xref: /linux/fs/ceph/export.c (revision 987b741c52c7c6c68d46fbaeb95b8d1087f10b7f)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/exportfs.h>
5 #include <linux/slab.h>
6 #include <asm/unaligned.h>
7 
8 #include "super.h"
9 #include "mds_client.h"
10 
11 /*
12  * Basic fh
13  */
14 struct ceph_nfs_fh {
15 	u64 ino;
16 } __attribute__ ((packed));
17 
18 /*
19  * Larger fh that includes parent ino.
20  */
21 struct ceph_nfs_confh {
22 	u64 ino, parent_ino;
23 } __attribute__ ((packed));
24 
25 /*
26  * fh for snapped inode
27  */
28 struct ceph_nfs_snapfh {
29 	u64 ino;
30 	u64 snapid;
31 	u64 parent_ino;
32 	u32 hash;
33 } __attribute__ ((packed));
34 
35 static int ceph_encode_snapfh(struct inode *inode, u32 *rawfh, int *max_len,
36 			      struct inode *parent_inode)
37 {
38 	static const int snap_handle_length =
39 		sizeof(struct ceph_nfs_snapfh) >> 2;
40 	struct ceph_nfs_snapfh *sfh = (void *)rawfh;
41 	u64 snapid = ceph_snap(inode);
42 	int ret;
43 	bool no_parent = true;
44 
45 	if (*max_len < snap_handle_length) {
46 		*max_len = snap_handle_length;
47 		ret = FILEID_INVALID;
48 		goto out;
49 	}
50 
51 	ret =  -EINVAL;
52 	if (snapid != CEPH_SNAPDIR) {
53 		struct inode *dir;
54 		struct dentry *dentry = d_find_alias(inode);
55 		if (!dentry)
56 			goto out;
57 
58 		rcu_read_lock();
59 		dir = d_inode_rcu(dentry->d_parent);
60 		if (ceph_snap(dir) != CEPH_SNAPDIR) {
61 			sfh->parent_ino = ceph_ino(dir);
62 			sfh->hash = ceph_dentry_hash(dir, dentry);
63 			no_parent = false;
64 		}
65 		rcu_read_unlock();
66 		dput(dentry);
67 	}
68 
69 	if (no_parent) {
70 		if (!S_ISDIR(inode->i_mode))
71 			goto out;
72 		sfh->parent_ino = sfh->ino;
73 		sfh->hash = 0;
74 	}
75 	sfh->ino = ceph_ino(inode);
76 	sfh->snapid = snapid;
77 
78 	*max_len = snap_handle_length;
79 	ret = FILEID_BTRFS_WITH_PARENT;
80 out:
81 	dout("encode_snapfh %llx.%llx ret=%d\n", ceph_vinop(inode), ret);
82 	return ret;
83 }
84 
85 static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
86 			  struct inode *parent_inode)
87 {
88 	static const int handle_length =
89 		sizeof(struct ceph_nfs_fh) >> 2;
90 	static const int connected_handle_length =
91 		sizeof(struct ceph_nfs_confh) >> 2;
92 	int type;
93 
94 	if (ceph_snap(inode) != CEPH_NOSNAP)
95 		return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
96 
97 	if (parent_inode && (*max_len < connected_handle_length)) {
98 		*max_len = connected_handle_length;
99 		return FILEID_INVALID;
100 	} else if (*max_len < handle_length) {
101 		*max_len = handle_length;
102 		return FILEID_INVALID;
103 	}
104 
105 	if (parent_inode) {
106 		struct ceph_nfs_confh *cfh = (void *)rawfh;
107 		dout("encode_fh %llx with parent %llx\n",
108 		     ceph_ino(inode), ceph_ino(parent_inode));
109 		cfh->ino = ceph_ino(inode);
110 		cfh->parent_ino = ceph_ino(parent_inode);
111 		*max_len = connected_handle_length;
112 		type = FILEID_INO32_GEN_PARENT;
113 	} else {
114 		struct ceph_nfs_fh *fh = (void *)rawfh;
115 		dout("encode_fh %llx\n", ceph_ino(inode));
116 		fh->ino = ceph_ino(inode);
117 		*max_len = handle_length;
118 		type = FILEID_INO32_GEN;
119 	}
120 	return type;
121 }
122 
123 static struct inode *__lookup_inode(struct super_block *sb, u64 ino)
124 {
125 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
126 	struct inode *inode;
127 	struct ceph_vino vino;
128 	int err;
129 
130 	vino.ino = ino;
131 	vino.snap = CEPH_NOSNAP;
132 
133 	if (ceph_vino_is_reserved(vino))
134 		return ERR_PTR(-ESTALE);
135 
136 	inode = ceph_find_inode(sb, vino);
137 	if (!inode) {
138 		struct ceph_mds_request *req;
139 		int mask;
140 
141 		req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
142 					       USE_ANY_MDS);
143 		if (IS_ERR(req))
144 			return ERR_CAST(req);
145 
146 		mask = CEPH_STAT_CAP_INODE;
147 		if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
148 			mask |= CEPH_CAP_XATTR_SHARED;
149 		req->r_args.lookupino.mask = cpu_to_le32(mask);
150 
151 		req->r_ino1 = vino;
152 		req->r_num_caps = 1;
153 		err = ceph_mdsc_do_request(mdsc, NULL, req);
154 		inode = req->r_target_inode;
155 		if (inode)
156 			ihold(inode);
157 		ceph_mdsc_put_request(req);
158 		if (!inode)
159 			return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
160 	}
161 	return inode;
162 }
163 
164 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
165 {
166 	struct inode *inode = __lookup_inode(sb, ino);
167 	if (IS_ERR(inode))
168 		return inode;
169 	if (inode->i_nlink == 0) {
170 		iput(inode);
171 		return ERR_PTR(-ESTALE);
172 	}
173 	return inode;
174 }
175 
176 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
177 {
178 	struct inode *inode = __lookup_inode(sb, ino);
179 	int err;
180 
181 	if (IS_ERR(inode))
182 		return ERR_CAST(inode);
183 	/* We need LINK caps to reliably check i_nlink */
184 	err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
185 	if (err) {
186 		iput(inode);
187 		return ERR_PTR(err);
188 	}
189 	/* -ESTALE if inode as been unlinked and no file is open */
190 	if ((inode->i_nlink == 0) && (atomic_read(&inode->i_count) == 1)) {
191 		iput(inode);
192 		return ERR_PTR(-ESTALE);
193 	}
194 	return d_obtain_alias(inode);
195 }
196 
197 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
198 					  struct ceph_nfs_snapfh *sfh,
199 					  bool want_parent)
200 {
201 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
202 	struct ceph_mds_request *req;
203 	struct inode *inode;
204 	struct ceph_vino vino;
205 	int mask;
206 	int err;
207 	bool unlinked = false;
208 
209 	if (want_parent) {
210 		vino.ino = sfh->parent_ino;
211 		if (sfh->snapid == CEPH_SNAPDIR)
212 			vino.snap = CEPH_NOSNAP;
213 		else if (sfh->ino == sfh->parent_ino)
214 			vino.snap = CEPH_SNAPDIR;
215 		else
216 			vino.snap = sfh->snapid;
217 	} else {
218 		vino.ino = sfh->ino;
219 		vino.snap = sfh->snapid;
220 	}
221 
222 	if (ceph_vino_is_reserved(vino))
223 		return ERR_PTR(-ESTALE);
224 
225 	inode = ceph_find_inode(sb, vino);
226 	if (inode)
227 		return d_obtain_alias(inode);
228 
229 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
230 				       USE_ANY_MDS);
231 	if (IS_ERR(req))
232 		return ERR_CAST(req);
233 
234 	mask = CEPH_STAT_CAP_INODE;
235 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
236 		mask |= CEPH_CAP_XATTR_SHARED;
237 	req->r_args.lookupino.mask = cpu_to_le32(mask);
238 	if (vino.snap < CEPH_NOSNAP) {
239 		req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
240 		if (!want_parent && sfh->ino != sfh->parent_ino) {
241 			req->r_args.lookupino.parent =
242 					cpu_to_le64(sfh->parent_ino);
243 			req->r_args.lookupino.hash =
244 					cpu_to_le32(sfh->hash);
245 		}
246 	}
247 
248 	req->r_ino1 = vino;
249 	req->r_num_caps = 1;
250 	err = ceph_mdsc_do_request(mdsc, NULL, req);
251 	inode = req->r_target_inode;
252 	if (inode) {
253 		if (vino.snap == CEPH_SNAPDIR) {
254 			if (inode->i_nlink == 0)
255 				unlinked = true;
256 			inode = ceph_get_snapdir(inode);
257 		} else if (ceph_snap(inode) == vino.snap) {
258 			ihold(inode);
259 		} else {
260 			/* mds does not support lookup snapped inode */
261 			inode = ERR_PTR(-EOPNOTSUPP);
262 		}
263 	} else {
264 		inode = ERR_PTR(-ESTALE);
265 	}
266 	ceph_mdsc_put_request(req);
267 
268 	if (want_parent) {
269 		dout("snapfh_to_parent %llx.%llx\n err=%d\n",
270 		     vino.ino, vino.snap, err);
271 	} else {
272 		dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
273 		      vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
274 	}
275 	if (IS_ERR(inode))
276 		return ERR_CAST(inode);
277 	/* see comments in ceph_get_parent() */
278 	return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
279 }
280 
281 /*
282  * convert regular fh to dentry
283  */
284 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
285 					struct fid *fid,
286 					int fh_len, int fh_type)
287 {
288 	struct ceph_nfs_fh *fh = (void *)fid->raw;
289 
290 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
291 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
292 		return __snapfh_to_dentry(sb, sfh, false);
293 	}
294 
295 	if (fh_type != FILEID_INO32_GEN  &&
296 	    fh_type != FILEID_INO32_GEN_PARENT)
297 		return NULL;
298 	if (fh_len < sizeof(*fh) / 4)
299 		return NULL;
300 
301 	dout("fh_to_dentry %llx\n", fh->ino);
302 	return __fh_to_dentry(sb, fh->ino);
303 }
304 
305 static struct dentry *__get_parent(struct super_block *sb,
306 				   struct dentry *child, u64 ino)
307 {
308 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
309 	struct ceph_mds_request *req;
310 	struct inode *inode;
311 	int mask;
312 	int err;
313 
314 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
315 				       USE_ANY_MDS);
316 	if (IS_ERR(req))
317 		return ERR_CAST(req);
318 
319 	if (child) {
320 		req->r_inode = d_inode(child);
321 		ihold(d_inode(child));
322 	} else {
323 		req->r_ino1 = (struct ceph_vino) {
324 			.ino = ino,
325 			.snap = CEPH_NOSNAP,
326 		};
327 	}
328 
329 	mask = CEPH_STAT_CAP_INODE;
330 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
331 		mask |= CEPH_CAP_XATTR_SHARED;
332 	req->r_args.getattr.mask = cpu_to_le32(mask);
333 
334 	req->r_num_caps = 1;
335 	err = ceph_mdsc_do_request(mdsc, NULL, req);
336 	if (err) {
337 		ceph_mdsc_put_request(req);
338 		return ERR_PTR(err);
339 	}
340 
341 	inode = req->r_target_inode;
342 	if (inode)
343 		ihold(inode);
344 	ceph_mdsc_put_request(req);
345 	if (!inode)
346 		return ERR_PTR(-ENOENT);
347 
348 	return d_obtain_alias(inode);
349 }
350 
351 static struct dentry *ceph_get_parent(struct dentry *child)
352 {
353 	struct inode *inode = d_inode(child);
354 	struct dentry *dn;
355 
356 	if (ceph_snap(inode) != CEPH_NOSNAP) {
357 		struct inode* dir;
358 		bool unlinked = false;
359 		/* do not support non-directory */
360 		if (!d_is_dir(child)) {
361 			dn = ERR_PTR(-EINVAL);
362 			goto out;
363 		}
364 		dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
365 		if (IS_ERR(dir)) {
366 			dn = ERR_CAST(dir);
367 			goto out;
368 		}
369 		/* There can be multiple paths to access snapped inode.
370 		 * For simplicity, treat snapdir of head inode as parent */
371 		if (ceph_snap(inode) != CEPH_SNAPDIR) {
372 			struct inode *snapdir = ceph_get_snapdir(dir);
373 			if (dir->i_nlink == 0)
374 				unlinked = true;
375 			iput(dir);
376 			if (IS_ERR(snapdir)) {
377 				dn = ERR_CAST(snapdir);
378 				goto out;
379 			}
380 			dir = snapdir;
381 		}
382 		/* If directory has already been deleted, futher get_parent
383 		 * will fail. Do not mark snapdir dentry as disconnected,
384 		 * this prevent exportfs from doing futher get_parent. */
385 		if (unlinked)
386 			dn = d_obtain_root(dir);
387 		else
388 			dn = d_obtain_alias(dir);
389 	} else {
390 		dn = __get_parent(child->d_sb, child, 0);
391 	}
392 out:
393 	dout("get_parent %p ino %llx.%llx err=%ld\n",
394 	     child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
395 	return dn;
396 }
397 
398 /*
399  * convert regular fh to parent
400  */
401 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
402 					struct fid *fid,
403 					int fh_len, int fh_type)
404 {
405 	struct ceph_nfs_confh *cfh = (void *)fid->raw;
406 	struct dentry *dentry;
407 
408 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
409 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
410 		return __snapfh_to_dentry(sb, sfh, true);
411 	}
412 
413 	if (fh_type != FILEID_INO32_GEN_PARENT)
414 		return NULL;
415 	if (fh_len < sizeof(*cfh) / 4)
416 		return NULL;
417 
418 	dout("fh_to_parent %llx\n", cfh->parent_ino);
419 	dentry = __get_parent(sb, NULL, cfh->ino);
420 	if (unlikely(dentry == ERR_PTR(-ENOENT)))
421 		dentry = __fh_to_dentry(sb, cfh->parent_ino);
422 	return dentry;
423 }
424 
425 static int __get_snap_name(struct dentry *parent, char *name,
426 			   struct dentry *child)
427 {
428 	struct inode *inode = d_inode(child);
429 	struct inode *dir = d_inode(parent);
430 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
431 	struct ceph_mds_request *req = NULL;
432 	char *last_name = NULL;
433 	unsigned next_offset = 2;
434 	int err = -EINVAL;
435 
436 	if (ceph_ino(inode) != ceph_ino(dir))
437 		goto out;
438 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
439 		if (ceph_snap(dir) == CEPH_NOSNAP) {
440 			strcpy(name, fsc->mount_options->snapdir_name);
441 			err = 0;
442 		}
443 		goto out;
444 	}
445 	if (ceph_snap(dir) != CEPH_SNAPDIR)
446 		goto out;
447 
448 	while (1) {
449 		struct ceph_mds_reply_info_parsed *rinfo;
450 		struct ceph_mds_reply_dir_entry *rde;
451 		int i;
452 
453 		req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
454 					       USE_AUTH_MDS);
455 		if (IS_ERR(req)) {
456 			err = PTR_ERR(req);
457 			req = NULL;
458 			goto out;
459 		}
460 		err = ceph_alloc_readdir_reply_buffer(req, inode);
461 		if (err)
462 			goto out;
463 
464 		req->r_direct_mode = USE_AUTH_MDS;
465 		req->r_readdir_offset = next_offset;
466 		req->r_args.readdir.flags =
467 				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
468 		if (last_name) {
469 			req->r_path2 = last_name;
470 			last_name = NULL;
471 		}
472 
473 		req->r_inode = dir;
474 		ihold(dir);
475 		req->r_dentry = dget(parent);
476 
477 		inode_lock(dir);
478 		err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
479 		inode_unlock(dir);
480 
481 		if (err < 0)
482 			goto out;
483 
484 		rinfo = &req->r_reply_info;
485 		for (i = 0; i < rinfo->dir_nr; i++) {
486 			rde = rinfo->dir_entries + i;
487 			BUG_ON(!rde->inode.in);
488 			if (ceph_snap(inode) ==
489 			    le64_to_cpu(rde->inode.in->snapid)) {
490 				memcpy(name, rde->name, rde->name_len);
491 				name[rde->name_len] = '\0';
492 				err = 0;
493 				goto out;
494 			}
495 		}
496 
497 		if (rinfo->dir_end)
498 			break;
499 
500 		BUG_ON(rinfo->dir_nr <= 0);
501 		rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
502 		next_offset += rinfo->dir_nr;
503 		last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
504 		if (!last_name) {
505 			err = -ENOMEM;
506 			goto out;
507 		}
508 
509 		ceph_mdsc_put_request(req);
510 		req = NULL;
511 	}
512 	err = -ENOENT;
513 out:
514 	if (req)
515 		ceph_mdsc_put_request(req);
516 	kfree(last_name);
517 	dout("get_snap_name %p ino %llx.%llx err=%d\n",
518 	     child, ceph_vinop(inode), err);
519 	return err;
520 }
521 
522 static int ceph_get_name(struct dentry *parent, char *name,
523 			 struct dentry *child)
524 {
525 	struct ceph_mds_client *mdsc;
526 	struct ceph_mds_request *req;
527 	struct inode *inode = d_inode(child);
528 	int err;
529 
530 	if (ceph_snap(inode) != CEPH_NOSNAP)
531 		return __get_snap_name(parent, name, child);
532 
533 	mdsc = ceph_inode_to_client(inode)->mdsc;
534 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
535 				       USE_ANY_MDS);
536 	if (IS_ERR(req))
537 		return PTR_ERR(req);
538 
539 	inode_lock(d_inode(parent));
540 
541 	req->r_inode = inode;
542 	ihold(inode);
543 	req->r_ino2 = ceph_vino(d_inode(parent));
544 	req->r_parent = d_inode(parent);
545 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
546 	req->r_num_caps = 2;
547 	err = ceph_mdsc_do_request(mdsc, NULL, req);
548 
549 	inode_unlock(d_inode(parent));
550 
551 	if (!err) {
552 		struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
553 		memcpy(name, rinfo->dname, rinfo->dname_len);
554 		name[rinfo->dname_len] = 0;
555 		dout("get_name %p ino %llx.%llx name %s\n",
556 		     child, ceph_vinop(inode), name);
557 	} else {
558 		dout("get_name %p ino %llx.%llx err %d\n",
559 		     child, ceph_vinop(inode), err);
560 	}
561 
562 	ceph_mdsc_put_request(req);
563 	return err;
564 }
565 
566 const struct export_operations ceph_export_ops = {
567 	.encode_fh = ceph_encode_fh,
568 	.fh_to_dentry = ceph_fh_to_dentry,
569 	.fh_to_parent = ceph_fh_to_parent,
570 	.get_parent = ceph_get_parent,
571 	.get_name = ceph_get_name,
572 };
573