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