry = dget(old_dentry);
1070 	req->r_parent = dir;
1071 	ihold(dir);
1072 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1073 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1074 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1075 	/* release LINK_SHARED on source inode (mds will lock it) */
1076 	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1077 	err = ceph_mdsc_do_request(mdsc, dir, req);
1078 	if (err) {
1079 		d_drop(dentry);
1080 	} else if (!req->r_reply_info.head->is_dentry) {
1081 		ihold(d_inode(old_dentry));
1082 		d_instantiate(dentry, d_inode(old_dentry));
1083 	}
1084 	ceph_mdsc_put_request(req);
1085 	return err;
1086 }
1087 
ceph_async_unlink_cb(struct ceph_mds_client * mdsc,struct ceph_mds_request * req)1088 static void ceph_async_unlink_cb(struct ceph_mds_client *mdsc,
1089 				 struct ceph_mds_request *req)
1090 {
1091 	struct dentry *dentry = req->r_dentry;
1092 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
1093 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1094 	int result = req->r_err ? req->r_err :
1095 			le32_to_cpu(req->r_reply_info.head->result);
1096 
1097 	if (!test_bit(CEPH_DENTRY_ASYNC_UNLINK_BIT, &di->flags))
1098 		pr_warn("%s dentry %p:%pd async unlink bit is not set\n",
1099 			__func__, dentry, dentry);
1100 
1101 	spin_lock(&fsc->async_unlink_conflict_lock);
1102 	hash_del_rcu(&di->hnode);
1103 	spin_unlock(&fsc->async_unlink_conflict_lock);
1104 
1105 	spin_lock(&dentry->d_lock);
1106 	di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK;
1107 	wake_up_bit(&di->flags, CEPH_DENTRY_ASYNC_UNLINK_BIT);
1108 	spin_unlock(&dentry->d_lock);
1109 
1110 	synchronize_rcu();
1111 
1112 	if (result == -EJUKEBOX)
1113 		goto out;
1114 
1115 	/* If op failed, mark everyone involved for errors */
1116 	if (result) {
1117 		int pathlen = 0;
1118 		u64 base = 0;
1119 		char *path = ceph_mdsc_build_path(dentry, &pathlen,
1120 						  &base, 0);
1121 
1122 		/* mark error on parent + clear complete */
1123 		mapping_set_error(req->r_parent->i_mapping, result);
1124 		ceph_dir_clear_complete(req->r_parent);
1125 
1126 		/* drop the dentry -- we don't know its status */
1127 		if (!d_unhashed(dentry))
1128 			d_drop(dentry);
1129 
1130 		/* mark inode itself for an error (since metadata is bogus) */
1131 		mapping_set_error(req->r_old_inode->i_mapping, result);
1132 
1133 		pr_warn("async unlink failure path=(%llx)%s result=%d!\n",
1134 			base, IS_ERR(path) ? "<<bad>>" : path, result);
1135 		ceph_mdsc_free_path(path, pathlen);
1136 	}
1137 out:
1138 	iput(req->r_old_inode);
1139 	ceph_mdsc_release_dir_caps(req);
1140 }
1141 
get_caps_for_async_unlink(struct inode * dir,struct dentry * dentry)1142 static int get_caps_for_async_unlink(struct inode *dir, struct dentry *dentry)
1143 {
1144 	struct ceph_inode_info *ci = ceph_inode(dir);
1145 	struct ceph_dentry_info *di;
1146 	int got = 0, want = CEPH_CAP_FILE_EXCL | CEPH_CAP_DIR_UNLINK;
1147 
1148 	spin_lock(&ci->i_ceph_lock);
1149 	if ((__ceph_caps_issued(ci, NULL) & want) == want) {
1150 		ceph_take_cap_refs(ci, want, false);
1151 		got = want;
1152 	}
1153 	spin_unlock(&ci->i_ceph_lock);
1154 
1155 	/* If we didn't get anything, return 0 */
1156 	if (!got)
1157 		return 0;
1158 
1159         spin_lock(&dentry->d_lock);
1160         di = ceph_dentry(dentry);
1161 	/*
1162 	 * - We are holding Fx, which implies Fs caps.
1163 	 * - Only support async unlink for primary linkage
1164 	 */
1165 	if (atomic_read(&ci->i_shared_gen) != di->lease_shared_gen ||
1166 	    !(di->flags & CEPH_DENTRY_PRIMARY_LINK))
1167 		want = 0;
1168         spin_unlock(&dentry->d_lock);
1169 
1170 	/* Do we still want what we've got? */
1171 	if (want == got)
1172 		return got;
1173 
1174 	ceph_put_cap_refs(ci, got);
1175 	return 0;
1176 }
1177 
1178 /*
1179  * rmdir and unlink are differ only by the metadata op code
1180  */
ceph_unlink(struct inode * dir,struct dentry * dentry)1181 static int ceph_unlink(struct inode *dir, struct dentry *dentry)
1182 {
1183 	struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb);
1184 	struct ceph_mds_client *mdsc = fsc->mdsc;
1185 	struct inode *inode = d_inode(dentry);
1186 	struct ceph_mds_request *req;
1187 	bool try_async = ceph_test_mount_opt(fsc, ASYNC_DIROPS);
1188 	int err = -EROFS;
1189 	int op;
1190 
1191 	if (ceph_snap(dir) == CEPH_SNAPDIR) {
1192 		/* rmdir .snap/foo is RMSNAP */
1193 		dout("rmsnap dir %p '%pd' dn %p\n", dir, dentry, dentry);
1194 		op = CEPH_MDS_OP_RMSNAP;
1195 	} else if (ceph_snap(dir) == CEPH_NOSNAP) {
1196 		dout("unlink/rmdir dir %p dn %p inode %p\n",
1197 		     dir, dentry, inode);
1198 		op = d_is_dir(dentry) ?
1199 			CEPH_MDS_OP_RMDIR : CEPH_MDS_OP_UNLINK;
1200 	} else
1201 		goto out;
1202 retry:
1203 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1204 	if (IS_ERR(req)) {
1205 		err = PTR_ERR(req);
1206 		goto out;
1207 	}
1208 	req->r_dentry = dget(dentry);
1209 	req->r_num_caps = 2;
1210 	req->r_parent = dir;
1211 	ihold(dir);
1212 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1213 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1214 	req->r_inode_drop = ceph_drop_caps_for_unlink(inode);
1215 
1216 	if (try_async && op == CEPH_MDS_OP_UNLINK &&
1217 	    (req->r_dir_caps = get_caps_for_async_unlink(dir, dentry))) {
1218 		struct ceph_dentry_info *di = ceph_dentry(dentry);
1219 
1220 		dout("async unlink on %llu/%.*s caps=%s", ceph_ino(dir),
1221 		     dentry->d_name.len, dentry->d_name.name,
1222 		     ceph_cap_string(req->r_dir_caps));
1223 		set_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags);
1224 		req->r_callback = ceph_async_unlink_cb;
1225 		req->r_old_inode = d_inode(dentry);
1226 		ihold(req->r_old_inode);
1227 
1228 		spin_lock(&dentry->d_lock);
1229 		di->flags |= CEPH_DENTRY_ASYNC_UNLINK;
1230 		spin_unlock(&dentry->d_lock);
1231 
1232 		spin_lock(&fsc->async_unlink_conflict_lock);
1233 		hash_add_rcu(fsc->async_unlink_conflict, &di->hnode,
1234 			     dentry->d_name.hash);
1235 		spin_unlock(&fsc->async_unlink_conflict_lock);
1236 
1237 		err = ceph_mdsc_submit_request(mdsc, dir, req);
1238 		if (!err) {
1239 			/*
1240 			 * We have enough caps, so we assume that the unlink
1241 			 * will succeed. Fix up the target inode and dcache.
1242 			 */
1243 			drop_nlink(inode);
1244 			d_delete(dentry);
1245 		} else {
1246 			spin_lock(&fsc->async_unlink_conflict_lock);
1247 			hash_del_rcu(&di->hnode);
1248 			spin_unlock(&fsc->async_unlink_conflict_lock);
1249 
1250 			spin_lock(&dentry->d_lock);
1251 			di->flags &= ~CEPH_DENTRY_ASYNC_UNLINK;
1252 			spin_unlock(&dentry->d_lock);
1253 
1254 			if (err == -EJUKEBOX) {
1255 				try_async = false;
1256 				ceph_mdsc_put_request(req);
1257 				goto retry;
1258 			}
1259 		}
1260 	} else {
1261 		set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1262 		err = ceph_mdsc_do_request(mdsc, dir, req);
1263 		if (!err && !req->r_reply_info.head->is_dentry)
1264 			d_delete(dentry);
1265 	}
1266 
1267 	ceph_mdsc_put_request(req);
1268 out:
1269 	return err;
1270 }
1271 
ceph_rename(struct user_namespace * mnt_userns,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)1272 static int ceph_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
1273 		       struct dentry *old_dentry, struct inode *new_dir,
1274 		       struct dentry *new_dentry, unsigned int flags)
1275 {
1276 	struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(old_dir->i_sb);
1277 	struct ceph_mds_request *req;
1278 	int op = CEPH_MDS_OP_RENAME;
1279 	int err;
1280 
1281 	if (flags)
1282 		return -EINVAL;
1283 
1284 	if (ceph_snap(old_dir) != ceph_snap(new_dir))
1285 		return -EXDEV;
1286 	if (ceph_snap(old_dir) != CEPH_NOSNAP) {
1287 		if (old_dir == new_dir && ceph_snap(old_dir) == CEPH_SNAPDIR)
1288 			op = CEPH_MDS_OP_RENAMESNAP;
1289 		else
1290 			return -EROFS;
1291 	}
1292 	/* don't allow cross-quota renames */
1293 	if ((old_dir != new_dir) &&
1294 	    (!ceph_quota_is_same_realm(old_dir, new_dir)))
1295 		return -EXDEV;
1296 
1297 	err = ceph_wait_on_conflict_unlink(new_dentry);
1298 	if (err)
1299 		return err;
1300 
1301 	dout("rename dir %p dentry %p to dir %p dentry %p\n",
1302 	     old_dir, old_dentry, new_dir, new_dentry);
1303 	req = ceph_mdsc_create_request(mdsc, op, USE_AUTH_MDS);
1304 	if (IS_ERR(req))
1305 		return PTR_ERR(req);
1306 	ihold(old_dir);
1307 	req->r_dentry = dget(new_dentry);
1308 	req->r_num_caps = 2;
1309 	req->r_old_dentry = dget(old_dentry);
1310 	req->r_old_dentry_dir = old_dir;
1311 	req->r_parent = new_dir;
1312 	ihold(new_dir);
1313 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
1314 	req->r_old_dentry_drop = CEPH_CAP_FILE_SHARED;
1315 	req->r_old_dentry_unless = CEPH_CAP_FILE_EXCL;
1316 	req->r_dentry_drop = CEPH_CAP_FILE_SHARED;
1317 	req->r_dentry_unless = CEPH_CAP_FILE_EXCL;
1318 	/* release LINK_RDCACHE on source inode (mds will lock it) */
1319 	req->r_old_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL;
1320 	if (d_really_is_positive(new_dentry)) {
1321 		req->r_inode_drop =
1322 			ceph_drop_caps_for_unlink(d_inode(new_dentry));
1323 	}
1324 	err = ceph_mdsc_do_request(mdsc, old_dir, req);
1325 	if (!err && !req->r_reply_info.head->is_dentry) {
1326 		/*
1327 		 * Normally d_move() is done by fill_trace (called by
1328 		 * do_request, above).  If there is no trace, we need
1329 		 * to do it here.
1330 		 */
1331 		d_move(old_dentry, new_dentry);
1332 	}
1333 	ceph_mdsc_put_request(req);
1334 	return err;
1335 }
1336 
1337 /*
1338  * Move dentry to tail of mdsc->dentry_leases list when lease is updated.
1339  * Leases at front of the list will expire first. (Assume all leases have
1340  * similar duration)
1341  *
1342  * Called under dentry->d_lock.
1343  */
__ceph_dentry_lease_touch(struct ceph_dentry_info * di)1344 void __ceph_dentry_lease_touch(struct ceph_dentry_info *di)
1345 {
1346 	struct dentry *dn = di->dentry;
1347 	struct ceph_mds_client *mdsc;
1348 
1349 	dout("dentry_lease_touch %p %p '%pd'\n", di, dn, dn);
1350 
1351 	di->flags |= CEPH_DENTRY_LEASE_LIST;
1352 	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1353 		di->flags |= CEPH_DENTRY_REFERENCED;
1354 		return;
1355 	}
1356 
1357 	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1358 	spin_lock(&mdsc->dentry_list_lock);
1359 	list_move_tail(&di->lease_list, &mdsc->dentry_leases);
1360 	spin_unlock(&mdsc->dentry_list_lock);
1361 }
1362 
__dentry_dir_lease_touch(struct ceph_mds_client * mdsc,struct ceph_dentry_info * di)1363 static void __dentry_dir_lease_touch(struct ceph_mds_client* mdsc,
1364 				     struct ceph_dentry_info *di)
1365 {
1366 	di->flags &= ~(CEPH_DENTRY_LEASE_LIST | CEPH_DENTRY_REFERENCED);
1367 	di->lease_gen = 0;
1368 	di->time = jiffies;
1369 	list_move_tail(&di->lease_list, &mdsc->dentry_dir_leases);
1370 }
1371 
1372 /*
1373  * When dir lease is used, add dentry to tail of mdsc->dentry_dir_leases
1374  * list if it's not in the list, otherwise set 'referenced' flag.
1375  *
1376  * Called under dentry->d_lock.
1377  */
__ceph_dentry_dir_lease_touch(struct ceph_dentry_info * di)1378 void __ceph_dentry_dir_lease_touch(struct ceph_dentry_info *di)
1379 {
1380 	struct dentry *dn = di->dentry;
1381 	struct ceph_mds_client *mdsc;
1382 
1383 	dout("dentry_dir_lease_touch %p %p '%pd' (offset 0x%llx)\n",
1384 	     di, dn, dn, di->offset);
1385 
1386 	if (!list_empty(&di->lease_list)) {
1387 		if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1388 			/* don't remove dentry from dentry lease list
1389 			 * if its lease is valid */
1390 			if (__dentry_lease_is_valid(di))
1391 				return;
1392 		} else {
1393 			di->flags |= CEPH_DENTRY_REFERENCED;
1394 			return;
1395 		}
1396 	}
1397 
1398 	if (di->flags & CEPH_DENTRY_SHRINK_LIST) {
1399 		di->flags |= CEPH_DENTRY_REFERENCED;
1400 		di->flags &= ~CEPH_DENTRY_LEASE_LIST;
1401 		return;
1402 	}
1403 
1404 	mdsc = ceph_sb_to_client(dn->d_sb)->mdsc;
1405 	spin_lock(&mdsc->dentry_list_lock);
1406 	__dentry_dir_lease_touch(mdsc, di),
1407 	spin_unlock(&mdsc->dentry_list_lock);
1408 }
1409 
__dentry_lease_unlist(struct ceph_dentry_info * di)1410 static void __dentry_lease_unlist(struct ceph_dentry_info *di)
1411 {
1412 	struct ceph_mds_client *mdsc;
1413 	if (di->flags & CEPH_DENTRY_SHRINK_LIST)
1414 		return;
1415 	if (list_empty(&di->lease_list))
1416 		return;
1417 
1418 	mdsc = ceph_sb_to_client(di->dentry->d_sb)->mdsc;
1419 	spin_lock(&mdsc->dentry_list_lock);
1420 	list_del_init(&di->lease_list);
1421 	spin_unlock(&mdsc->dentry_list_lock);
1422 }
1423 
1424 enum {
1425 	KEEP	= 0,
1426 	DELETE	= 1,
1427 	TOUCH	= 2,
1428 	STOP	= 4,
1429 };
1430 
1431 struct ceph_lease_walk_control {
1432 	bool dir_lease;
1433 	bool expire_dir_lease;
1434 	unsigned long nr_to_scan;
1435 	unsigned long dir_lease_ttl;
1436 };
1437 
1438 static unsigned long
__dentry_leases_walk(struct ceph_mds_client * mdsc,struct ceph_lease_walk_control * lwc,int (* check)(struct dentry *,void *))1439 __dentry_leases_walk(struct ceph_mds_client *mdsc,
1440 		     struct ceph_lease_walk_control *lwc,
1441 		     int (*check)(struct dentry*, void*))
1442 {
1443 	struct ceph_dentry_info *di, *tmp;
1444 	struct dentry *dentry, *last = NULL;
1445 	struct list_head* list;
1446         LIST_HEAD(dispose);
1447 	unsigned long freed = 0;
1448 	int ret = 0;
1449 
1450 	list = lwc->dir_lease ? &mdsc->dentry_dir_leases : &mdsc->dentry_leases;
1451 	spin_lock(&mdsc->dentry_list_lock);
1452 	list_for_each_entry_safe(di, tmp, list, lease_list) {
1453 		if (!lwc->nr_to_scan)
1454 			break;
1455 		--lwc->nr_to_scan;
1456 
1457 		dentry = di->dentry;
1458 		if (last == dentry)
1459 			break;
1460 
1461 		if (!spin_trylock(&dentry->d_lock))
1462 			continue;
1463 
1464 		if (__lockref_is_dead(&dentry->d_lockref)) {
1465 			list_del_init(&di->lease_list);
1466 			goto next;
1467 		}
1468 
1469 		ret = check(dentry, lwc);
1470 		if (ret & TOUCH) {
1471 			/* move it into tail of dir lease list */
1472 			__dentry_dir_lease_touch(mdsc, di);
1473 			if (!last)
1474 				last = dentry;
1475 		}
1476 		if (ret & DELETE) {
1477 			/* stale lease */
1478 			di->flags &= ~CEPH_DENTRY_REFERENCED;
1479 			if (dentry->d_lockref.count > 0) {
1480 				/* update_dentry_lease() will re-add
1481 				 * it to lease list, or
1482 				 * ceph_d_delete() will return 1 when
1483 				 * last reference is dropped */
1484 				list_del_init(&di->lease_list);
1485 			} else {
1486 				di->flags |= CEPH_DENTRY_SHRINK_LIST;
1487 				list_move_tail(&di->lease_list, &dispose);
1488 				dget_dlock(dentry);
1489 			}
1490 		}
1491 next:
1492 		spin_unlock(&dentry->d_lock);
1493 		if (ret & STOP)
1494 			break;
1495 	}
1496 	spin_unlock(&mdsc->dentry_list_lock);
1497 
1498 	while (!list_empty(&dispose)) {
1499 		di = list_first_entry(&dispose, struct ceph_dentry_info,
1500 				      lease_list);
1501 		dentry = di->dentry;
1502 		spin_lock(&dentry->d_lock);
1503 
1504 		list_del_init(&di->lease_list);
1505 		di->flags &= ~CEPH_DENTRY_SHRINK_LIST;
1506 		if (di->flags & CEPH_DENTRY_REFERENCED) {
1507 			spin_lock(&mdsc->dentry_list_lock);
1508 			if (di->flags & CEPH_DENTRY_LEASE_LIST) {
1509 				list_add_tail(&di->lease_list,
1510 					      &mdsc->dentry_leases);
1511 			} else {
1512 				__dentry_dir_lease_touch(mdsc, di);
1513 			}
1514 			spin_unlock(&mdsc->dentry_list_lock);
1515 		} else {
1516 			freed++;
1517 		}
1518 
1519 		spin_unlock(&dentry->d_lock);
1520 		/* ceph_d_delete() does the trick */
1521 		dput(dentry);
1522 	}
1523 	return freed;
1524 }
1525 
__dentry_lease_check(struct dentry * dentry,void * arg)1526 static int __dentry_lease_check(struct dentry *dentry, void *arg)
1527 {
1528 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1529 	int ret;
1530 
1531 	if (__dentry_lease_is_valid(di))
1532 		return STOP;
1533 	ret = __dir_lease_try_check(dentry);
1534 	if (ret == -EBUSY)
1535 		return KEEP;
1536 	if (ret > 0)
1537 		return TOUCH;
1538 	return DELETE;
1539 }
1540 
__dir_lease_check(struct dentry * dentry,void * arg)1541 static int __dir_lease_check(struct dentry *dentry, void *arg)
1542 {
1543 	struct ceph_lease_walk_control *lwc = arg;
1544 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1545 
1546 	int ret = __dir_lease_try_check(dentry);
1547 	if (ret == -EBUSY)
1548 		return KEEP;
1549 	if (ret > 0) {
1550 		if (time_before(jiffies, di->time + lwc->dir_lease_ttl))
1551 			return STOP;
1552 		/* Move dentry to tail of dir lease list if we don't want
1553 		 * to delete it. So dentries in the list are checked in a
1554 		 * round robin manner */
1555 		if (!lwc->expire_dir_lease)
1556 			return TOUCH;
1557 		if (dentry->d_lockref.count > 0 ||
1558 		    (di->flags & CEPH_DENTRY_REFERENCED))
1559 			return TOUCH;
1560 		/* invalidate dir lease */
1561 		di->lease_shared_gen = 0;
1562 	}
1563 	return DELETE;
1564 }
1565 
ceph_trim_dentries(struct ceph_mds_client * mdsc)1566 int ceph_trim_dentries(struct ceph_mds_client *mdsc)
1567 {
1568 	struct ceph_lease_walk_control lwc;
1569 	unsigned long count;
1570 	unsigned long freed;
1571 
1572 	spin_lock(&mdsc->caps_list_lock);
1573         if (mdsc->caps_use_max > 0 &&
1574             mdsc->caps_use_count > mdsc->caps_use_max)
1575 		count = mdsc->caps_use_count - mdsc->caps_use_max;
1576 	else
1577 		count = 0;
1578         spin_unlock(&mdsc->caps_list_lock);
1579 
1580 	lwc.dir_lease = false;
1581 	lwc.nr_to_scan  = CEPH_CAPS_PER_RELEASE * 2;
1582 	freed = __dentry_leases_walk(mdsc, &lwc, __dentry_lease_check);
1583 	if (!lwc.nr_to_scan) /* more invalid leases */
1584 		return -EAGAIN;
1585 
1586 	if (lwc.nr_to_scan < CEPH_CAPS_PER_RELEASE)
1587 		lwc.nr_to_scan = CEPH_CAPS_PER_RELEASE;
1588 
1589 	lwc.dir_lease = true;
1590 	lwc.expire_dir_lease = freed < count;
1591 	lwc.dir_lease_ttl = mdsc->fsc->mount_options->caps_wanted_delay_max * HZ;
1592 	freed +=__dentry_leases_walk(mdsc, &lwc, __dir_lease_check);
1593 	if (!lwc.nr_to_scan) /* more to check */
1594 		return -EAGAIN;
1595 
1596 	return freed > 0 ? 1 : 0;
1597 }
1598 
1599 /*
1600  * Ensure a dentry lease will no longer revalidate.
1601  */
ceph_invalidate_dentry_lease(struct dentry * dentry)1602 void ceph_invalidate_dentry_lease(struct dentry *dentry)
1603 {
1604 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1605 	spin_lock(&dentry->d_lock);
1606 	di->time = jiffies;
1607 	di->lease_shared_gen = 0;
1608 	di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1609 	__dentry_lease_unlist(di);
1610 	spin_unlock(&dentry->d_lock);
1611 }
1612 
1613 /*
1614  * Check if dentry lease is valid.  If not, delete the lease.  Try to
1615  * renew if the least is more than half up.
1616  */
__dentry_lease_is_valid(struct ceph_dentry_info * di)1617 static bool __dentry_lease_is_valid(struct ceph_dentry_info *di)
1618 {
1619 	struct ceph_mds_session *session;
1620 
1621 	if (!di->lease_gen)
1622 		return false;
1623 
1624 	session = di->lease_session;
1625 	if (session) {
1626 		u32 gen;
1627 		unsigned long ttl;
1628 
1629 		gen = atomic_read(&session->s_cap_gen);
1630 		ttl = session->s_cap_ttl;
1631 
1632 		if (di->lease_gen == gen &&
1633 		    time_before(jiffies, ttl) &&
1634 		    time_before(jiffies, di->time))
1635 			return true;
1636 	}
1637 	di->lease_gen = 0;
1638 	return false;
1639 }
1640 
dentry_lease_is_valid(struct dentry * dentry,unsigned int flags)1641 static int dentry_lease_is_valid(struct dentry *dentry, unsigned int flags)
1642 {
1643 	struct ceph_dentry_info *di;
1644 	struct ceph_mds_session *session = NULL;
1645 	u32 seq = 0;
1646 	int valid = 0;
1647 
1648 	spin_lock(&dentry->d_lock);
1649 	di = ceph_dentry(dentry);
1650 	if (di && __dentry_lease_is_valid(di)) {
1651 		valid = 1;
1652 
1653 		if (di->lease_renew_after &&
1654 		    time_after(jiffies, di->lease_renew_after)) {
1655 			/*
1656 			 * We should renew. If we're in RCU walk mode
1657 			 * though, we can't do that so just return
1658 			 * -ECHILD.
1659 			 */
1660 			if (flags & LOOKUP_RCU) {
1661 				valid = -ECHILD;
1662 			} else {
1663 				session = ceph_get_mds_session(di->lease_session);
1664 				seq = di->lease_seq;
1665 				di->lease_renew_after = 0;
1666 				di->lease_renew_from = jiffies;
1667 			}
1668 		}
1669 	}
1670 	spin_unlock(&dentry->d_lock);
1671 
1672 	if (session) {
1673 		ceph_mdsc_lease_send_msg(session, dentry,
1674 					 CEPH_MDS_LEASE_RENEW, seq);
1675 		ceph_put_mds_session(session);
1676 	}
1677 	dout("dentry_lease_is_valid - dentry %p = %d\n", dentry, valid);
1678 	return valid;
1679 }
1680 
1681 /*
1682  * Called under dentry->d_lock.
1683  */
__dir_lease_try_check(const struct dentry * dentry)1684 static int __dir_lease_try_check(const struct dentry *dentry)
1685 {
1686 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1687 	struct inode *dir;
1688 	struct ceph_inode_info *ci;
1689 	int valid = 0;
1690 
1691 	if (!di->lease_shared_gen)
1692 		return 0;
1693 	if (IS_ROOT(dentry))
1694 		return 0;
1695 
1696 	dir = d_inode(dentry->d_parent);
1697 	ci = ceph_inode(dir);
1698 
1699 	if (spin_trylock(&ci->i_ceph_lock)) {
1700 		if (atomic_read(&ci->i_shared_gen) == di->lease_shared_gen &&
1701 		    __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 0))
1702 			valid = 1;
1703 		spin_unlock(&ci->i_ceph_lock);
1704 	} else {
1705 		valid = -EBUSY;
1706 	}
1707 
1708 	if (!valid)
1709 		di->lease_shared_gen = 0;
1710 	return valid;
1711 }
1712 
1713 /*
1714  * Check if directory-wide content lease/cap is valid.
1715  */
dir_lease_is_valid(struct inode * dir,struct dentry * dentry,struct ceph_mds_client * mdsc)1716 static int dir_lease_is_valid(struct inode *dir, struct dentry *dentry,
1717 			      struct ceph_mds_client *mdsc)
1718 {
1719 	struct ceph_inode_info *ci = ceph_inode(dir);
1720 	int valid;
1721 	int shared_gen;
1722 
1723 	spin_lock(&ci->i_ceph_lock);
1724 	valid = __ceph_caps_issued_mask(ci, CEPH_CAP_FILE_SHARED, 1);
1725 	if (valid) {
1726 		__ceph_touch_fmode(ci, mdsc, CEPH_FILE_MODE_RD);
1727 		shared_gen = atomic_read(&ci->i_shared_gen);
1728 	}
1729 	spin_unlock(&ci->i_ceph_lock);
1730 	if (valid) {
1731 		struct ceph_dentry_info *di;
1732 		spin_lock(&dentry->d_lock);
1733 		di = ceph_dentry(dentry);
1734 		if (dir == d_inode(dentry->d_parent) &&
1735 		    di && di->lease_shared_gen == shared_gen)
1736 			__ceph_dentry_dir_lease_touch(di);
1737 		else
1738 			valid = 0;
1739 		spin_unlock(&dentry->d_lock);
1740 	}
1741 	dout("dir_lease_is_valid dir %p v%u dentry %p = %d\n",
1742 	     dir, (unsigned)atomic_read(&ci->i_shared_gen), dentry, valid);
1743 	return valid;
1744 }
1745 
1746 /*
1747  * Check if cached dentry can be trusted.
1748  */
ceph_d_revalidate(struct dentry * dentry,unsigned int flags)1749 static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
1750 {
1751 	int valid = 0;
1752 	struct dentry *parent;
1753 	struct inode *dir, *inode;
1754 	struct ceph_mds_client *mdsc;
1755 
1756 	if (flags & LOOKUP_RCU) {
1757 		parent = READ_ONCE(dentry->d_parent);
1758 		dir = d_inode_rcu(parent);
1759 		if (!dir)
1760 			return -ECHILD;
1761 		inode = d_inode_rcu(dentry);
1762 	} else {
1763 		parent = dget_parent(dentry);
1764 		dir = d_inode(parent);
1765 		inode = d_inode(dentry);
1766 	}
1767 
1768 	dout("d_revalidate %p '%pd' inode %p offset 0x%llx\n", dentry,
1769 	     dentry, inode, ceph_dentry(dentry)->offset);
1770 
1771 	mdsc = ceph_sb_to_client(dir->i_sb)->mdsc;
1772 
1773 	/* always trust cached snapped dentries, snapdir dentry */
1774 	if (ceph_snap(dir) != CEPH_NOSNAP) {
1775 		dout("d_revalidate %p '%pd' inode %p is SNAPPED\n", dentry,
1776 		     dentry, inode);
1777 		valid = 1;
1778 	} else if (inode && ceph_snap(inode) == CEPH_SNAPDIR) {
1779 		valid = 1;
1780 	} else {
1781 		valid = dentry_lease_is_valid(dentry, flags);
1782 		if (valid == -ECHILD)
1783 			return valid;
1784 		if (valid || dir_lease_is_valid(dir, dentry, mdsc)) {
1785 			if (inode)
1786 				valid = ceph_is_any_caps(inode);
1787 			else
1788 				valid = 1;
1789 		}
1790 	}
1791 
1792 	if (!valid) {
1793 		struct ceph_mds_request *req;
1794 		int op, err;
1795 		u32 mask;
1796 
1797 		if (flags & LOOKUP_RCU)
1798 			return -ECHILD;
1799 
1800 		percpu_counter_inc(&mdsc->metric.d_lease_mis);
1801 
1802 		op = ceph_snap(dir) == CEPH_SNAPDIR ?
1803 			CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP;
1804 		req = ceph_mdsc_create_request(mdsc, op, USE_ANY_MDS);
1805 		if (!IS_ERR(req)) {
1806 			req->r_dentry = dget(dentry);
1807 			req->r_num_caps = 2;
1808 			req->r_parent = dir;
1809 			ihold(dir);
1810 
1811 			mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED;
1812 			if (ceph_security_xattr_wanted(dir))
1813 				mask |= CEPH_CAP_XATTR_SHARED;
1814 			req->r_args.getattr.mask = cpu_to_le32(mask);
1815 
1816 			err = ceph_mdsc_do_request(mdsc, NULL, req);
1817 			switch (err) {
1818 			case 0:
1819 				if (d_really_is_positive(dentry) &&
1820 				    d_inode(dentry) == req->r_target_inode)
1821 					valid = 1;
1822 				break;
1823 			case -ENOENT:
1824 				if (d_really_is_negative(dentry))
1825 					valid = 1;
1826 				fallthrough;
1827 			default:
1828 				break;
1829 			}
1830 			ceph_mdsc_put_request(req);
1831 			dout("d_revalidate %p lookup result=%d\n",
1832 			     dentry, err);
1833 		}
1834 	} else {
1835 		percpu_counter_inc(&mdsc->metric.d_lease_hit);
1836 	}
1837 
1838 	dout("d_revalidate %p %s\n", dentry, valid ? "valid" : "invalid");
1839 	if (!valid)
1840 		ceph_dir_clear_complete(dir);
1841 
1842 	if (!(flags & LOOKUP_RCU))
1843 		dput(parent);
1844 	return valid;
1845 }
1846 
1847 /*
1848  * Delete unused dentry that doesn't have valid lease
1849  *
1850  * Called under dentry->d_lock.
1851  */
ceph_d_delete(const struct dentry * dentry)1852 static int ceph_d_delete(const struct dentry *dentry)
1853 {
1854 	struct ceph_dentry_info *di;
1855 
1856 	/* won't release caps */
1857 	if (d_really_is_negative(dentry))
1858 		return 0;
1859 	if (ceph_snap(d_inode(dentry)) != CEPH_NOSNAP)
1860 		return 0;
1861 	/* vaild lease? */
1862 	di = ceph_dentry(dentry);
1863 	if (di) {
1864 		if (__dentry_lease_is_valid(di))
1865 			return 0;
1866 		if (__dir_lease_try_check(dentry))
1867 			return 0;
1868 	}
1869 	return 1;
1870 }
1871 
1872 /*
1873  * Release our ceph_dentry_info.
1874  */
ceph_d_release(struct dentry * dentry)1875 static void ceph_d_release(struct dentry *dentry)
1876 {
1877 	struct ceph_dentry_info *di = ceph_dentry(dentry);
1878 	struct ceph_fs_client *fsc = ceph_sb_to_client(dentry->d_sb);
1879 
1880 	dout("d_release %p\n", dentry);
1881 
1882 	atomic64_dec(&fsc->mdsc->metric.total_dentries);
1883 
1884 	spin_lock(&dentry->d_lock);
1885 	__dentry_lease_unlist(di);
1886 	dentry->d_fsdata = NULL;
1887 	spin_unlock(&dentry->d_lock);
1888 
1889 	ceph_put_mds_session(di->lease_session);
1890 	kmem_cache_free(ceph_dentry_cachep, di);
1891 }
1892 
1893 /*
1894  * When the VFS prunes a dentry from the cache, we need to clear the
1895  * complete flag on the parent directory.
1896  *
1897  * Called under dentry->d_lock.
1898  */
ceph_d_prune(struct dentry * dentry)1899 static void ceph_d_prune(struct dentry *dentry)
1900 {
1901 	struct ceph_inode_info *dir_ci;
1902 	struct ceph_dentry_info *di;
1903 
1904 	dout("ceph_d_prune %pd %p\n", dentry, dentry);
1905 
1906 	/* do we have a valid parent? */
1907 	if (IS_ROOT(dentry))
1908 		return;
1909 
1910 	/* we hold d_lock, so d_parent is stable */
1911 	dir_ci = ceph_inode(d_inode(dentry->d_parent));
1912 	if (dir_ci->i_vino.snap == CEPH_SNAPDIR)
1913 		return;
1914 
1915 	/* who calls d_delete() should also disable dcache readdir */
1916 	if (d_really_is_negative(dentry))
1917 		return;
1918 
1919 	/* d_fsdata does not get cleared until d_release */
1920 	if (!d_unhashed(dentry)) {
1921 		__ceph_dir_clear_complete(dir_ci);
1922 		return;
1923 	}
1924 
1925 	/* Disable dcache readdir just in case that someone called d_drop()
1926 	 * or d_invalidate(), but MDS didn't revoke CEPH_CAP_FILE_SHARED
1927 	 * properly (dcache readdir is still enabled) */
1928 	di = ceph_dentry(dentry);
1929 	if (di->offset > 0 &&
1930 	    di->lease_shared_gen == atomic_read(&dir_ci->i_shared_gen))
1931 		__ceph_dir_clear_ordered(dir_ci);
1932 }
1933 
1934 /*
1935  * read() on a dir.  This weird interface hack only works if mounted
1936  * with '-o dirstat'.
1937  */
ceph_read_dir(struct file * file,char __user * buf,size_t size,loff_t * ppos)1938 static ssize_t ceph_read_dir(struct file *file, char __user *buf, size_t size,
1939 			     loff_t *ppos)
1940 {
1941 	struct ceph_dir_file_info *dfi = file->private_data;
1942 	struct inode *inode = file_inode(file);
1943 	struct ceph_inode_info *ci = ceph_inode(inode);
1944 	int left;
1945 	const int bufsize = 1024;
1946 
1947 	if (!ceph_test_mount_opt(ceph_sb_to_client(inode->i_sb), DIRSTAT))
1948 		return -EISDIR;
1949 
1950 	if (!dfi->dir_info) {
1951 		dfi->dir_info = kmalloc(bufsize, GFP_KERNEL);
1952 		if (!dfi->dir_info)
1953 			return -ENOMEM;
1954 		dfi->dir_info_len =
1955 			snprintf(dfi->dir_info, bufsize,
1956 				"entries:   %20lld\n"
1957 				" files:    %20lld\n"
1958 				" subdirs:  %20lld\n"
1959 				"rentries:  %20lld\n"
1960 				" rfiles:   %20lld\n"
1961 				" rsubdirs: %20lld\n"
1962 				"rbytes:    %20lld\n"
1963 				"rctime:    %10lld.%09ld\n",
1964 				ci->i_files + ci->i_subdirs,
1965 				ci->i_files,
1966 				ci->i_subdirs,
1967 				ci->i_rfiles + ci->i_rsubdirs,
1968 				ci->i_rfiles,
1969 				ci->i_rsubdirs,
1970 				ci->i_rbytes,
1971 				ci->i_rctime.tv_sec,
1972 				ci->i_rctime.tv_nsec);
1973 	}
1974 
1975 	if (*ppos >= dfi->dir_info_len)
1976 		return 0;
1977 	size = min_t(unsigned, size, dfi->dir_info_len-*ppos);
1978 	left = copy_to_user(buf, dfi->dir_info + *ppos, size);
1979 	if (left == size)
1980 		return -EFAULT;
1981 	*ppos += (size - left);
1982 	return size - left;
1983 }
1984 
1985 
1986 
1987 /*
1988  * Return name hash for a given dentry.  This is dependent on
1989  * the parent directory's hash function.
1990  */
ceph_dentry_hash(struct inode * dir,struct dentry * dn)1991 unsigned ceph_dentry_hash(struct inode *dir, struct dentry *dn)
1992 {
1993 	struct ceph_inode_info *dci = ceph_inode(dir);
1994 	unsigned hash;
1995 
1996 	switch (dci->i_dir_layout.dl_dir_hash) {
1997 	case 0:	/* for backward compat */
1998 	case CEPH_STR_HASH_LINUX:
1999 		return dn->d_name.hash;
2000 
2001 	default:
2002 		spin_lock(&dn->d_lock);
2003 		hash = ceph_str_hash(dci->i_dir_layout.dl_dir_hash,
2004 				     dn->d_name.name, dn->d_name.len);
2005 		spin_unlock(&dn->d_lock);
2006 		return hash;
2007 	}
2008 }
2009 
2010 const struct file_operations ceph_dir_fops = {
2011 	.read = ceph_read_dir,
2012 	.iterate = ceph_readdir,
2013 	.llseek = ceph_dir_llseek,
2014 	.open = ceph_open,
2015 	.release = ceph_release,
2016 	.unlocked_ioctl = ceph_ioctl,
2017 	.compat_ioctl = compat_ptr_ioctl,
2018 	.fsync = ceph_fsync,
2019 	.lock = ceph_lock,
2020 	.flock = ceph_flock,
2021 };
2022 
2023 const struct file_operations ceph_snapdir_fops = {
2024 	.iterate = ceph_readdir,
2025 	.llseek = ceph_dir_llseek,
2026 	.open = ceph_open,
2027 	.release = ceph_release,
2028 };
2029 
2030 const struct inode_operations ceph_dir_iops = {
2031 	.lookup = ceph_lookup,
2032 	.permission = ceph_permission,
2033 	.getattr = ceph_getattr,
2034 	.setattr = ceph_setattr,
2035 	.listxattr = ceph_listxattr,
2036 	.get_acl = ceph_get_acl,
2037 	.set_acl = ceph_set_acl,
2038 	.mknod = ceph_mknod,
2039 	.symlink = ceph_symlink,
2040 	.mkdir = ceph_mkdir,
2041 	.link = ceph_link,
2042 	.unlink = ceph_unlink,
2043 	.rmdir = ceph_unlink,
2044 	.rename = ceph_rename,
2045 	.create = ceph_create,
2046 	.atomic_open = ceph_atomic_open,
2047 };
2048 
2049 const struct inode_operations ceph_snapdir_iops = {
2050 	.lookup = ceph_lookup,
2051 	.permission = ceph_permission,
2052 	.getattr = ceph_getattr,
2053 	.mkdir = ceph_mkdir,
2054 	.rmdir = ceph_unlink,
2055 	.rename = ceph_rename,
2056 };
2057 
2058 const struct dentry_operations ceph_dentry_ops = {
2059 	.d_revalidate = ceph_d_revalidate,
2060 	.d_delete = ceph_d_delete,
2061 	.d_release = ceph_d_release,
2062 	.d_prune = ceph_d_prune,
2063 	.d_init = ceph_d_init,
2064 };
2065