xref: /linux/fs/ceph/export.c (revision 6c7353836a91b1479e6b81791cdc163fb04b4834)
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 #include "crypto.h"
11 
12 /*
13  * Basic fh
14  */
15 struct ceph_nfs_fh {
16 	u64 ino;
17 } __attribute__ ((packed));
18 
19 /*
20  * Larger fh that includes parent ino.
21  */
22 struct ceph_nfs_confh {
23 	u64 ino, parent_ino;
24 } __attribute__ ((packed));
25 
26 /*
27  * fh for snapped inode
28  */
29 struct ceph_nfs_snapfh {
30 	u64 ino;
31 	u64 snapid;
32 	u64 parent_ino;
33 	u32 hash;
34 } __attribute__ ((packed));
35 
36 static int ceph_encode_snapfh(struct inode *inode, u32 *rawfh, int *max_len,
37 			      struct inode *parent_inode)
38 {
39 	struct ceph_client *cl = ceph_inode_to_client(inode);
40 	static const int snap_handle_length =
41 		sizeof(struct ceph_nfs_snapfh) >> 2;
42 	struct ceph_nfs_snapfh *sfh = (void *)rawfh;
43 	u64 snapid = ceph_snap(inode);
44 	int ret;
45 	bool no_parent = true;
46 
47 	if (*max_len < snap_handle_length) {
48 		*max_len = snap_handle_length;
49 		ret = FILEID_INVALID;
50 		goto out;
51 	}
52 
53 	ret =  -EINVAL;
54 	if (snapid != CEPH_SNAPDIR) {
55 		struct inode *dir;
56 		struct dentry *dentry = d_find_alias(inode);
57 		if (!dentry)
58 			goto out;
59 
60 		rcu_read_lock();
61 		dir = d_inode_rcu(dentry->d_parent);
62 		if (ceph_snap(dir) != CEPH_SNAPDIR) {
63 			sfh->parent_ino = ceph_ino(dir);
64 			sfh->hash = ceph_dentry_hash(dir, dentry);
65 			no_parent = false;
66 		}
67 		rcu_read_unlock();
68 		dput(dentry);
69 	}
70 
71 	if (no_parent) {
72 		if (!S_ISDIR(inode->i_mode))
73 			goto out;
74 		sfh->parent_ino = sfh->ino;
75 		sfh->hash = 0;
76 	}
77 	sfh->ino = ceph_ino(inode);
78 	sfh->snapid = snapid;
79 
80 	*max_len = snap_handle_length;
81 	ret = FILEID_BTRFS_WITH_PARENT;
82 out:
83 	doutc(cl, "%p %llx.%llx ret=%d\n", inode, ceph_vinop(inode), ret);
84 	return ret;
85 }
86 
87 static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
88 			  struct inode *parent_inode)
89 {
90 	struct ceph_client *cl = ceph_inode_to_client(inode);
91 	static const int handle_length =
92 		sizeof(struct ceph_nfs_fh) >> 2;
93 	static const int connected_handle_length =
94 		sizeof(struct ceph_nfs_confh) >> 2;
95 	int type;
96 
97 	if (ceph_snap(inode) != CEPH_NOSNAP)
98 		return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
99 
100 	if (parent_inode && (*max_len < connected_handle_length)) {
101 		*max_len = connected_handle_length;
102 		return FILEID_INVALID;
103 	} else if (*max_len < handle_length) {
104 		*max_len = handle_length;
105 		return FILEID_INVALID;
106 	}
107 
108 	if (parent_inode) {
109 		struct ceph_nfs_confh *cfh = (void *)rawfh;
110 		doutc(cl, "%p %llx.%llx with parent %p %llx.%llx\n", inode,
111 		      ceph_vinop(inode), parent_inode, ceph_vinop(parent_inode));
112 		cfh->ino = ceph_ino(inode);
113 		cfh->parent_ino = ceph_ino(parent_inode);
114 		*max_len = connected_handle_length;
115 		type = FILEID_INO32_GEN_PARENT;
116 	} else {
117 		struct ceph_nfs_fh *fh = (void *)rawfh;
118 		doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
119 		fh->ino = ceph_ino(inode);
120 		*max_len = handle_length;
121 		type = FILEID_INO32_GEN;
122 	}
123 	return type;
124 }
125 
126 static struct inode *__lookup_inode(struct super_block *sb, u64 ino)
127 {
128 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
129 	struct inode *inode;
130 	struct ceph_vino vino;
131 	int err;
132 
133 	vino.ino = ino;
134 	vino.snap = CEPH_NOSNAP;
135 
136 	if (ceph_vino_is_reserved(vino))
137 		return ERR_PTR(-ESTALE);
138 
139 	inode = ceph_find_inode(sb, vino);
140 	if (!inode) {
141 		struct ceph_mds_request *req;
142 		int mask;
143 
144 		req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
145 					       USE_ANY_MDS);
146 		if (IS_ERR(req))
147 			return ERR_CAST(req);
148 
149 		mask = CEPH_STAT_CAP_INODE;
150 		if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
151 			mask |= CEPH_CAP_XATTR_SHARED;
152 		req->r_args.lookupino.mask = cpu_to_le32(mask);
153 
154 		req->r_ino1 = vino;
155 		req->r_num_caps = 1;
156 		err = ceph_mdsc_do_request(mdsc, NULL, req);
157 		inode = req->r_target_inode;
158 		if (inode)
159 			ihold(inode);
160 		ceph_mdsc_put_request(req);
161 		if (!inode)
162 			return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
163 	} else {
164 		if (ceph_inode_is_shutdown(inode)) {
165 			iput(inode);
166 			return ERR_PTR(-ESTALE);
167 		}
168 	}
169 	return inode;
170 }
171 
172 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
173 {
174 	struct inode *inode = __lookup_inode(sb, ino);
175 	if (IS_ERR(inode))
176 		return inode;
177 	if (inode->i_nlink == 0) {
178 		iput(inode);
179 		return ERR_PTR(-ESTALE);
180 	}
181 	return inode;
182 }
183 
184 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
185 {
186 	struct inode *inode = __lookup_inode(sb, ino);
187 	struct ceph_inode_info *ci = ceph_inode(inode);
188 	int err;
189 
190 	if (IS_ERR(inode))
191 		return ERR_CAST(inode);
192 	/* We need LINK caps to reliably check i_nlink */
193 	err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
194 	if (err) {
195 		iput(inode);
196 		return ERR_PTR(err);
197 	}
198 	/* -ESTALE if inode as been unlinked and no file is open */
199 	if ((inode->i_nlink == 0) && !__ceph_is_file_opened(ci)) {
200 		iput(inode);
201 		return ERR_PTR(-ESTALE);
202 	}
203 	return d_obtain_alias(inode);
204 }
205 
206 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
207 					  struct ceph_nfs_snapfh *sfh,
208 					  bool want_parent)
209 {
210 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
211 	struct ceph_client *cl = mdsc->fsc->client;
212 	struct ceph_mds_request *req;
213 	struct inode *inode;
214 	struct ceph_vino vino;
215 	int mask;
216 	int err;
217 	bool unlinked = false;
218 
219 	if (want_parent) {
220 		vino.ino = sfh->parent_ino;
221 		if (sfh->snapid == CEPH_SNAPDIR)
222 			vino.snap = CEPH_NOSNAP;
223 		else if (sfh->ino == sfh->parent_ino)
224 			vino.snap = CEPH_SNAPDIR;
225 		else
226 			vino.snap = sfh->snapid;
227 	} else {
228 		vino.ino = sfh->ino;
229 		vino.snap = sfh->snapid;
230 	}
231 
232 	if (ceph_vino_is_reserved(vino))
233 		return ERR_PTR(-ESTALE);
234 
235 	inode = ceph_find_inode(sb, vino);
236 	if (inode) {
237 		if (ceph_inode_is_shutdown(inode)) {
238 			iput(inode);
239 			return ERR_PTR(-ESTALE);
240 		}
241 		return d_obtain_alias(inode);
242 	}
243 
244 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
245 				       USE_ANY_MDS);
246 	if (IS_ERR(req))
247 		return ERR_CAST(req);
248 
249 	mask = CEPH_STAT_CAP_INODE;
250 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
251 		mask |= CEPH_CAP_XATTR_SHARED;
252 	req->r_args.lookupino.mask = cpu_to_le32(mask);
253 	if (vino.snap < CEPH_NOSNAP) {
254 		req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
255 		if (!want_parent && sfh->ino != sfh->parent_ino) {
256 			req->r_args.lookupino.parent =
257 					cpu_to_le64(sfh->parent_ino);
258 			req->r_args.lookupino.hash =
259 					cpu_to_le32(sfh->hash);
260 		}
261 	}
262 
263 	req->r_ino1 = vino;
264 	req->r_num_caps = 1;
265 	err = ceph_mdsc_do_request(mdsc, NULL, req);
266 	inode = req->r_target_inode;
267 	if (inode) {
268 		if (vino.snap == CEPH_SNAPDIR) {
269 			if (inode->i_nlink == 0)
270 				unlinked = true;
271 			inode = ceph_get_snapdir(inode);
272 		} else if (ceph_snap(inode) == vino.snap) {
273 			ihold(inode);
274 		} else {
275 			/* mds does not support lookup snapped inode */
276 			inode = ERR_PTR(-EOPNOTSUPP);
277 		}
278 	} else {
279 		inode = ERR_PTR(-ESTALE);
280 	}
281 	ceph_mdsc_put_request(req);
282 
283 	if (want_parent) {
284 		doutc(cl, "%llx.%llx\n err=%d\n", vino.ino, vino.snap, err);
285 	} else {
286 		doutc(cl, "%llx.%llx parent %llx hash %x err=%d", vino.ino,
287 		      vino.snap, sfh->parent_ino, sfh->hash, err);
288 	}
289 	/* see comments in ceph_get_parent() */
290 	return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
291 }
292 
293 /*
294  * convert regular fh to dentry
295  */
296 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
297 					struct fid *fid,
298 					int fh_len, int fh_type)
299 {
300 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
301 	struct ceph_nfs_fh *fh = (void *)fid->raw;
302 
303 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
304 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
305 		return __snapfh_to_dentry(sb, sfh, false);
306 	}
307 
308 	if (fh_type != FILEID_INO32_GEN  &&
309 	    fh_type != FILEID_INO32_GEN_PARENT)
310 		return NULL;
311 	if (fh_len < sizeof(*fh) / 4)
312 		return NULL;
313 
314 	doutc(fsc->client, "%llx\n", fh->ino);
315 	return __fh_to_dentry(sb, fh->ino);
316 }
317 
318 static struct dentry *__get_parent(struct super_block *sb,
319 				   struct dentry *child, u64 ino)
320 {
321 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
322 	struct ceph_mds_request *req;
323 	struct inode *inode;
324 	int mask;
325 	int err;
326 
327 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
328 				       USE_ANY_MDS);
329 	if (IS_ERR(req))
330 		return ERR_CAST(req);
331 
332 	if (child) {
333 		req->r_inode = d_inode(child);
334 		ihold(d_inode(child));
335 	} else {
336 		req->r_ino1 = (struct ceph_vino) {
337 			.ino = ino,
338 			.snap = CEPH_NOSNAP,
339 		};
340 	}
341 
342 	mask = CEPH_STAT_CAP_INODE;
343 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
344 		mask |= CEPH_CAP_XATTR_SHARED;
345 	req->r_args.getattr.mask = cpu_to_le32(mask);
346 
347 	req->r_num_caps = 1;
348 	err = ceph_mdsc_do_request(mdsc, NULL, req);
349 	if (err) {
350 		ceph_mdsc_put_request(req);
351 		return ERR_PTR(err);
352 	}
353 
354 	inode = req->r_target_inode;
355 	if (inode)
356 		ihold(inode);
357 	ceph_mdsc_put_request(req);
358 	if (!inode)
359 		return ERR_PTR(-ENOENT);
360 
361 	return d_obtain_alias(inode);
362 }
363 
364 static struct dentry *ceph_get_parent(struct dentry *child)
365 {
366 	struct inode *inode = d_inode(child);
367 	struct ceph_client *cl = ceph_inode_to_client(inode);
368 	struct dentry *dn;
369 
370 	if (ceph_snap(inode) != CEPH_NOSNAP) {
371 		struct inode* dir;
372 		bool unlinked = false;
373 		/* do not support non-directory */
374 		if (!d_is_dir(child)) {
375 			dn = ERR_PTR(-EINVAL);
376 			goto out;
377 		}
378 		dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
379 		if (IS_ERR(dir)) {
380 			dn = ERR_CAST(dir);
381 			goto out;
382 		}
383 		/* There can be multiple paths to access snapped inode.
384 		 * For simplicity, treat snapdir of head inode as parent */
385 		if (ceph_snap(inode) != CEPH_SNAPDIR) {
386 			struct inode *snapdir = ceph_get_snapdir(dir);
387 			if (dir->i_nlink == 0)
388 				unlinked = true;
389 			iput(dir);
390 			if (IS_ERR(snapdir)) {
391 				dn = ERR_CAST(snapdir);
392 				goto out;
393 			}
394 			dir = snapdir;
395 		}
396 		/* If directory has already been deleted, futher get_parent
397 		 * will fail. Do not mark snapdir dentry as disconnected,
398 		 * this prevent exportfs from doing futher get_parent. */
399 		if (unlinked)
400 			dn = d_obtain_root(dir);
401 		else
402 			dn = d_obtain_alias(dir);
403 	} else {
404 		dn = __get_parent(child->d_sb, child, 0);
405 	}
406 out:
407 	doutc(cl, "child %p %p %llx.%llx err=%ld\n", child, inode,
408 	      ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
409 	return dn;
410 }
411 
412 /*
413  * convert regular fh to parent
414  */
415 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
416 					struct fid *fid,
417 					int fh_len, int fh_type)
418 {
419 	struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
420 	struct ceph_nfs_confh *cfh = (void *)fid->raw;
421 	struct dentry *dentry;
422 
423 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
424 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
425 		return __snapfh_to_dentry(sb, sfh, true);
426 	}
427 
428 	if (fh_type != FILEID_INO32_GEN_PARENT)
429 		return NULL;
430 	if (fh_len < sizeof(*cfh) / 4)
431 		return NULL;
432 
433 	doutc(fsc->client, "%llx\n", cfh->parent_ino);
434 	dentry = __get_parent(sb, NULL, cfh->ino);
435 	if (unlikely(dentry == ERR_PTR(-ENOENT)))
436 		dentry = __fh_to_dentry(sb, cfh->parent_ino);
437 	return dentry;
438 }
439 
440 static int __get_snap_name(struct dentry *parent, char *name,
441 			   struct dentry *child)
442 {
443 	struct inode *inode = d_inode(child);
444 	struct inode *dir = d_inode(parent);
445 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
446 	struct ceph_mds_request *req = NULL;
447 	char *last_name = NULL;
448 	unsigned next_offset = 2;
449 	int err = -EINVAL;
450 
451 	if (ceph_ino(inode) != ceph_ino(dir))
452 		goto out;
453 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
454 		if (ceph_snap(dir) == CEPH_NOSNAP) {
455 			strcpy(name, fsc->mount_options->snapdir_name);
456 			err = 0;
457 		}
458 		goto out;
459 	}
460 	if (ceph_snap(dir) != CEPH_SNAPDIR)
461 		goto out;
462 
463 	while (1) {
464 		struct ceph_mds_reply_info_parsed *rinfo;
465 		struct ceph_mds_reply_dir_entry *rde;
466 		int i;
467 
468 		req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
469 					       USE_AUTH_MDS);
470 		if (IS_ERR(req)) {
471 			err = PTR_ERR(req);
472 			req = NULL;
473 			goto out;
474 		}
475 		err = ceph_alloc_readdir_reply_buffer(req, inode);
476 		if (err)
477 			goto out;
478 
479 		req->r_direct_mode = USE_AUTH_MDS;
480 		req->r_readdir_offset = next_offset;
481 		req->r_args.readdir.flags =
482 				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
483 		if (last_name) {
484 			req->r_path2 = last_name;
485 			last_name = NULL;
486 		}
487 
488 		req->r_inode = dir;
489 		ihold(dir);
490 		req->r_dentry = dget(parent);
491 
492 		inode_lock(dir);
493 		err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
494 		inode_unlock(dir);
495 
496 		if (err < 0)
497 			goto out;
498 
499 		rinfo = &req->r_reply_info;
500 		for (i = 0; i < rinfo->dir_nr; i++) {
501 			rde = rinfo->dir_entries + i;
502 			BUG_ON(!rde->inode.in);
503 			if (ceph_snap(inode) ==
504 			    le64_to_cpu(rde->inode.in->snapid)) {
505 				memcpy(name, rde->name, rde->name_len);
506 				name[rde->name_len] = '\0';
507 				err = 0;
508 				goto out;
509 			}
510 		}
511 
512 		if (rinfo->dir_end)
513 			break;
514 
515 		BUG_ON(rinfo->dir_nr <= 0);
516 		rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
517 		next_offset += rinfo->dir_nr;
518 		last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
519 		if (!last_name) {
520 			err = -ENOMEM;
521 			goto out;
522 		}
523 
524 		ceph_mdsc_put_request(req);
525 		req = NULL;
526 	}
527 	err = -ENOENT;
528 out:
529 	if (req)
530 		ceph_mdsc_put_request(req);
531 	kfree(last_name);
532 	doutc(fsc->client, "child dentry %p %p %llx.%llx err=%d\n", child,
533 	      inode, ceph_vinop(inode), err);
534 	return err;
535 }
536 
537 static int ceph_get_name(struct dentry *parent, char *name,
538 			 struct dentry *child)
539 {
540 	struct ceph_mds_client *mdsc;
541 	struct ceph_mds_request *req;
542 	struct inode *dir = d_inode(parent);
543 	struct inode *inode = d_inode(child);
544 	struct ceph_mds_reply_info_parsed *rinfo;
545 	int err;
546 
547 	if (ceph_snap(inode) != CEPH_NOSNAP)
548 		return __get_snap_name(parent, name, child);
549 
550 	mdsc = ceph_inode_to_fs_client(inode)->mdsc;
551 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
552 				       USE_ANY_MDS);
553 	if (IS_ERR(req))
554 		return PTR_ERR(req);
555 
556 	inode_lock(dir);
557 	req->r_inode = inode;
558 	ihold(inode);
559 	req->r_ino2 = ceph_vino(d_inode(parent));
560 	req->r_parent = dir;
561 	ihold(dir);
562 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
563 	req->r_num_caps = 2;
564 	err = ceph_mdsc_do_request(mdsc, NULL, req);
565 	inode_unlock(dir);
566 
567 	if (err)
568 		goto out;
569 
570 	rinfo = &req->r_reply_info;
571 	if (!IS_ENCRYPTED(dir)) {
572 		memcpy(name, rinfo->dname, rinfo->dname_len);
573 		name[rinfo->dname_len] = 0;
574 	} else {
575 		struct fscrypt_str oname = FSTR_INIT(NULL, 0);
576 		struct ceph_fname fname = { .dir	= dir,
577 					    .name	= rinfo->dname,
578 					    .ctext	= rinfo->altname,
579 					    .name_len	= rinfo->dname_len,
580 					    .ctext_len	= rinfo->altname_len };
581 
582 		err = ceph_fname_alloc_buffer(dir, &oname);
583 		if (err < 0)
584 			goto out;
585 
586 		err = ceph_fname_to_usr(&fname, NULL, &oname, NULL);
587 		if (!err) {
588 			memcpy(name, oname.name, oname.len);
589 			name[oname.len] = 0;
590 		}
591 		ceph_fname_free_buffer(dir, &oname);
592 	}
593 out:
594 	doutc(mdsc->fsc->client, "child dentry %p %p %llx.%llx err %d %s%s\n",
595 	      child, inode, ceph_vinop(inode), err, err ? "" : "name ",
596 	      err ? "" : name);
597 	ceph_mdsc_put_request(req);
598 	return err;
599 }
600 
601 const struct export_operations ceph_export_ops = {
602 	.encode_fh = ceph_encode_fh,
603 	.fh_to_dentry = ceph_fh_to_dentry,
604 	.fh_to_parent = ceph_fh_to_parent,
605 	.get_parent = ceph_get_parent,
606 	.get_name = ceph_get_name,
607 };
608