xref: /openbmc/linux/fs/ceph/export.c (revision 2f4e3926)
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 	static const int snap_handle_length =
40 		sizeof(struct ceph_nfs_snapfh) >> 2;
41 	struct ceph_nfs_snapfh *sfh = (void *)rawfh;
42 	u64 snapid = ceph_snap(inode);
43 	int ret;
44 	bool no_parent = true;
45 
46 	if (*max_len < snap_handle_length) {
47 		*max_len = snap_handle_length;
48 		ret = FILEID_INVALID;
49 		goto out;
50 	}
51 
52 	ret =  -EINVAL;
53 	if (snapid != CEPH_SNAPDIR) {
54 		struct inode *dir;
55 		struct dentry *dentry = d_find_alias(inode);
56 		if (!dentry)
57 			goto out;
58 
59 		rcu_read_lock();
60 		dir = d_inode_rcu(dentry->d_parent);
61 		if (ceph_snap(dir) != CEPH_SNAPDIR) {
62 			sfh->parent_ino = ceph_ino(dir);
63 			sfh->hash = ceph_dentry_hash(dir, dentry);
64 			no_parent = false;
65 		}
66 		rcu_read_unlock();
67 		dput(dentry);
68 	}
69 
70 	if (no_parent) {
71 		if (!S_ISDIR(inode->i_mode))
72 			goto out;
73 		sfh->parent_ino = sfh->ino;
74 		sfh->hash = 0;
75 	}
76 	sfh->ino = ceph_ino(inode);
77 	sfh->snapid = snapid;
78 
79 	*max_len = snap_handle_length;
80 	ret = FILEID_BTRFS_WITH_PARENT;
81 out:
82 	dout("encode_snapfh %llx.%llx ret=%d\n", ceph_vinop(inode), ret);
83 	return ret;
84 }
85 
86 static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
87 			  struct inode *parent_inode)
88 {
89 	static const int handle_length =
90 		sizeof(struct ceph_nfs_fh) >> 2;
91 	static const int connected_handle_length =
92 		sizeof(struct ceph_nfs_confh) >> 2;
93 	int type;
94 
95 	if (ceph_snap(inode) != CEPH_NOSNAP)
96 		return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
97 
98 	if (parent_inode && (*max_len < connected_handle_length)) {
99 		*max_len = connected_handle_length;
100 		return FILEID_INVALID;
101 	} else if (*max_len < handle_length) {
102 		*max_len = handle_length;
103 		return FILEID_INVALID;
104 	}
105 
106 	if (parent_inode) {
107 		struct ceph_nfs_confh *cfh = (void *)rawfh;
108 		dout("encode_fh %llx with parent %llx\n",
109 		     ceph_ino(inode), ceph_ino(parent_inode));
110 		cfh->ino = ceph_ino(inode);
111 		cfh->parent_ino = ceph_ino(parent_inode);
112 		*max_len = connected_handle_length;
113 		type = FILEID_INO32_GEN_PARENT;
114 	} else {
115 		struct ceph_nfs_fh *fh = (void *)rawfh;
116 		dout("encode_fh %llx\n", ceph_ino(inode));
117 		fh->ino = ceph_ino(inode);
118 		*max_len = handle_length;
119 		type = FILEID_INO32_GEN;
120 	}
121 	return type;
122 }
123 
124 static struct inode *__lookup_inode(struct super_block *sb, u64 ino)
125 {
126 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
127 	struct inode *inode;
128 	struct ceph_vino vino;
129 	int err;
130 
131 	vino.ino = ino;
132 	vino.snap = CEPH_NOSNAP;
133 
134 	if (ceph_vino_is_reserved(vino))
135 		return ERR_PTR(-ESTALE);
136 
137 	inode = ceph_find_inode(sb, vino);
138 	if (!inode) {
139 		struct ceph_mds_request *req;
140 		int mask;
141 
142 		req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
143 					       USE_ANY_MDS);
144 		if (IS_ERR(req))
145 			return ERR_CAST(req);
146 
147 		mask = CEPH_STAT_CAP_INODE;
148 		if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
149 			mask |= CEPH_CAP_XATTR_SHARED;
150 		req->r_args.lookupino.mask = cpu_to_le32(mask);
151 
152 		req->r_ino1 = vino;
153 		req->r_num_caps = 1;
154 		err = ceph_mdsc_do_request(mdsc, NULL, req);
155 		inode = req->r_target_inode;
156 		if (inode)
157 			ihold(inode);
158 		ceph_mdsc_put_request(req);
159 		if (!inode)
160 			return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
161 	} else {
162 		if (ceph_inode_is_shutdown(inode)) {
163 			iput(inode);
164 			return ERR_PTR(-ESTALE);
165 		}
166 	}
167 	return inode;
168 }
169 
170 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
171 {
172 	struct inode *inode = __lookup_inode(sb, ino);
173 	if (IS_ERR(inode))
174 		return inode;
175 	if (inode->i_nlink == 0) {
176 		iput(inode);
177 		return ERR_PTR(-ESTALE);
178 	}
179 	return inode;
180 }
181 
182 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
183 {
184 	struct inode *inode = __lookup_inode(sb, ino);
185 	struct ceph_inode_info *ci = ceph_inode(inode);
186 	int err;
187 
188 	if (IS_ERR(inode))
189 		return ERR_CAST(inode);
190 	/* We need LINK caps to reliably check i_nlink */
191 	err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
192 	if (err) {
193 		iput(inode);
194 		return ERR_PTR(err);
195 	}
196 	/* -ESTALE if inode as been unlinked and no file is open */
197 	if ((inode->i_nlink == 0) && !__ceph_is_file_opened(ci)) {
198 		iput(inode);
199 		return ERR_PTR(-ESTALE);
200 	}
201 	return d_obtain_alias(inode);
202 }
203 
204 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
205 					  struct ceph_nfs_snapfh *sfh,
206 					  bool want_parent)
207 {
208 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
209 	struct ceph_mds_request *req;
210 	struct inode *inode;
211 	struct ceph_vino vino;
212 	int mask;
213 	int err;
214 	bool unlinked = false;
215 
216 	if (want_parent) {
217 		vino.ino = sfh->parent_ino;
218 		if (sfh->snapid == CEPH_SNAPDIR)
219 			vino.snap = CEPH_NOSNAP;
220 		else if (sfh->ino == sfh->parent_ino)
221 			vino.snap = CEPH_SNAPDIR;
222 		else
223 			vino.snap = sfh->snapid;
224 	} else {
225 		vino.ino = sfh->ino;
226 		vino.snap = sfh->snapid;
227 	}
228 
229 	if (ceph_vino_is_reserved(vino))
230 		return ERR_PTR(-ESTALE);
231 
232 	inode = ceph_find_inode(sb, vino);
233 	if (inode) {
234 		if (ceph_inode_is_shutdown(inode)) {
235 			iput(inode);
236 			return ERR_PTR(-ESTALE);
237 		}
238 		return d_obtain_alias(inode);
239 	}
240 
241 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
242 				       USE_ANY_MDS);
243 	if (IS_ERR(req))
244 		return ERR_CAST(req);
245 
246 	mask = CEPH_STAT_CAP_INODE;
247 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
248 		mask |= CEPH_CAP_XATTR_SHARED;
249 	req->r_args.lookupino.mask = cpu_to_le32(mask);
250 	if (vino.snap < CEPH_NOSNAP) {
251 		req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
252 		if (!want_parent && sfh->ino != sfh->parent_ino) {
253 			req->r_args.lookupino.parent =
254 					cpu_to_le64(sfh->parent_ino);
255 			req->r_args.lookupino.hash =
256 					cpu_to_le32(sfh->hash);
257 		}
258 	}
259 
260 	req->r_ino1 = vino;
261 	req->r_num_caps = 1;
262 	err = ceph_mdsc_do_request(mdsc, NULL, req);
263 	inode = req->r_target_inode;
264 	if (inode) {
265 		if (vino.snap == CEPH_SNAPDIR) {
266 			if (inode->i_nlink == 0)
267 				unlinked = true;
268 			inode = ceph_get_snapdir(inode);
269 		} else if (ceph_snap(inode) == vino.snap) {
270 			ihold(inode);
271 		} else {
272 			/* mds does not support lookup snapped inode */
273 			inode = ERR_PTR(-EOPNOTSUPP);
274 		}
275 	} else {
276 		inode = ERR_PTR(-ESTALE);
277 	}
278 	ceph_mdsc_put_request(req);
279 
280 	if (want_parent) {
281 		dout("snapfh_to_parent %llx.%llx\n err=%d\n",
282 		     vino.ino, vino.snap, err);
283 	} else {
284 		dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
285 		      vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
286 	}
287 	if (IS_ERR(inode))
288 		return ERR_CAST(inode);
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_nfs_fh *fh = (void *)fid->raw;
301 
302 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
303 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
304 		return __snapfh_to_dentry(sb, sfh, false);
305 	}
306 
307 	if (fh_type != FILEID_INO32_GEN  &&
308 	    fh_type != FILEID_INO32_GEN_PARENT)
309 		return NULL;
310 	if (fh_len < sizeof(*fh) / 4)
311 		return NULL;
312 
313 	dout("fh_to_dentry %llx\n", fh->ino);
314 	return __fh_to_dentry(sb, fh->ino);
315 }
316 
317 static struct dentry *__get_parent(struct super_block *sb,
318 				   struct dentry *child, u64 ino)
319 {
320 	struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
321 	struct ceph_mds_request *req;
322 	struct inode *inode;
323 	int mask;
324 	int err;
325 
326 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
327 				       USE_ANY_MDS);
328 	if (IS_ERR(req))
329 		return ERR_CAST(req);
330 
331 	if (child) {
332 		req->r_inode = d_inode(child);
333 		ihold(d_inode(child));
334 	} else {
335 		req->r_ino1 = (struct ceph_vino) {
336 			.ino = ino,
337 			.snap = CEPH_NOSNAP,
338 		};
339 	}
340 
341 	mask = CEPH_STAT_CAP_INODE;
342 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
343 		mask |= CEPH_CAP_XATTR_SHARED;
344 	req->r_args.getattr.mask = cpu_to_le32(mask);
345 
346 	req->r_num_caps = 1;
347 	err = ceph_mdsc_do_request(mdsc, NULL, req);
348 	if (err) {
349 		ceph_mdsc_put_request(req);
350 		return ERR_PTR(err);
351 	}
352 
353 	inode = req->r_target_inode;
354 	if (inode)
355 		ihold(inode);
356 	ceph_mdsc_put_request(req);
357 	if (!inode)
358 		return ERR_PTR(-ENOENT);
359 
360 	return d_obtain_alias(inode);
361 }
362 
363 static struct dentry *ceph_get_parent(struct dentry *child)
364 {
365 	struct inode *inode = d_inode(child);
366 	struct dentry *dn;
367 
368 	if (ceph_snap(inode) != CEPH_NOSNAP) {
369 		struct inode* dir;
370 		bool unlinked = false;
371 		/* do not support non-directory */
372 		if (!d_is_dir(child)) {
373 			dn = ERR_PTR(-EINVAL);
374 			goto out;
375 		}
376 		dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
377 		if (IS_ERR(dir)) {
378 			dn = ERR_CAST(dir);
379 			goto out;
380 		}
381 		/* There can be multiple paths to access snapped inode.
382 		 * For simplicity, treat snapdir of head inode as parent */
383 		if (ceph_snap(inode) != CEPH_SNAPDIR) {
384 			struct inode *snapdir = ceph_get_snapdir(dir);
385 			if (dir->i_nlink == 0)
386 				unlinked = true;
387 			iput(dir);
388 			if (IS_ERR(snapdir)) {
389 				dn = ERR_CAST(snapdir);
390 				goto out;
391 			}
392 			dir = snapdir;
393 		}
394 		/* If directory has already been deleted, futher get_parent
395 		 * will fail. Do not mark snapdir dentry as disconnected,
396 		 * this prevent exportfs from doing futher get_parent. */
397 		if (unlinked)
398 			dn = d_obtain_root(dir);
399 		else
400 			dn = d_obtain_alias(dir);
401 	} else {
402 		dn = __get_parent(child->d_sb, child, 0);
403 	}
404 out:
405 	dout("get_parent %p ino %llx.%llx err=%ld\n",
406 	     child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
407 	return dn;
408 }
409 
410 /*
411  * convert regular fh to parent
412  */
413 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
414 					struct fid *fid,
415 					int fh_len, int fh_type)
416 {
417 	struct ceph_nfs_confh *cfh = (void *)fid->raw;
418 	struct dentry *dentry;
419 
420 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
421 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
422 		return __snapfh_to_dentry(sb, sfh, true);
423 	}
424 
425 	if (fh_type != FILEID_INO32_GEN_PARENT)
426 		return NULL;
427 	if (fh_len < sizeof(*cfh) / 4)
428 		return NULL;
429 
430 	dout("fh_to_parent %llx\n", cfh->parent_ino);
431 	dentry = __get_parent(sb, NULL, cfh->ino);
432 	if (unlikely(dentry == ERR_PTR(-ENOENT)))
433 		dentry = __fh_to_dentry(sb, cfh->parent_ino);
434 	return dentry;
435 }
436 
437 static int __get_snap_name(struct dentry *parent, char *name,
438 			   struct dentry *child)
439 {
440 	struct inode *inode = d_inode(child);
441 	struct inode *dir = d_inode(parent);
442 	struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
443 	struct ceph_mds_request *req = NULL;
444 	char *last_name = NULL;
445 	unsigned next_offset = 2;
446 	int err = -EINVAL;
447 
448 	if (ceph_ino(inode) != ceph_ino(dir))
449 		goto out;
450 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
451 		if (ceph_snap(dir) == CEPH_NOSNAP) {
452 			strcpy(name, fsc->mount_options->snapdir_name);
453 			err = 0;
454 		}
455 		goto out;
456 	}
457 	if (ceph_snap(dir) != CEPH_SNAPDIR)
458 		goto out;
459 
460 	while (1) {
461 		struct ceph_mds_reply_info_parsed *rinfo;
462 		struct ceph_mds_reply_dir_entry *rde;
463 		int i;
464 
465 		req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
466 					       USE_AUTH_MDS);
467 		if (IS_ERR(req)) {
468 			err = PTR_ERR(req);
469 			req = NULL;
470 			goto out;
471 		}
472 		err = ceph_alloc_readdir_reply_buffer(req, inode);
473 		if (err)
474 			goto out;
475 
476 		req->r_direct_mode = USE_AUTH_MDS;
477 		req->r_readdir_offset = next_offset;
478 		req->r_args.readdir.flags =
479 				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
480 		if (last_name) {
481 			req->r_path2 = last_name;
482 			last_name = NULL;
483 		}
484 
485 		req->r_inode = dir;
486 		ihold(dir);
487 		req->r_dentry = dget(parent);
488 
489 		inode_lock(dir);
490 		err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
491 		inode_unlock(dir);
492 
493 		if (err < 0)
494 			goto out;
495 
496 		rinfo = &req->r_reply_info;
497 		for (i = 0; i < rinfo->dir_nr; i++) {
498 			rde = rinfo->dir_entries + i;
499 			BUG_ON(!rde->inode.in);
500 			if (ceph_snap(inode) ==
501 			    le64_to_cpu(rde->inode.in->snapid)) {
502 				memcpy(name, rde->name, rde->name_len);
503 				name[rde->name_len] = '\0';
504 				err = 0;
505 				goto out;
506 			}
507 		}
508 
509 		if (rinfo->dir_end)
510 			break;
511 
512 		BUG_ON(rinfo->dir_nr <= 0);
513 		rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
514 		next_offset += rinfo->dir_nr;
515 		last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
516 		if (!last_name) {
517 			err = -ENOMEM;
518 			goto out;
519 		}
520 
521 		ceph_mdsc_put_request(req);
522 		req = NULL;
523 	}
524 	err = -ENOENT;
525 out:
526 	if (req)
527 		ceph_mdsc_put_request(req);
528 	kfree(last_name);
529 	dout("get_snap_name %p ino %llx.%llx err=%d\n",
530 	     child, ceph_vinop(inode), err);
531 	return err;
532 }
533 
534 static int ceph_get_name(struct dentry *parent, char *name,
535 			 struct dentry *child)
536 {
537 	struct ceph_mds_client *mdsc;
538 	struct ceph_mds_request *req;
539 	struct inode *dir = d_inode(parent);
540 	struct inode *inode = d_inode(child);
541 	struct ceph_mds_reply_info_parsed *rinfo;
542 	int err;
543 
544 	if (ceph_snap(inode) != CEPH_NOSNAP)
545 		return __get_snap_name(parent, name, child);
546 
547 	mdsc = ceph_inode_to_fs_client(inode)->mdsc;
548 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
549 				       USE_ANY_MDS);
550 	if (IS_ERR(req))
551 		return PTR_ERR(req);
552 
553 	inode_lock(dir);
554 	req->r_inode = inode;
555 	ihold(inode);
556 	req->r_ino2 = ceph_vino(d_inode(parent));
557 	req->r_parent = dir;
558 	ihold(dir);
559 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
560 	req->r_num_caps = 2;
561 	err = ceph_mdsc_do_request(mdsc, NULL, req);
562 	inode_unlock(dir);
563 
564 	if (err)
565 		goto out;
566 
567 	rinfo = &req->r_reply_info;
568 	if (!IS_ENCRYPTED(dir)) {
569 		memcpy(name, rinfo->dname, rinfo->dname_len);
570 		name[rinfo->dname_len] = 0;
571 	} else {
572 		struct fscrypt_str oname = FSTR_INIT(NULL, 0);
573 		struct ceph_fname fname = { .dir	= dir,
574 					    .name	= rinfo->dname,
575 					    .ctext	= rinfo->altname,
576 					    .name_len	= rinfo->dname_len,
577 					    .ctext_len	= rinfo->altname_len };
578 
579 		err = ceph_fname_alloc_buffer(dir, &oname);
580 		if (err < 0)
581 			goto out;
582 
583 		err = ceph_fname_to_usr(&fname, NULL, &oname, NULL);
584 		if (!err) {
585 			memcpy(name, oname.name, oname.len);
586 			name[oname.len] = 0;
587 		}
588 		ceph_fname_free_buffer(dir, &oname);
589 	}
590 out:
591 	dout("get_name %p ino %llx.%llx err %d %s%s\n",
592 		     child, ceph_vinop(inode), err,
593 		     err ? "" : "name ", err ? "" : name);
594 	ceph_mdsc_put_request(req);
595 	return err;
596 }
597 
598 const struct export_operations ceph_export_ops = {
599 	.encode_fh = ceph_encode_fh,
600 	.fh_to_dentry = ceph_fh_to_dentry,
601 	.fh_to_parent = ceph_fh_to_parent,
602 	.get_parent = ceph_get_parent,
603 	.get_name = ceph_get_name,
604 };
605