1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/exportfs.h>
5 #include <linux/slab.h>
6 #include <asm/unaligned.h>
9 #include "mds_client.h"
16 } __attribute__ ((packed));
19 * Larger fh that includes parent ino.
21 struct ceph_nfs_confh {
23 } __attribute__ ((packed));
26 * fh for snapped inode
28 struct ceph_nfs_snapfh {
33 } __attribute__ ((packed));
35 static int ceph_encode_snapfh(struct inode *inode, u32 *rawfh, int *max_len,
36 struct inode *parent_inode)
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);
43 bool no_parent = true;
45 if (*max_len < snap_handle_length) {
46 *max_len = snap_handle_length;
52 if (snapid != CEPH_SNAPDIR) {
54 struct dentry *dentry = d_find_alias(inode);
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);
70 if (!S_ISDIR(inode->i_mode))
72 sfh->parent_ino = sfh->ino;
75 sfh->ino = ceph_ino(inode);
78 *max_len = snap_handle_length;
79 ret = FILEID_BTRFS_WITH_PARENT;
81 dout("encode_snapfh %llx.%llx ret=%d\n", ceph_vinop(inode), ret);
85 static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
86 struct inode *parent_inode)
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;
94 if (ceph_snap(inode) != CEPH_NOSNAP)
95 return ceph_encode_snapfh(inode, rawfh, max_len, parent_inode);
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;
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;
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;
123 static struct inode *__lookup_inode(struct super_block *sb, u64 ino)
125 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
127 struct ceph_vino vino;
131 vino.snap = CEPH_NOSNAP;
132 inode = ceph_find_inode(sb, vino);
134 struct ceph_mds_request *req;
137 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
140 return ERR_CAST(req);
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);
149 err = ceph_mdsc_do_request(mdsc, NULL, req);
150 inode = req->r_target_inode;
153 ceph_mdsc_put_request(req);
155 return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
160 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
162 struct inode *inode = __lookup_inode(sb, ino);
165 if (inode->i_nlink == 0) {
167 return ERR_PTR(-ESTALE);
172 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
174 struct inode *inode = __lookup_inode(sb, ino);
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);
185 /* -ESTALE if inode as been unlinked and no file is open */
186 if ((inode->i_nlink == 0) && (atomic_read(&inode->i_count) == 1)) {
188 return ERR_PTR(-ESTALE);
190 return d_obtain_alias(inode);
193 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
194 struct ceph_nfs_snapfh *sfh,
197 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
198 struct ceph_mds_request *req;
200 struct ceph_vino vino;
203 bool unlinked = false;
206 vino.ino = sfh->parent_ino;
207 if (sfh->snapid == CEPH_SNAPDIR)
208 vino.snap = CEPH_NOSNAP;
209 else if (sfh->ino == sfh->parent_ino)
210 vino.snap = CEPH_SNAPDIR;
212 vino.snap = sfh->snapid;
215 vino.snap = sfh->snapid;
217 inode = ceph_find_inode(sb, vino);
219 return d_obtain_alias(inode);
221 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
224 return ERR_CAST(req);
226 mask = CEPH_STAT_CAP_INODE;
227 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
228 mask |= CEPH_CAP_XATTR_SHARED;
229 req->r_args.lookupino.mask = cpu_to_le32(mask);
230 if (vino.snap < CEPH_NOSNAP) {
231 req->r_args.lookupino.snapid = cpu_to_le64(vino.snap);
232 if (!want_parent && sfh->ino != sfh->parent_ino) {
233 req->r_args.lookupino.parent =
234 cpu_to_le64(sfh->parent_ino);
235 req->r_args.lookupino.hash =
236 cpu_to_le32(sfh->hash);
242 err = ceph_mdsc_do_request(mdsc, NULL, req);
243 inode = req->r_target_inode;
245 if (vino.snap == CEPH_SNAPDIR) {
246 if (inode->i_nlink == 0)
248 inode = ceph_get_snapdir(inode);
249 } else if (ceph_snap(inode) == vino.snap) {
252 /* mds does not support lookup snapped inode */
257 ceph_mdsc_put_request(req);
260 dout("snapfh_to_parent %llx.%llx\n err=%d\n",
261 vino.ino, vino.snap, err);
263 dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
264 vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
267 return ERR_PTR(-ESTALE);
268 /* see comments in ceph_get_parent() */
269 return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
273 * convert regular fh to dentry
275 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
277 int fh_len, int fh_type)
279 struct ceph_nfs_fh *fh = (void *)fid->raw;
281 if (fh_type == FILEID_BTRFS_WITH_PARENT) {
282 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
283 return __snapfh_to_dentry(sb, sfh, false);
286 if (fh_type != FILEID_INO32_GEN &&
287 fh_type != FILEID_INO32_GEN_PARENT)
289 if (fh_len < sizeof(*fh) / 4)
292 dout("fh_to_dentry %llx\n", fh->ino);
293 return __fh_to_dentry(sb, fh->ino);
296 static struct dentry *__get_parent(struct super_block *sb,
297 struct dentry *child, u64 ino)
299 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
300 struct ceph_mds_request *req;
305 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
308 return ERR_CAST(req);
311 req->r_inode = d_inode(child);
312 ihold(d_inode(child));
314 req->r_ino1 = (struct ceph_vino) {
320 mask = CEPH_STAT_CAP_INODE;
321 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
322 mask |= CEPH_CAP_XATTR_SHARED;
323 req->r_args.getattr.mask = cpu_to_le32(mask);
326 err = ceph_mdsc_do_request(mdsc, NULL, req);
328 ceph_mdsc_put_request(req);
332 inode = req->r_target_inode;
335 ceph_mdsc_put_request(req);
337 return ERR_PTR(-ENOENT);
339 return d_obtain_alias(inode);
342 static struct dentry *ceph_get_parent(struct dentry *child)
344 struct inode *inode = d_inode(child);
347 if (ceph_snap(inode) != CEPH_NOSNAP) {
349 bool unlinked = false;
350 /* do not support non-directory */
351 if (!d_is_dir(child)) {
352 dn = ERR_PTR(-EINVAL);
355 dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
360 /* There can be multiple paths to access snapped inode.
361 * For simplicity, treat snapdir of head inode as parent */
362 if (ceph_snap(inode) != CEPH_SNAPDIR) {
363 struct inode *snapdir = ceph_get_snapdir(dir);
364 if (dir->i_nlink == 0)
367 if (IS_ERR(snapdir)) {
368 dn = ERR_CAST(snapdir);
373 /* If directory has already been deleted, futher get_parent
374 * will fail. Do not mark snapdir dentry as disconnected,
375 * this prevent exportfs from doing futher get_parent. */
377 dn = d_obtain_root(dir);
379 dn = d_obtain_alias(dir);
381 dn = __get_parent(child->d_sb, child, 0);
384 dout("get_parent %p ino %llx.%llx err=%ld\n",
385 child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
390 * convert regular fh to parent
392 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
394 int fh_len, int fh_type)
396 struct ceph_nfs_confh *cfh = (void *)fid->raw;
397 struct dentry *dentry;
399 if (fh_type == FILEID_BTRFS_WITH_PARENT) {
400 struct ceph_nfs_snapfh *sfh = (void *)fid->raw;
401 return __snapfh_to_dentry(sb, sfh, true);
404 if (fh_type != FILEID_INO32_GEN_PARENT)
406 if (fh_len < sizeof(*cfh) / 4)
409 dout("fh_to_parent %llx\n", cfh->parent_ino);
410 dentry = __get_parent(sb, NULL, cfh->ino);
411 if (unlikely(dentry == ERR_PTR(-ENOENT)))
412 dentry = __fh_to_dentry(sb, cfh->parent_ino);
416 static int __get_snap_name(struct dentry *parent, char *name,
417 struct dentry *child)
419 struct inode *inode = d_inode(child);
420 struct inode *dir = d_inode(parent);
421 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
422 struct ceph_mds_request *req = NULL;
423 char *last_name = NULL;
424 unsigned next_offset = 2;
427 if (ceph_ino(inode) != ceph_ino(dir))
429 if (ceph_snap(inode) == CEPH_SNAPDIR) {
430 if (ceph_snap(dir) == CEPH_NOSNAP) {
431 strcpy(name, fsc->mount_options->snapdir_name);
436 if (ceph_snap(dir) != CEPH_SNAPDIR)
440 struct ceph_mds_reply_info_parsed *rinfo;
441 struct ceph_mds_reply_dir_entry *rde;
444 req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
451 err = ceph_alloc_readdir_reply_buffer(req, inode);
455 req->r_direct_mode = USE_AUTH_MDS;
456 req->r_readdir_offset = next_offset;
457 req->r_args.readdir.flags =
458 cpu_to_le16(CEPH_READDIR_REPLY_BITFLAGS);
460 req->r_path2 = last_name;
466 req->r_dentry = dget(parent);
469 err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
475 rinfo = &req->r_reply_info;
476 for (i = 0; i < rinfo->dir_nr; i++) {
477 rde = rinfo->dir_entries + i;
478 BUG_ON(!rde->inode.in);
479 if (ceph_snap(inode) ==
480 le64_to_cpu(rde->inode.in->snapid)) {
481 memcpy(name, rde->name, rde->name_len);
482 name[rde->name_len] = '\0';
491 BUG_ON(rinfo->dir_nr <= 0);
492 rde = rinfo->dir_entries + (rinfo->dir_nr - 1);
493 next_offset += rinfo->dir_nr;
494 last_name = kstrndup(rde->name, rde->name_len, GFP_KERNEL);
500 ceph_mdsc_put_request(req);
506 ceph_mdsc_put_request(req);
508 dout("get_snap_name %p ino %llx.%llx err=%d\n",
509 child, ceph_vinop(inode), err);
513 static int ceph_get_name(struct dentry *parent, char *name,
514 struct dentry *child)
516 struct ceph_mds_client *mdsc;
517 struct ceph_mds_request *req;
518 struct inode *inode = d_inode(child);
521 if (ceph_snap(inode) != CEPH_NOSNAP)
522 return __get_snap_name(parent, name, child);
524 mdsc = ceph_inode_to_client(inode)->mdsc;
525 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
530 inode_lock(d_inode(parent));
532 req->r_inode = inode;
534 req->r_ino2 = ceph_vino(d_inode(parent));
535 req->r_parent = d_inode(parent);
536 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
538 err = ceph_mdsc_do_request(mdsc, NULL, req);
540 inode_unlock(d_inode(parent));
543 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
544 memcpy(name, rinfo->dname, rinfo->dname_len);
545 name[rinfo->dname_len] = 0;
546 dout("get_name %p ino %llx.%llx name %s\n",
547 child, ceph_vinop(inode), name);
549 dout("get_name %p ino %llx.%llx err %d\n",
550 child, ceph_vinop(inode), err);
553 ceph_mdsc_put_request(req);
557 const struct export_operations ceph_export_ops = {
558 .encode_fh = ceph_encode_fh,
559 .fh_to_dentry = ceph_fh_to_dentry,
560 .fh_to_parent = ceph_fh_to_parent,
561 .get_parent = ceph_get_parent,
562 .get_name = ceph_get_name,