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;
133 if (ceph_vino_is_reserved(vino))
134 return ERR_PTR(-ESTALE);
136 inode = ceph_find_inode(sb, vino);
138 struct ceph_mds_request *req;
141 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
144 return ERR_CAST(req);
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);
153 err = ceph_mdsc_do_request(mdsc, NULL, req);
154 inode = req->r_target_inode;
157 ceph_mdsc_put_request(req);
159 return err < 0 ? ERR_PTR(err) : ERR_PTR(-ESTALE);
161 if (ceph_inode_is_shutdown(inode)) {
163 return ERR_PTR(-ESTALE);
169 struct inode *ceph_lookup_inode(struct super_block *sb, u64 ino)
171 struct inode *inode = __lookup_inode(sb, ino);
174 if (inode->i_nlink == 0) {
176 return ERR_PTR(-ESTALE);
181 static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
183 struct inode *inode = __lookup_inode(sb, ino);
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);
194 /* -ESTALE if inode as been unlinked and no file is open */
195 if ((inode->i_nlink == 0) && (atomic_read(&inode->i_count) == 1)) {
197 return ERR_PTR(-ESTALE);
199 return d_obtain_alias(inode);
202 static struct dentry *__snapfh_to_dentry(struct super_block *sb,
203 struct ceph_nfs_snapfh *sfh,
206 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
207 struct ceph_mds_request *req;
209 struct ceph_vino vino;
212 bool unlinked = false;
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;
221 vino.snap = sfh->snapid;
224 vino.snap = sfh->snapid;
227 if (ceph_vino_is_reserved(vino))
228 return ERR_PTR(-ESTALE);
230 inode = ceph_find_inode(sb, vino);
232 if (ceph_inode_is_shutdown(inode)) {
234 return ERR_PTR(-ESTALE);
236 return d_obtain_alias(inode);
239 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
242 return ERR_CAST(req);
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);
260 err = ceph_mdsc_do_request(mdsc, NULL, req);
261 inode = req->r_target_inode;
263 if (vino.snap == CEPH_SNAPDIR) {
264 if (inode->i_nlink == 0)
266 inode = ceph_get_snapdir(inode);
267 } else if (ceph_snap(inode) == vino.snap) {
270 /* mds does not support lookup snapped inode */
271 inode = ERR_PTR(-EOPNOTSUPP);
274 inode = ERR_PTR(-ESTALE);
276 ceph_mdsc_put_request(req);
279 dout("snapfh_to_parent %llx.%llx\n err=%d\n",
280 vino.ino, vino.snap, err);
282 dout("snapfh_to_dentry %llx.%llx parent %llx hash %x err=%d",
283 vino.ino, vino.snap, sfh->parent_ino, sfh->hash, err);
286 return ERR_CAST(inode);
287 /* see comments in ceph_get_parent() */
288 return unlinked ? d_obtain_root(inode) : d_obtain_alias(inode);
292 * convert regular fh to dentry
294 static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
296 int fh_len, int fh_type)
298 struct ceph_nfs_fh *fh = (void *)fid->raw;
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);
305 if (fh_type != FILEID_INO32_GEN &&
306 fh_type != FILEID_INO32_GEN_PARENT)
308 if (fh_len < sizeof(*fh) / 4)
311 dout("fh_to_dentry %llx\n", fh->ino);
312 return __fh_to_dentry(sb, fh->ino);
315 static struct dentry *__get_parent(struct super_block *sb,
316 struct dentry *child, u64 ino)
318 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
319 struct ceph_mds_request *req;
324 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
327 return ERR_CAST(req);
330 req->r_inode = d_inode(child);
331 ihold(d_inode(child));
333 req->r_ino1 = (struct ceph_vino) {
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);
345 err = ceph_mdsc_do_request(mdsc, NULL, req);
347 ceph_mdsc_put_request(req);
351 inode = req->r_target_inode;
354 ceph_mdsc_put_request(req);
356 return ERR_PTR(-ENOENT);
358 return d_obtain_alias(inode);
361 static struct dentry *ceph_get_parent(struct dentry *child)
363 struct inode *inode = d_inode(child);
366 if (ceph_snap(inode) != CEPH_NOSNAP) {
368 bool unlinked = false;
369 /* do not support non-directory */
370 if (!d_is_dir(child)) {
371 dn = ERR_PTR(-EINVAL);
374 dir = __lookup_inode(inode->i_sb, ceph_ino(inode));
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)
386 if (IS_ERR(snapdir)) {
387 dn = ERR_CAST(snapdir);
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. */
396 dn = d_obtain_root(dir);
398 dn = d_obtain_alias(dir);
400 dn = __get_parent(child->d_sb, child, 0);
403 dout("get_parent %p ino %llx.%llx err=%ld\n",
404 child, ceph_vinop(inode), (long)PTR_ERR_OR_ZERO(dn));
409 * convert regular fh to parent
411 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
413 int fh_len, int fh_type)
415 struct ceph_nfs_confh *cfh = (void *)fid->raw;
416 struct dentry *dentry;
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);
423 if (fh_type != FILEID_INO32_GEN_PARENT)
425 if (fh_len < sizeof(*cfh) / 4)
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);
435 static int __get_snap_name(struct dentry *parent, char *name,
436 struct dentry *child)
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;
446 if (ceph_ino(inode) != ceph_ino(dir))
448 if (ceph_snap(inode) == CEPH_SNAPDIR) {
449 if (ceph_snap(dir) == CEPH_NOSNAP) {
450 strcpy(name, fsc->mount_options->snapdir_name);
455 if (ceph_snap(dir) != CEPH_SNAPDIR)
459 struct ceph_mds_reply_info_parsed *rinfo;
460 struct ceph_mds_reply_dir_entry *rde;
463 req = ceph_mdsc_create_request(fsc->mdsc, CEPH_MDS_OP_LSSNAP,
470 err = ceph_alloc_readdir_reply_buffer(req, inode);
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);
479 req->r_path2 = last_name;
485 req->r_dentry = dget(parent);
488 err = ceph_mdsc_do_request(fsc->mdsc, NULL, req);
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';
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);
519 ceph_mdsc_put_request(req);
525 ceph_mdsc_put_request(req);
527 dout("get_snap_name %p ino %llx.%llx err=%d\n",
528 child, ceph_vinop(inode), err);
532 static int ceph_get_name(struct dentry *parent, char *name,
533 struct dentry *child)
535 struct ceph_mds_client *mdsc;
536 struct ceph_mds_request *req;
537 struct inode *inode = d_inode(child);
540 if (ceph_snap(inode) != CEPH_NOSNAP)
541 return __get_snap_name(parent, name, child);
543 mdsc = ceph_inode_to_client(inode)->mdsc;
544 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
549 inode_lock(d_inode(parent));
551 req->r_inode = 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);
558 err = ceph_mdsc_do_request(mdsc, NULL, req);
560 inode_unlock(d_inode(parent));
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);
569 dout("get_name %p ino %llx.%llx err %d\n",
570 child, ceph_vinop(inode), err);
573 ceph_mdsc_put_request(req);
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,