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 <linux/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 #define BYTES_PER_U32 (sizeof(u32))
37 #define CEPH_FH_BASIC_SIZE \
38 (sizeof(struct ceph_nfs_fh) / BYTES_PER_U32)
39 #define CEPH_FH_WITH_PARENT_SIZE \
40 (sizeof(struct ceph_nfs_confh) / BYTES_PER_U32)
41 #define CEPH_FH_SNAPPED_INODE_SIZE \
42 (sizeof(struct ceph_nfs_snapfh) / BYTES_PER_U32)
43
ceph_encode_snapfh(struct inode * inode,u32 * rawfh,int * max_len,struct inode * parent_inode)44 static int ceph_encode_snapfh(struct inode *inode, u32 *rawfh, int *max_len,
45 struct inode *parent_inode)
46 {
47 struct ceph_client *cl = ceph_inode_to_client(inode);
48 static const int snap_handle_length = CEPH_FH_SNAPPED_INODE_SIZE;
49 struct ceph_nfs_snapfh *sfh = (void *)rawfh;
50 u64 snapid = ceph_snap(inode);
51 int ret;
52 bool no_parent = true;
53
54 if (*max_len < snap_handle_length) {
55 *max_len = snap_handle_length;
56 ret = FILEID_INVALID;
57 goto out;
58 }
59
60 ret = -EINVAL;
61 if (snapid != CEPH_SNAPDIR) {
62 struct inode *dir;
63 struct dentry *dentry = d_find_alias(inode);
64 if (!dentry)
65 goto out;
66
67 rcu_read_lock();
68 dir = d_inode_rcu(dentry->d_parent);
69 if (ceph_snap(dir) != CEPH_SNAPDIR) {
70 sfh->parent_ino = ceph_ino(dir);
71 sfh->hash = ceph_dentry_hash(dir, dentry);
72 no_parent = false;
73 }
74 rcu_read_unlock();
75 dput(dentry);
76 }
77
78 if (no_parent) {
79 if (!S_ISDIR(inode->i_mode))
80 goto out;
81 sfh->parent_ino = sfh->ino;
82 sfh->hash = 0;
83 }
84 sfh->ino = ceph_ino(inode);
85 sfh->snapid = snapid;
86
87 *max_len = snap_handle_length;
88 ret = FILEID_BTRFS_WITH_PARENT;
89 out:
90 doutc(cl, "%p %llx.%llx ret=%d\n", inode, ceph_vinop(inode), ret);
91 return ret;
92 }
93
ceph_encode_fh(struct inode * inode,u32 * rawfh,int * max_len,struct inode * parent_inode)94 static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
95 struct inode *parent_inode)
96 {
97 struct ceph_client *cl = ceph_inode_to_client(inode);
98 static const int handle_length = CEPH_FH_BASIC_SIZE;
99 static const int connected_handle_length = CEPH_FH_WITH_PARENT_SIZE;
100 int type;
101
102 if (ceph_snap(inode) != CEPH_NOSNAP)
103 return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
104
105 if (parent_inode && (*max_len < connected_handle_length)) {
106 *max_len = connected_handle_length;
107 return FILEID_INVALID;
108 } else if (*max_len < handle_length) {
109 *max_len = handle_length;
110 return FILEID_INVALID;
111 }
112
113 if (parent_inode) {
114 struct ceph_nfs_confh *cfh = (void *)rawfh;
115 doutc(cl, "%p %llx.%llx with parent %p %llx.%llx\n", inode,
116 ceph_vinop(inode), parent_inode, ceph_vinop(parent_inode));
117 cfh->ino = ceph_ino(inode);
118 cfh->parent_ino = ceph_ino(parent_inode);
119 *max_len = connected_handle_length;
120 type = FILEID_INO32_GEN_PARENT;
121 } else {
122 struct ceph_nfs_fh *fh = (void *)rawfh;
123 doutc(cl, "%p %llx.%llx\n", inode, ceph_vinop(inode));
124 fh->ino = ceph_ino(inode);
125 *max_len = handle_length;
126 type = FILEID_INO32_GEN;
127 }
128 return type;
129 }
130
__lookup_inode(struct super_block * sb,u64 ino)131 static struct inode *__lookup_inode(struct super_block *sb, u64 ino)
132 {
133 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
134 struct inode *inode;
135 struct ceph_vino vino;
136 int err;
137
138 vino.ino = ino;
139 vino.snap = CEPH_NOSNAP;
140
141 if (ceph_vino_is_reserved(vino))
142 return ERR_PTR(-ESTALE);
143
144 inode = ceph_find_inode(sb, vino);
145 if (!inode) {
146 struct ceph_mds_request *req;
147 int mask;
148
149 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
150 USE_ANY_MDS);
151 if (IS_ERR(req))
152 return ERR_CAST(req);
153
154 mask = CEPH_STAT_CAP_INODE;
155 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
156 mask |= CEPH_CAP_XATTR_SHARED;
157 req->r_args.lookupino.mask = cpu_to_le32(mask);
158
159 req->r_ino1 = vino;
160 req->r_num_caps = 1;
161 err = ceph_mdsc_do_request(mdsc, NULL, req);
162 inode = req->r_target_inode;
163 if (inode)
164 ihold(inode);
165 ceph_mdsc_put_request(req);
166 if (!inode)
167 return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
168 } else {
169 if (ceph_inode_is_shutdown(inode)) {
170 iput(inode);
171 return ERR_PTR(-ESTALE);
172 }
173 }
174 return inode;
175 }
176
ceph_lookup_inode(struct super_block * sb,u64 ino)177 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
178 {
179 struct inode *inode = __lookup_inode(sb, ino);
180 if (IS_ERR(inode))
181 return inode;
182 if (inode->i_nlink == 0) {
183 iput(inode);
184 return ERR_PTR(-ESTALE);
185 }
186 return inode;
187 }
188
__fh_to_dentry(struct super_block * sb,u64 ino)189 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
190 {
191 struct inode *inode = __lookup_inode(sb, ino);
192 struct ceph_inode_info *ci = ceph_inode(inode);
193 int err;
194
195 if (IS_ERR(inode))
196 return ERR_CAST(inode);
197 /* We need LINK caps to reliably check i_nlink */
198 err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
199 if (err) {
200 iput(inode);
201 return ERR_PTR(err);
202 }
203 /* -ESTALE if inode as been unlinked and no file is open */
204 if ((inode->i_nlink == 0) && !__ceph_is_file_opened(ci)) {
205 iput(inode);
206 return ERR_PTR(-ESTALE);
207 }
208 return d_obtain_alias(inode);
209 }
210
__snapfh_to_dentry(struct super_block * sb,struct ceph_nfs_snapfh * sfh,bool want_parent)211 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
212 struct ceph_nfs_snapfh *sfh,
213 bool want_parent)
214 {
215 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
216 struct ceph_client *cl = mdsc->fsc->client;
217 struct ceph_mds_request *req;
218 struct inode *inode;
219 struct ceph_vino vino;
220 int mask;
221 int err;
222 bool unlinked = false;
223
224 if (want_parent) {
225 vino.ino = sfh->parent_ino;
226 if (sfh->snapid == CEPH_SNAPDIR)
227 vino.snap = CEPH_NOSNAP;
228 else if (sfh->ino == sfh->parent_ino)
229 vino.snap = CEPH_SNAPDIR;
230 else
231 vino.snap = sfh->snapid;
232 } else {
233 vino.ino = sfh->ino;
234 vino.snap = sfh->snapid;
235 }
236
237 if (ceph_vino_is_reserved(vino))
238 return ERR_PTR(-ESTALE);
239
240 inode = ceph_find_inode(sb, vino);
241 if (inode) {
242 if (ceph_inode_is_shutdown(inode)) {
243 iput(inode);
244 return ERR_PTR(-ESTALE);
245 }
246 return d_obtain_alias(inode);
247 }
248
249 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
250 USE_ANY_MDS);
251 if (IS_ERR(req))
252 return ERR_CAST(req);
253
254 mask = CEPH_STAT_CAP_INODE;
255 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
256 mask |= CEPH_CAP_XATTR_SHARED;
257 req->r_args.lookupino.mask = cpu_to_le32(mask);
258 if (vino.snap < CEPH_NOSNAP) {
259 req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
260 if (!want_parent && sfh->ino != sfh->parent_ino) {
261 req->r_args.lookupino.parent =
262 cpu_to_le64(sfh->parent_ino);
263 req->r_args.lookupino.hash =
264 cpu_to_le32(sfh->hash);
265 }
266 }
267
268 req->r_ino1 = vino;
269 req->r_num_caps = 1;
270 err = ceph_mdsc_do_request(mdsc, NULL, req);
271 inode = req->r_target_inode;
272 if (inode) {
273 if (vino.snap == CEPH_SNAPDIR) {
274 if (inode->i_nlink == 0)
275 unlinked = true;
276 inode = ceph_get_snapdir(inode);
277 } else if (ceph_snap(inode) == vino.snap) {
278 ihold(inode);
279 } else {
280 /* mds does not support lookup snapped inode */
281 inode = ERR_PTR(-EOPNOTSUPP);
282 }
283 } else {
284 inode = ERR_PTR(-ESTALE);
285 }
286 ceph_mdsc_put_request(req);
287
288 if (want_parent) {
289 doutc(cl, "%llx.%llx\n err=%d\n", vino.ino, vino.snap, err);
290 } else {
291 doutc(cl, "%llx.%llx parent %llx hash %x err=%d", vino.ino,
292 vino.snap, sfh->parent_ino, sfh->hash, err);
293 }
294 /* see comments in ceph_get_parent() */
295 return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
296 }
297
298 /*
299 * convert regular fh to dentry
300 */
ceph_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)301 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
302 struct fid *fid,
303 int fh_len, int fh_type)
304 {
305 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
306 struct ceph_nfs_fh *fh = (void *)fid->raw;
307
308 if (fh_type == FILEID_BTRFS_WITH_PARENT) {
309 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
310 return __snapfh_to_dentry(sb, sfh, false);
311 }
312
313 if (fh_type != FILEID_INO32_GEN &&
314 fh_type != FILEID_INO32_GEN_PARENT)
315 return NULL;
316 if (fh_len < sizeof(*fh) / BYTES_PER_U32)
317 return NULL;
318
319 doutc(fsc->client, "%llx\n", fh->ino);
320 return __fh_to_dentry(sb, fh->ino);
321 }
322
__get_parent(struct super_block * sb,struct dentry * child,u64 ino)323 static struct dentry *__get_parent(struct super_block *sb,
324 struct dentry *child, u64 ino)
325 {
326 struct ceph_mds_client *mdsc = ceph_sb_to_fs_client(sb)->mdsc;
327 struct ceph_mds_request *req;
328 struct inode *inode;
329 int mask;
330 int err;
331
332 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
333 USE_ANY_MDS);
334 if (IS_ERR(req))
335 return ERR_CAST(req);
336
337 if (child) {
338 req->r_inode = d_inode(child);
339 ihold(d_inode(child));
340 } else {
341 req->r_ino1 = (struct ceph_vino) {
342 .ino = ino,
343 .snap = CEPH_NOSNAP,
344 };
345 }
346
347 mask = CEPH_STAT_CAP_INODE;
348 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
349 mask |= CEPH_CAP_XATTR_SHARED;
350 req->r_args.getattr.mask = cpu_to_le32(mask);
351
352 req->r_num_caps = 1;
353 err = ceph_mdsc_do_request(mdsc, NULL, req);
354 if (err) {
355 ceph_mdsc_put_request(req);
356 return ERR_PTR(err);
357 }
358
359 inode = req->r_target_inode;
360 if (inode)
361 ihold(inode);
362 ceph_mdsc_put_request(req);
363 if (!inode)
364 return ERR_PTR(-ENOENT);
365
366 return d_obtain_alias(inode);
367 }
368
ceph_get_parent(struct dentry * child)369 static struct dentry *ceph_get_parent(struct dentry *child)
370 {
371 struct inode *inode = d_inode(child);
372 struct ceph_client *cl = ceph_inode_to_client(inode);
373 struct dentry *dn;
374
375 if (ceph_snap(inode) != CEPH_NOSNAP) {
376 struct inode* dir;
377 bool unlinked = false;
378 /* do not support non-directory */
379 if (!d_is_dir(child)) {
380 dn = ERR_PTR(-EINVAL);
381 goto out;
382 }
383 dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
384 if (IS_ERR(dir)) {
385 dn = ERR_CAST(dir);
386 goto out;
387 }
388 /* There can be multiple paths to access snapped inode.
389 * For simplicity, treat snapdir of head inode as parent */
390 if (ceph_snap(inode) != CEPH_SNAPDIR) {
391 struct inode *snapdir = ceph_get_snapdir(dir);
392 if (dir->i_nlink == 0)
393 unlinked = true;
394 iput(dir);
395 if (IS_ERR(snapdir)) {
396 dn = ERR_CAST(snapdir);
397 goto out;
398 }
399 dir = snapdir;
400 }
401 /* If directory has already been deleted, further get_parent
402 * will fail. Do not mark snapdir dentry as disconnected,
403 * this prevents exportfs from doing further get_parent. */
404 if (unlinked)
405 dn = d_obtain_root(dir);
406 else
407 dn = d_obtain_alias(dir);
408 } else {
409 dn = __get_parent(child->d_sb, child, 0);
410 }
411 out:
412 doutc(cl, "child %p %p %llx.%llx err=%ld\n", child, inode,
413 ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
414 return dn;
415 }
416
417 /*
418 * convert regular fh to parent
419 */
ceph_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)420 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
421 struct fid *fid,
422 int fh_len, int fh_type)
423 {
424 struct ceph_fs_client *fsc = ceph_sb_to_fs_client(sb);
425 struct ceph_nfs_confh *cfh = (void *)fid->raw;
426 struct dentry *dentry;
427
428 if (fh_type == FILEID_BTRFS_WITH_PARENT) {
429 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
430 return __snapfh_to_dentry(sb, sfh, true);
431 }
432
433 if (fh_type != FILEID_INO32_GEN_PARENT)
434 return NULL;
435 if (fh_len < sizeof(*cfh) / BYTES_PER_U32)
436 return NULL;
437
438 doutc(fsc->client, "%llx\n", cfh->parent_ino);
439 dentry = __get_parent(sb, NULL, cfh->ino);
440 if (unlikely(dentry == ERR_PTR(-ENOENT)))
441 dentry = __fh_to_dentry(sb, cfh->parent_ino);
442 return dentry;
443 }
444
__get_snap_name(struct dentry * parent,char * name,struct dentry * child)445 static int __get_snap_name(struct dentry *parent, char *name,
446 struct dentry *child)
447 {
448 struct inode *inode = d_inode(child);
449 struct inode *dir = d_inode(parent);
450 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
451 struct ceph_mds_request *req = NULL;
452 char *last_name = NULL;
453 unsigned next_offset = 2;
454 int err = -EINVAL;
455
456 if (ceph_ino(inode) != ceph_ino(dir))
457 goto out;
458 if (ceph_snap(inode) == CEPH_SNAPDIR) {
459 if (ceph_snap(dir) == CEPH_NOSNAP) {
460 /*
461 * .get_name() from struct export_operations
462 * assumes that its 'name' parameter is pointing
463 * to a NAME_MAX+1 sized buffer
464 */
465 strscpy(name, fsc->mount_options->snapdir_name,
466 NAME_MAX + 1);
467 err = 0;
468 }
469 goto out;
470 }
471 if (ceph_snap(dir) != CEPH_SNAPDIR)
472 goto out;
473
474 while (1) {
475 struct ceph_mds_reply_info_parsed *rinfo;
476 struct ceph_mds_reply_dir_entry *rde;
477 int i;
478
479 req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
480 USE_AUTH_MDS);
481 if (IS_ERR(req)) {
482 err = PTR_ERR(req);
483 req = NULL;
484 goto out;
485 }
486 err = ceph_alloc_readdir_reply_buffer(req, inode);
487 if (err)
488 goto out;
489
490 req->r_direct_mode = USE_AUTH_MDS;
491 req->r_readdir_offset = next_offset;
492 req->r_args.readdir.flags =
493 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
494 if (last_name) {
495 req->r_path2 = last_name;
496 last_name = NULL;
497 }
498
499 req->r_inode = dir;
500 ihold(dir);
501 req->r_dentry = dget(parent);
502
503 inode_lock(dir);
504 err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
505 inode_unlock(dir);
506
507 if (err < 0)
508 goto out;
509
510 rinfo = &req->r_reply_info;
511 for (i = 0; i < rinfo->dir_nr; i++) {
512 rde = rinfo->dir_entries + i;
513 BUG_ON(!rde->inode.in);
514 if (ceph_snap(inode) ==
515 le64_to_cpu(rde->inode.in->snapid)) {
516 memcpy(name, rde->name, rde->name_len);
517 name[rde->name_len] = '\0';
518 err = 0;
519 goto out;
520 }
521 }
522
523 if (rinfo->dir_end)
524 break;
525
526 BUG_ON(rinfo->dir_nr <= 0);
527 rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
528 next_offset += rinfo->dir_nr;
529 last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
530 if (!last_name) {
531 err = -ENOMEM;
532 goto out;
533 }
534
535 ceph_mdsc_put_request(req);
536 req = NULL;
537 }
538 err = -ENOENT;
539 out:
540 if (req)
541 ceph_mdsc_put_request(req);
542 kfree(last_name);
543 doutc(fsc->client, "child dentry %p %p %llx.%llx err=%d\n", child,
544 inode, ceph_vinop(inode), err);
545 return err;
546 }
547
ceph_get_name(struct dentry * parent,char * name,struct dentry * child)548 static int ceph_get_name(struct dentry *parent, char *name,
549 struct dentry *child)
550 {
551 struct ceph_mds_client *mdsc;
552 struct ceph_mds_request *req;
553 struct inode *dir = d_inode(parent);
554 struct inode *inode = d_inode(child);
555 struct ceph_mds_reply_info_parsed *rinfo;
556 int err;
557
558 if (ceph_snap(inode) != CEPH_NOSNAP)
559 return __get_snap_name(parent, name, child);
560
561 mdsc = ceph_inode_to_fs_client(inode)->mdsc;
562 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
563 USE_ANY_MDS);
564 if (IS_ERR(req))
565 return PTR_ERR(req);
566
567 inode_lock(dir);
568 req->r_inode = inode;
569 ihold(inode);
570 req->r_ino2 = ceph_vino(d_inode(parent));
571 req->r_parent = dir;
572 ihold(dir);
573 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
574 req->r_num_caps = 2;
575 err = ceph_mdsc_do_request(mdsc, NULL, req);
576 inode_unlock(dir);
577
578 if (err)
579 goto out;
580
581 rinfo = &req->r_reply_info;
582 if (!IS_ENCRYPTED(dir)) {
583 memcpy(name, rinfo->dname, rinfo->dname_len);
584 name[rinfo->dname_len] = 0;
585 } else {
586 struct fscrypt_str oname = FSTR_INIT(NULL, 0);
587 struct ceph_fname fname = { .dir = dir,
588 .name = rinfo->dname,
589 .ctext = rinfo->altname,
590 .name_len = rinfo->dname_len,
591 .ctext_len = rinfo->altname_len };
592
593 err = ceph_fname_alloc_buffer(dir, &oname);
594 if (err < 0)
595 goto out;
596
597 err = ceph_fname_to_usr(&fname, NULL, &oname, NULL);
598 if (!err) {
599 memcpy(name, oname.name, oname.len);
600 name[oname.len] = 0;
601 }
602 ceph_fname_free_buffer(dir, &oname);
603 }
604 out:
605 doutc(mdsc->fsc->client, "child dentry %p %p %llx.%llx err %d %s%s\n",
606 child, inode, ceph_vinop(inode), err, err ? "" : "name ",
607 err ? "" : name);
608 ceph_mdsc_put_request(req);
609 return err;
610 }
611
612 const struct export_operations ceph_export_ops = {
613 .encode_fh = ceph_encode_fh,
614 .fh_to_dentry = ceph_fh_to_dentry,
615 .fh_to_parent = ceph_fh_to_parent,
616 .get_parent = ceph_get_parent,
617 .get_name = ceph_get_name,
618 };
619