xref: /openbmc/linux/fs/ceph/export.c (revision aac28965)
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 	} else {
161 		if (ceph_inode_is_shutdown(inode)) {
162 			iput(inode);
163 			return ERR_PTR(-ESTALE);
164 		}
165 	}
166 	return inode;
167 }
168 
169 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
170 {
171 	struct inode *inode = __lookup_inode(sb, ino);
172 	if (IS_ERR(inode))
173 		return inode;
174 	if (inode->i_nlink == 0) {
175 		iput(inode);
176 		return ERR_PTR(-ESTALE);
177 	}
178 	return inode;
179 }
180 
181 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
182 {
183 	struct inode *inode = __lookup_inode(sb, ino);
184 	int err;
185 
186 	if (IS_ERR(inode))
187 		return ERR_CAST(inode);
188 	/* We need LINK caps to reliably check i_nlink */
189 	err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
190 	if (err) {
191 		iput(inode);
192 		return ERR_PTR(err);
193 	}
194 	/* -ESTALE if inode as been unlinked and no file is open */
195 	if ((inode->i_nlink == 0) && (atomic_read(&inode->i_count) == 1)) {
196 		iput(inode);
197 		return ERR_PTR(-ESTALE);
198 	}
199 	return d_obtain_alias(inode);
200 }
201 
202 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
203 					  struct ceph_nfs_snapfh *sfh,
204 					  bool want_parent)
205 {
206 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
207 	struct ceph_mds_request *req;
208 	struct inode *inode;
209 	struct ceph_vino vino;
210 	int mask;
211 	int err;
212 	bool unlinked = false;
213 
214 	if (want_parent) {
215 		vino.ino = sfh->parent_ino;
216 		if (sfh->snapid == CEPH_SNAPDIR)
217 			vino.snap = CEPH_NOSNAP;
218 		else if (sfh->ino == sfh->parent_ino)
219 			vino.snap = CEPH_SNAPDIR;
220 		else
221 			vino.snap = sfh->snapid;
222 	} else {
223 		vino.ino = sfh->ino;
224 		vino.snap = sfh->snapid;
225 	}
226 
227 	if (ceph_vino_is_reserved(vino))
228 		return ERR_PTR(-ESTALE);
229 
230 	inode = ceph_find_inode(sb, vino);
231 	if (inode) {
232 		if (ceph_inode_is_shutdown(inode)) {
233 			iput(inode);
234 			return ERR_PTR(-ESTALE);
235 		}
236 		return d_obtain_alias(inode);
237 	}
238 
239 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
240 				       USE_ANY_MDS);
241 	if (IS_ERR(req))
242 		return ERR_CAST(req);
243 
244 	mask = CEPH_STAT_CAP_INODE;
245 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
246 		mask |= CEPH_CAP_XATTR_SHARED;
247 	req->r_args.lookupino.mask = cpu_to_le32(mask);
248 	if (vino.snap < CEPH_NOSNAP) {
249 		req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
250 		if (!want_parent && sfh->ino != sfh->parent_ino) {
251 			req->r_args.lookupino.parent =
252 					cpu_to_le64(sfh->parent_ino);
253 			req->r_args.lookupino.hash =
254 					cpu_to_le32(sfh->hash);
255 		}
256 	}
257 
258 	req->r_ino1 = vino;
259 	req->r_num_caps = 1;
260 	err = ceph_mdsc_do_request(mdsc, NULL, req);
261 	inode = req->r_target_inode;
262 	if (inode) {
263 		if (vino.snap == CEPH_SNAPDIR) {
264 			if (inode->i_nlink == 0)
265 				unlinked = true;
266 			inode = ceph_get_snapdir(inode);
267 		} else if (ceph_snap(inode) == vino.snap) {
268 			ihold(inode);
269 		} else {
270 			/* mds does not support lookup snapped inode */
271 			inode = ERR_PTR(-EOPNOTSUPP);
272 		}
273 	} else {
274 		inode = ERR_PTR(-ESTALE);
275 	}
276 	ceph_mdsc_put_request(req);
277 
278 	if (want_parent) {
279 		dout("snapfh_to_parent %llx.%llx\n err=%d\n",
280 		     vino.ino, vino.snap, err);
281 	} else {
282 		dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
283 		      vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
284 	}
285 	if (IS_ERR(inode))
286 		return ERR_CAST(inode);
287 	/* see comments in ceph_get_parent() */
288 	return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
289 }
290 
291 /*
292  * convert regular fh to dentry
293  */
294 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
295 					struct fid *fid,
296 					int fh_len, int fh_type)
297 {
298 	struct ceph_nfs_fh *fh = (void *)fid->raw;
299 
300 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
301 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
302 		return __snapfh_to_dentry(sb, sfh, false);
303 	}
304 
305 	if (fh_type != FILEID_INO32_GEN  &&
306 	    fh_type != FILEID_INO32_GEN_PARENT)
307 		return NULL;
308 	if (fh_len < sizeof(*fh) / 4)
309 		return NULL;
310 
311 	dout("fh_to_dentry %llx\n", fh->ino);
312 	return __fh_to_dentry(sb, fh->ino);
313 }
314 
315 static struct dentry *__get_parent(struct super_block *sb,
316 				   struct dentry *child, u64 ino)
317 {
318 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
319 	struct ceph_mds_request *req;
320 	struct inode *inode;
321 	int mask;
322 	int err;
323 
324 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
325 				       USE_ANY_MDS);
326 	if (IS_ERR(req))
327 		return ERR_CAST(req);
328 
329 	if (child) {
330 		req->r_inode = d_inode(child);
331 		ihold(d_inode(child));
332 	} else {
333 		req->r_ino1 = (struct ceph_vino) {
334 			.ino = ino,
335 			.snap = CEPH_NOSNAP,
336 		};
337 	}
338 
339 	mask = CEPH_STAT_CAP_INODE;
340 	if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
341 		mask |= CEPH_CAP_XATTR_SHARED;
342 	req->r_args.getattr.mask = cpu_to_le32(mask);
343 
344 	req->r_num_caps = 1;
345 	err = ceph_mdsc_do_request(mdsc, NULL, req);
346 	if (err) {
347 		ceph_mdsc_put_request(req);
348 		return ERR_PTR(err);
349 	}
350 
351 	inode = req->r_target_inode;
352 	if (inode)
353 		ihold(inode);
354 	ceph_mdsc_put_request(req);
355 	if (!inode)
356 		return ERR_PTR(-ENOENT);
357 
358 	return d_obtain_alias(inode);
359 }
360 
361 static struct dentry *ceph_get_parent(struct dentry *child)
362 {
363 	struct inode *inode = d_inode(child);
364 	struct dentry *dn;
365 
366 	if (ceph_snap(inode) != CEPH_NOSNAP) {
367 		struct inode* dir;
368 		bool unlinked = false;
369 		/* do not support non-directory */
370 		if (!d_is_dir(child)) {
371 			dn = ERR_PTR(-EINVAL);
372 			goto out;
373 		}
374 		dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
375 		if (IS_ERR(dir)) {
376 			dn = ERR_CAST(dir);
377 			goto out;
378 		}
379 		/* There can be multiple paths to access snapped inode.
380 		 * For simplicity, treat snapdir of head inode as parent */
381 		if (ceph_snap(inode) != CEPH_SNAPDIR) {
382 			struct inode *snapdir = ceph_get_snapdir(dir);
383 			if (dir->i_nlink == 0)
384 				unlinked = true;
385 			iput(dir);
386 			if (IS_ERR(snapdir)) {
387 				dn = ERR_CAST(snapdir);
388 				goto out;
389 			}
390 			dir = snapdir;
391 		}
392 		/* If directory has already been deleted, futher get_parent
393 		 * will fail. Do not mark snapdir dentry as disconnected,
394 		 * this prevent exportfs from doing futher get_parent. */
395 		if (unlinked)
396 			dn = d_obtain_root(dir);
397 		else
398 			dn = d_obtain_alias(dir);
399 	} else {
400 		dn = __get_parent(child->d_sb, child, 0);
401 	}
402 out:
403 	dout("get_parent %p ino %llx.%llx err=%ld\n",
404 	     child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
405 	return dn;
406 }
407 
408 /*
409  * convert regular fh to parent
410  */
411 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
412 					struct fid *fid,
413 					int fh_len, int fh_type)
414 {
415 	struct ceph_nfs_confh *cfh = (void *)fid->raw;
416 	struct dentry *dentry;
417 
418 	if (fh_type == FILEID_BTRFS_WITH_PARENT) {
419 		struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
420 		return __snapfh_to_dentry(sb, sfh, true);
421 	}
422 
423 	if (fh_type != FILEID_INO32_GEN_PARENT)
424 		return NULL;
425 	if (fh_len < sizeof(*cfh) / 4)
426 		return NULL;
427 
428 	dout("fh_to_parent %llx\n", cfh->parent_ino);
429 	dentry = __get_parent(sb, NULL, cfh->ino);
430 	if (unlikely(dentry == ERR_PTR(-ENOENT)))
431 		dentry = __fh_to_dentry(sb, cfh->parent_ino);
432 	return dentry;
433 }
434 
435 static int __get_snap_name(struct dentry *parent, char *name,
436 			   struct dentry *child)
437 {
438 	struct inode *inode = d_inode(child);
439 	struct inode *dir = d_inode(parent);
440 	struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
441 	struct ceph_mds_request *req = NULL;
442 	char *last_name = NULL;
443 	unsigned next_offset = 2;
444 	int err = -EINVAL;
445 
446 	if (ceph_ino(inode) != ceph_ino(dir))
447 		goto out;
448 	if (ceph_snap(inode) == CEPH_SNAPDIR) {
449 		if (ceph_snap(dir) == CEPH_NOSNAP) {
450 			strcpy(name, fsc->mount_options->snapdir_name);
451 			err = 0;
452 		}
453 		goto out;
454 	}
455 	if (ceph_snap(dir) != CEPH_SNAPDIR)
456 		goto out;
457 
458 	while (1) {
459 		struct ceph_mds_reply_info_parsed *rinfo;
460 		struct ceph_mds_reply_dir_entry *rde;
461 		int i;
462 
463 		req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
464 					       USE_AUTH_MDS);
465 		if (IS_ERR(req)) {
466 			err = PTR_ERR(req);
467 			req = NULL;
468 			goto out;
469 		}
470 		err = ceph_alloc_readdir_reply_buffer(req, inode);
471 		if (err)
472 			goto out;
473 
474 		req->r_direct_mode = USE_AUTH_MDS;
475 		req->r_readdir_offset = next_offset;
476 		req->r_args.readdir.flags =
477 				cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
478 		if (last_name) {
479 			req->r_path2 = last_name;
480 			last_name = NULL;
481 		}
482 
483 		req->r_inode = dir;
484 		ihold(dir);
485 		req->r_dentry = dget(parent);
486 
487 		inode_lock(dir);
488 		err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
489 		inode_unlock(dir);
490 
491 		if (err < 0)
492 			goto out;
493 
494 		rinfo = &req->r_reply_info;
495 		for (i = 0; i < rinfo->dir_nr; i++) {
496 			rde = rinfo->dir_entries + i;
497 			BUG_ON(!rde->inode.in);
498 			if (ceph_snap(inode) ==
499 			    le64_to_cpu(rde->inode.in->snapid)) {
500 				memcpy(name, rde->name, rde->name_len);
501 				name[rde->name_len] = '\0';
502 				err = 0;
503 				goto out;
504 			}
505 		}
506 
507 		if (rinfo->dir_end)
508 			break;
509 
510 		BUG_ON(rinfo->dir_nr <= 0);
511 		rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
512 		next_offset += rinfo->dir_nr;
513 		last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
514 		if (!last_name) {
515 			err = -ENOMEM;
516 			goto out;
517 		}
518 
519 		ceph_mdsc_put_request(req);
520 		req = NULL;
521 	}
522 	err = -ENOENT;
523 out:
524 	if (req)
525 		ceph_mdsc_put_request(req);
526 	kfree(last_name);
527 	dout("get_snap_name %p ino %llx.%llx err=%d\n",
528 	     child, ceph_vinop(inode), err);
529 	return err;
530 }
531 
532 static int ceph_get_name(struct dentry *parent, char *name,
533 			 struct dentry *child)
534 {
535 	struct ceph_mds_client *mdsc;
536 	struct ceph_mds_request *req;
537 	struct inode *inode = d_inode(child);
538 	int err;
539 
540 	if (ceph_snap(inode) != CEPH_NOSNAP)
541 		return __get_snap_name(parent, name, child);
542 
543 	mdsc = ceph_inode_to_client(inode)->mdsc;
544 	req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
545 				       USE_ANY_MDS);
546 	if (IS_ERR(req))
547 		return PTR_ERR(req);
548 
549 	inode_lock(d_inode(parent));
550 
551 	req->r_inode = inode;
552 	ihold(inode);
553 	req->r_ino2 = ceph_vino(d_inode(parent));
554 	req->r_parent = d_inode(parent);
555 	ihold(req->r_parent);
556 	set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
557 	req->r_num_caps = 2;
558 	err = ceph_mdsc_do_request(mdsc, NULL, req);
559 
560 	inode_unlock(d_inode(parent));
561 
562 	if (!err) {
563 		struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
564 		memcpy(name, rinfo->dname, rinfo->dname_len);
565 		name[rinfo->dname_len] = 0;
566 		dout("get_name %p ino %llx.%llx name %s\n",
567 		     child, ceph_vinop(inode), name);
568 	} else {
569 		dout("get_name %p ino %llx.%llx err %d\n",
570 		     child, ceph_vinop(inode), err);
571 	}
572 
573 	ceph_mdsc_put_request(req);
574 	return err;
575 }
576 
577 const struct export_operations ceph_export_ops = {
578 	.encode_fh = ceph_encode_fh,
579 	.fh_to_dentry = ceph_fh_to_dentry,
580 	.fh_to_parent = ceph_fh_to_parent,
581 	.get_parent = ceph_get_parent,
582 	.get_name = ceph_get_name,
583 };
584