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
ceph_encode_snapfh(struct inode * inode,u32 * rawfh,int * max_len,struct inode * parent_inode)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
ceph_encode_fh(struct inode * inode,u32 * rawfh,int * max_len,struct inode * parent_inode)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
__lookup_inode(struct super_block * sb,u64 ino)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 inode = ceph_find_inode(sb, vino);
133 if (!inode) {
134 struct ceph_mds_request *req;
135 int mask;
136
137 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
138 USE_ANY_MDS);
139 if (IS_ERR(req))
140 return ERR_CAST(req);
141
142 mask = CEPH_STAT_CAP_INODE;
143 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
144 mask |= CEPH_CAP_XATTR_SHARED;
145 req->r_args.lookupino.mask = cpu_to_le32(mask);
146
147 req->r_ino1 = vino;
148 req->r_num_caps = 1;
149 err = ceph_mdsc_do_request(mdsc, NULL, req);
150 inode = req->r_target_inode;
151 if (inode)
152 ihold(inode);
153 ceph_mdsc_put_request(req);
154 if (!inode)
155 return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
156 }
157 return inode;
158 }
159
ceph_lookup_inode(struct super_block * sb,u64 ino)160 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
161 {
162 struct inode *inode = __lookup_inode(sb, ino);
163 if (IS_ERR(inode))
164 return inode;
165 if (inode->i_nlink == 0) {
166 iput(inode);
167 return ERR_PTR(-ESTALE);
168 }
169 return inode;
170 }
171
__fh_to_dentry(struct super_block * sb,u64 ino)172 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
173 {
174 struct inode *inode = __lookup_inode(sb, ino);
175 int err;
176
177 if (IS_ERR(inode))
178 return ERR_CAST(inode);
179 /* We need LINK caps to reliably check i_nlink */
180 err = ceph_do_getattr(inode, CEPH_CAP_LINK_SHARED, false);
181 if (err)
182 return ERR_PTR(err);
183 /* -ESTALE if inode as been unlinked and no file is open */
184 if ((inode->i_nlink == 0) && (atomic_read(&inode->i_count) == 1)) {
185 iput(inode);
186 return ERR_PTR(-ESTALE);
187 }
188 return d_obtain_alias(inode);
189 }
190
__snapfh_to_dentry(struct super_block * sb,struct ceph_nfs_snapfh * sfh,bool want_parent)191 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
192 struct ceph_nfs_snapfh *sfh,
193 bool want_parent)
194 {
195 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
196 struct ceph_mds_request *req;
197 struct inode *inode;
198 struct ceph_vino vino;
199 int mask;
200 int err;
201 bool unlinked = false;
202
203 if (want_parent) {
204 vino.ino = sfh->parent_ino;
205 if (sfh->snapid == CEPH_SNAPDIR)
206 vino.snap = CEPH_NOSNAP;
207 else if (sfh->ino == sfh->parent_ino)
208 vino.snap = CEPH_SNAPDIR;
209 else
210 vino.snap = sfh->snapid;
211 } else {
212 vino.ino = sfh->ino;
213 vino.snap = sfh->snapid;
214 }
215 inode = ceph_find_inode(sb, vino);
216 if (inode)
217 return d_obtain_alias(inode);
218
219 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
220 USE_ANY_MDS);
221 if (IS_ERR(req))
222 return ERR_CAST(req);
223
224 mask = CEPH_STAT_CAP_INODE;
225 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
226 mask |= CEPH_CAP_XATTR_SHARED;
227 req->r_args.lookupino.mask = cpu_to_le32(mask);
228 if (vino.snap < CEPH_NOSNAP) {
229 req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
230 if (!want_parent && sfh->ino != sfh->parent_ino) {
231 req->r_args.lookupino.parent =
232 cpu_to_le64(sfh->parent_ino);
233 req->r_args.lookupino.hash =
234 cpu_to_le32(sfh->hash);
235 }
236 }
237
238 req->r_ino1 = vino;
239 req->r_num_caps = 1;
240 err = ceph_mdsc_do_request(mdsc, NULL, req);
241 inode = req->r_target_inode;
242 if (inode) {
243 if (vino.snap == CEPH_SNAPDIR) {
244 if (inode->i_nlink == 0)
245 unlinked = true;
246 inode = ceph_get_snapdir(inode);
247 } else if (ceph_snap(inode) == vino.snap) {
248 ihold(inode);
249 } else {
250 /* mds does not support lookup snapped inode */
251 err = -EOPNOTSUPP;
252 inode = NULL;
253 }
254 }
255 ceph_mdsc_put_request(req);
256
257 if (want_parent) {
258 dout("snapfh_to_parent %llx.%llx\n err=%d\n",
259 vino.ino, vino.snap, err);
260 } else {
261 dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
262 vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
263 }
264 if (!inode)
265 return ERR_PTR(-ESTALE);
266 /* see comments in ceph_get_parent() */
267 return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
268 }
269
270 /*
271 * convert regular fh to dentry
272 */
ceph_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)273 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
274 struct fid *fid,
275 int fh_len, int fh_type)
276 {
277 struct ceph_nfs_fh *fh = (void *)fid->raw;
278
279 if (fh_type == FILEID_BTRFS_WITH_PARENT) {
280 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
281 return __snapfh_to_dentry(sb, sfh, false);
282 }
283
284 if (fh_type != FILEID_INO32_GEN &&
285 fh_type != FILEID_INO32_GEN_PARENT)
286 return NULL;
287 if (fh_len < sizeof(*fh) / 4)
288 return NULL;
289
290 dout("fh_to_dentry %llx\n", fh->ino);
291 return __fh_to_dentry(sb, fh->ino);
292 }
293
__get_parent(struct super_block * sb,struct dentry * child,u64 ino)294 static struct dentry *__get_parent(struct super_block *sb,
295 struct dentry *child, u64 ino)
296 {
297 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
298 struct ceph_mds_request *req;
299 struct inode *inode;
300 int mask;
301 int err;
302
303 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
304 USE_ANY_MDS);
305 if (IS_ERR(req))
306 return ERR_CAST(req);
307
308 if (child) {
309 req->r_inode = d_inode(child);
310 ihold(d_inode(child));
311 } else {
312 req->r_ino1 = (struct ceph_vino) {
313 .ino = ino,
314 .snap = CEPH_NOSNAP,
315 };
316 }
317
318 mask = CEPH_STAT_CAP_INODE;
319 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
320 mask |= CEPH_CAP_XATTR_SHARED;
321 req->r_args.getattr.mask = cpu_to_le32(mask);
322
323 req->r_num_caps = 1;
324 err = ceph_mdsc_do_request(mdsc, NULL, req);
325 if (err) {
326 ceph_mdsc_put_request(req);
327 return ERR_PTR(err);
328 }
329
330 inode = req->r_target_inode;
331 if (inode)
332 ihold(inode);
333 ceph_mdsc_put_request(req);
334 if (!inode)
335 return ERR_PTR(-ENOENT);
336
337 return d_obtain_alias(inode);
338 }
339
ceph_get_parent(struct dentry * child)340 static struct dentry *ceph_get_parent(struct dentry *child)
341 {
342 struct inode *inode = d_inode(child);
343 struct dentry *dn;
344
345 if (ceph_snap(inode) != CEPH_NOSNAP) {
346 struct inode* dir;
347 bool unlinked = false;
348 /* do not support non-directory */
349 if (!d_is_dir(child)) {
350 dn = ERR_PTR(-EINVAL);
351 goto out;
352 }
353 dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
354 if (IS_ERR(dir)) {
355 dn = ERR_CAST(dir);
356 goto out;
357 }
358 /* There can be multiple paths to access snapped inode.
359 * For simplicity, treat snapdir of head inode as parent */
360 if (ceph_snap(inode) != CEPH_SNAPDIR) {
361 struct inode *snapdir = ceph_get_snapdir(dir);
362 if (dir->i_nlink == 0)
363 unlinked = true;
364 iput(dir);
365 if (IS_ERR(snapdir)) {
366 dn = ERR_CAST(snapdir);
367 goto out;
368 }
369 dir = snapdir;
370 }
371 /* If directory has already been deleted, futher get_parent
372 * will fail. Do not mark snapdir dentry as disconnected,
373 * this prevent exportfs from doing futher get_parent. */
374 if (unlinked)
375 dn = d_obtain_root(dir);
376 else
377 dn = d_obtain_alias(dir);
378 } else {
379 dn = __get_parent(child->d_sb, child, 0);
380 }
381 out:
382 dout("get_parent %p ino %llx.%llx err=%ld\n",
383 child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
384 return dn;
385 }
386
387 /*
388 * convert regular fh to parent
389 */
ceph_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)390 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
391 struct fid *fid,
392 int fh_len, int fh_type)
393 {
394 struct ceph_nfs_confh *cfh = (void *)fid->raw;
395 struct dentry *dentry;
396
397 if (fh_type == FILEID_BTRFS_WITH_PARENT) {
398 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
399 return __snapfh_to_dentry(sb, sfh, true);
400 }
401
402 if (fh_type != FILEID_INO32_GEN_PARENT)
403 return NULL;
404 if (fh_len < sizeof(*cfh) / 4)
405 return NULL;
406
407 dout("fh_to_parent %llx\n", cfh->parent_ino);
408 dentry = __get_parent(sb, NULL, cfh->ino);
409 if (unlikely(dentry == ERR_PTR(-ENOENT)))
410 dentry = __fh_to_dentry(sb, cfh->parent_ino);
411 return dentry;
412 }
413
__get_snap_name(struct dentry * parent,char * name,struct dentry * child)414 static int __get_snap_name(struct dentry *parent, char *name,
415 struct dentry *child)
416 {
417 struct inode *inode = d_inode(child);
418 struct inode *dir = d_inode(parent);
419 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
420 struct ceph_mds_request *req = NULL;
421 char *last_name = NULL;
422 unsigned next_offset = 2;
423 int err = -EINVAL;
424
425 if (ceph_ino(inode) != ceph_ino(dir))
426 goto out;
427 if (ceph_snap(inode) == CEPH_SNAPDIR) {
428 if (ceph_snap(dir) == CEPH_NOSNAP) {
429 strcpy(name, fsc->mount_options->snapdir_name);
430 err = 0;
431 }
432 goto out;
433 }
434 if (ceph_snap(dir) != CEPH_SNAPDIR)
435 goto out;
436
437 while (1) {
438 struct ceph_mds_reply_info_parsed *rinfo;
439 struct ceph_mds_reply_dir_entry *rde;
440 int i;
441
442 req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
443 USE_AUTH_MDS);
444 if (IS_ERR(req)) {
445 err = PTR_ERR(req);
446 req = NULL;
447 goto out;
448 }
449 err = ceph_alloc_readdir_reply_buffer(req, inode);
450 if (err)
451 goto out;
452
453 req->r_direct_mode = USE_AUTH_MDS;
454 req->r_readdir_offset = next_offset;
455 req->r_args.readdir.flags =
456 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
457 if (last_name) {
458 req->r_path2 = last_name;
459 last_name = NULL;
460 }
461
462 req->r_inode = dir;
463 ihold(dir);
464 req->r_dentry = dget(parent);
465
466 inode_lock(dir);
467 err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
468 inode_unlock(dir);
469
470 if (err < 0)
471 goto out;
472
473 rinfo = &req->r_reply_info;
474 for (i = 0; i < rinfo->dir_nr; i++) {
475 rde = rinfo->dir_entries + i;
476 BUG_ON(!rde->inode.in);
477 if (ceph_snap(inode) ==
478 le64_to_cpu(rde->inode.in->snapid)) {
479 memcpy(name, rde->name, rde->name_len);
480 name[rde->name_len] = '\0';
481 err = 0;
482 goto out;
483 }
484 }
485
486 if (rinfo->dir_end)
487 break;
488
489 BUG_ON(rinfo->dir_nr <= 0);
490 rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
491 next_offset += rinfo->dir_nr;
492 last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
493 if (!last_name) {
494 err = -ENOMEM;
495 goto out;
496 }
497
498 ceph_mdsc_put_request(req);
499 req = NULL;
500 }
501 err = -ENOENT;
502 out:
503 if (req)
504 ceph_mdsc_put_request(req);
505 kfree(last_name);
506 dout("get_snap_name %p ino %llx.%llx err=%d\n",
507 child, ceph_vinop(inode), err);
508 return err;
509 }
510
ceph_get_name(struct dentry * parent,char * name,struct dentry * child)511 static int ceph_get_name(struct dentry *parent, char *name,
512 struct dentry *child)
513 {
514 struct ceph_mds_client *mdsc;
515 struct ceph_mds_request *req;
516 struct inode *inode = d_inode(child);
517 int err;
518
519 if (ceph_snap(inode) != CEPH_NOSNAP)
520 return __get_snap_name(parent, name, child);
521
522 mdsc = ceph_inode_to_client(inode)->mdsc;
523 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
524 USE_ANY_MDS);
525 if (IS_ERR(req))
526 return PTR_ERR(req);
527
528 inode_lock(d_inode(parent));
529
530 req->r_inode = inode;
531 ihold(inode);
532 req->r_ino2 = ceph_vino(d_inode(parent));
533 req->r_parent = d_inode(parent);
534 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
535 req->r_num_caps = 2;
536 err = ceph_mdsc_do_request(mdsc, NULL, req);
537
538 inode_unlock(d_inode(parent));
539
540 if (!err) {
541 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
542 memcpy(name, rinfo->dname, rinfo->dname_len);
543 name[rinfo->dname_len] = 0;
544 dout("get_name %p ino %llx.%llx name %s\n",
545 child, ceph_vinop(inode), name);
546 } else {
547 dout("get_name %p ino %llx.%llx err %d\n",
548 child, ceph_vinop(inode), err);
549 }
550
551 ceph_mdsc_put_request(req);
552 return err;
553 }
554
555 const struct export_operations ceph_export_ops = {
556 .encode_fh = ceph_encode_fh,
557 .fh_to_dentry = ceph_fh_to_dentry,
558 .fh_to_parent = ceph_fh_to_parent,
559 .get_parent = ceph_get_parent,
560 .get_name = ceph_get_name,
561 };
562