Loading...
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 */
14struct ceph_nfs_fh {
15 u64 ino;
16} __attribute__ ((packed));
17
18/*
19 * Larger fh that includes parent ino.
20 */
21struct ceph_nfs_confh {
22 u64 ino, parent_ino;
23} __attribute__ ((packed));
24
25static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
26 struct inode *parent_inode)
27{
28 int type;
29 struct ceph_nfs_fh *fh = (void *)rawfh;
30 struct ceph_nfs_confh *cfh = (void *)rawfh;
31 int connected_handle_length = sizeof(*cfh)/4;
32 int handle_length = sizeof(*fh)/4;
33
34 /* don't re-export snaps */
35 if (ceph_snap(inode) != CEPH_NOSNAP)
36 return -EINVAL;
37
38 if (parent_inode && (*max_len < connected_handle_length)) {
39 *max_len = connected_handle_length;
40 return FILEID_INVALID;
41 } else if (*max_len < handle_length) {
42 *max_len = handle_length;
43 return FILEID_INVALID;
44 }
45
46 if (parent_inode) {
47 dout("encode_fh %llx with parent %llx\n",
48 ceph_ino(inode), ceph_ino(parent_inode));
49 cfh->ino = ceph_ino(inode);
50 cfh->parent_ino = ceph_ino(parent_inode);
51 *max_len = connected_handle_length;
52 type = FILEID_INO32_GEN_PARENT;
53 } else {
54 dout("encode_fh %llx\n", ceph_ino(inode));
55 fh->ino = ceph_ino(inode);
56 *max_len = handle_length;
57 type = FILEID_INO32_GEN;
58 }
59 return type;
60}
61
62static struct dentry *__fh_to_dentry(struct super_block *sb, u64 ino)
63{
64 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
65 struct inode *inode;
66 struct ceph_vino vino;
67 int err;
68
69 vino.ino = ino;
70 vino.snap = CEPH_NOSNAP;
71 inode = ceph_find_inode(sb, vino);
72 if (!inode) {
73 struct ceph_mds_request *req;
74 int mask;
75
76 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
77 USE_ANY_MDS);
78 if (IS_ERR(req))
79 return ERR_CAST(req);
80
81 mask = CEPH_STAT_CAP_INODE;
82 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
83 mask |= CEPH_CAP_XATTR_SHARED;
84 req->r_args.getattr.mask = cpu_to_le32(mask);
85
86 req->r_ino1 = vino;
87 req->r_num_caps = 1;
88 err = ceph_mdsc_do_request(mdsc, NULL, req);
89 inode = req->r_target_inode;
90 if (inode)
91 ihold(inode);
92 ceph_mdsc_put_request(req);
93 if (!inode)
94 return ERR_PTR(-ESTALE);
95 if (inode->i_nlink == 0) {
96 iput(inode);
97 return ERR_PTR(-ESTALE);
98 }
99 }
100
101 return d_obtain_alias(inode);
102}
103
104/*
105 * convert regular fh to dentry
106 */
107static struct dentry *ceph_fh_to_dentry(struct super_block *sb,
108 struct fid *fid,
109 int fh_len, int fh_type)
110{
111 struct ceph_nfs_fh *fh = (void *)fid->raw;
112
113 if (fh_type != FILEID_INO32_GEN &&
114 fh_type != FILEID_INO32_GEN_PARENT)
115 return NULL;
116 if (fh_len < sizeof(*fh) / 4)
117 return NULL;
118
119 dout("fh_to_dentry %llx\n", fh->ino);
120 return __fh_to_dentry(sb, fh->ino);
121}
122
123static struct dentry *__get_parent(struct super_block *sb,
124 struct dentry *child, u64 ino)
125{
126 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
127 struct ceph_mds_request *req;
128 struct inode *inode;
129 int mask;
130 int err;
131
132 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPPARENT,
133 USE_ANY_MDS);
134 if (IS_ERR(req))
135 return ERR_CAST(req);
136
137 if (child) {
138 req->r_inode = d_inode(child);
139 ihold(d_inode(child));
140 } else {
141 req->r_ino1 = (struct ceph_vino) {
142 .ino = ino,
143 .snap = CEPH_NOSNAP,
144 };
145 }
146
147 mask = CEPH_STAT_CAP_INODE;
148 if (ceph_security_xattr_wanted(d_inode(sb->s_root)))
149 mask |= CEPH_CAP_XATTR_SHARED;
150 req->r_args.getattr.mask = cpu_to_le32(mask);
151
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_PTR(-ENOENT);
160
161 return d_obtain_alias(inode);
162}
163
164static struct dentry *ceph_get_parent(struct dentry *child)
165{
166 /* don't re-export snaps */
167 if (ceph_snap(d_inode(child)) != CEPH_NOSNAP)
168 return ERR_PTR(-EINVAL);
169
170 dout("get_parent %p ino %llx.%llx\n",
171 child, ceph_vinop(d_inode(child)));
172 return __get_parent(child->d_sb, child, 0);
173}
174
175/*
176 * convert regular fh to parent
177 */
178static struct dentry *ceph_fh_to_parent(struct super_block *sb,
179 struct fid *fid,
180 int fh_len, int fh_type)
181{
182 struct ceph_nfs_confh *cfh = (void *)fid->raw;
183 struct dentry *dentry;
184
185 if (fh_type != FILEID_INO32_GEN_PARENT)
186 return NULL;
187 if (fh_len < sizeof(*cfh) / 4)
188 return NULL;
189
190 dout("fh_to_parent %llx\n", cfh->parent_ino);
191 dentry = __get_parent(sb, NULL, cfh->ino);
192 if (unlikely(dentry == ERR_PTR(-ENOENT)))
193 dentry = __fh_to_dentry(sb, cfh->parent_ino);
194 return dentry;
195}
196
197static int ceph_get_name(struct dentry *parent, char *name,
198 struct dentry *child)
199{
200 struct ceph_mds_client *mdsc;
201 struct ceph_mds_request *req;
202 int err;
203
204 mdsc = ceph_inode_to_client(d_inode(child))->mdsc;
205 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPNAME,
206 USE_ANY_MDS);
207 if (IS_ERR(req))
208 return PTR_ERR(req);
209
210 inode_lock(d_inode(parent));
211
212 req->r_inode = d_inode(child);
213 ihold(d_inode(child));
214 req->r_ino2 = ceph_vino(d_inode(parent));
215 req->r_parent = d_inode(parent);
216 set_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags);
217 req->r_num_caps = 2;
218 err = ceph_mdsc_do_request(mdsc, NULL, req);
219
220 inode_unlock(d_inode(parent));
221
222 if (!err) {
223 struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
224 memcpy(name, rinfo->dname, rinfo->dname_len);
225 name[rinfo->dname_len] = 0;
226 dout("get_name %p ino %llx.%llx name %s\n",
227 child, ceph_vinop(d_inode(child)), name);
228 } else {
229 dout("get_name %p ino %llx.%llx err %d\n",
230 child, ceph_vinop(d_inode(child)), err);
231 }
232
233 ceph_mdsc_put_request(req);
234 return err;
235}
236
237const struct export_operations ceph_export_ops = {
238 .encode_fh = ceph_encode_fh,
239 .fh_to_dentry = ceph_fh_to_dentry,
240 .fh_to_parent = ceph_fh_to_parent,
241 .get_parent = ceph_get_parent,
242 .get_name = ceph_get_name,
243};
1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/exportfs.h>
4#include <linux/slab.h>
5#include <asm/unaligned.h>
6
7#include "super.h"
8#include "mds_client.h"
9
10/*
11 * NFS export support
12 *
13 * NFS re-export of a ceph mount is, at present, only semireliable.
14 * The basic issue is that the Ceph architectures doesn't lend itself
15 * well to generating filehandles that will remain valid forever.
16 *
17 * So, we do our best. If you're lucky, your inode will be in the
18 * client's cache. If it's not, and you have a connectable fh, then
19 * the MDS server may be able to find it for you. Otherwise, you get
20 * ESTALE.
21 *
22 * There are ways to this more reliable, but in the non-connectable fh
23 * case, we won't every work perfectly, and in the connectable case,
24 * some changes are needed on the MDS side to work better.
25 */
26
27/*
28 * Basic fh
29 */
30struct ceph_nfs_fh {
31 u64 ino;
32} __attribute__ ((packed));
33
34/*
35 * Larger 'connectable' fh that includes parent ino and name hash.
36 * Use this whenever possible, as it works more reliably.
37 */
38struct ceph_nfs_confh {
39 u64 ino, parent_ino;
40 u32 parent_name_hash;
41} __attribute__ ((packed));
42
43/*
44 * The presence of @parent_inode here tells us whether NFS wants a
45 * connectable file handle. However, we want to make a connectionable
46 * file handle unconditionally so that the MDS gets as much of a hint
47 * as possible. That means we only use @parent_dentry to indicate
48 * whether nfsd wants a connectable fh, and whether we should indicate
49 * failure from a too-small @max_len.
50 */
51static int ceph_encode_fh(struct inode *inode, u32 *rawfh, int *max_len,
52 struct inode *parent_inode)
53{
54 int type;
55 struct ceph_nfs_fh *fh = (void *)rawfh;
56 struct ceph_nfs_confh *cfh = (void *)rawfh;
57 int connected_handle_length = sizeof(*cfh)/4;
58 int handle_length = sizeof(*fh)/4;
59 struct dentry *dentry = d_find_alias(inode);
60 struct dentry *parent;
61
62 /* don't re-export snaps */
63 if (ceph_snap(inode) != CEPH_NOSNAP)
64 return -EINVAL;
65
66 /* if we found an alias, generate a connectable fh */
67 if (*max_len >= connected_handle_length && dentry) {
68 dout("encode_fh %p connectable\n", dentry);
69 spin_lock(&dentry->d_lock);
70 parent = dentry->d_parent;
71 cfh->ino = ceph_ino(inode);
72 cfh->parent_ino = ceph_ino(parent->d_inode);
73 cfh->parent_name_hash = ceph_dentry_hash(parent->d_inode,
74 dentry);
75 *max_len = connected_handle_length;
76 type = 2;
77 spin_unlock(&dentry->d_lock);
78 } else if (*max_len >= handle_length) {
79 if (parent_inode) {
80 /* nfsd wants connectable */
81 *max_len = connected_handle_length;
82 type = 255;
83 } else {
84 dout("encode_fh %p\n", dentry);
85 fh->ino = ceph_ino(inode);
86 *max_len = handle_length;
87 type = 1;
88 }
89 } else {
90 *max_len = handle_length;
91 type = 255;
92 }
93 return type;
94}
95
96/*
97 * convert regular fh to dentry
98 *
99 * FIXME: we should try harder by querying the mds for the ino.
100 */
101static struct dentry *__fh_to_dentry(struct super_block *sb,
102 struct ceph_nfs_fh *fh)
103{
104 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
105 struct inode *inode;
106 struct dentry *dentry;
107 struct ceph_vino vino;
108 int err;
109
110 dout("__fh_to_dentry %llx\n", fh->ino);
111 vino.ino = fh->ino;
112 vino.snap = CEPH_NOSNAP;
113 inode = ceph_find_inode(sb, vino);
114 if (!inode) {
115 struct ceph_mds_request *req;
116
117 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
118 USE_ANY_MDS);
119 if (IS_ERR(req))
120 return ERR_CAST(req);
121
122 req->r_ino1 = vino;
123 req->r_num_caps = 1;
124 err = ceph_mdsc_do_request(mdsc, NULL, req);
125 inode = req->r_target_inode;
126 if (inode)
127 ihold(inode);
128 ceph_mdsc_put_request(req);
129 if (!inode)
130 return ERR_PTR(-ESTALE);
131 }
132
133 dentry = d_obtain_alias(inode);
134 if (IS_ERR(dentry)) {
135 pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n",
136 fh->ino, inode);
137 iput(inode);
138 return dentry;
139 }
140 err = ceph_init_dentry(dentry);
141 if (err < 0) {
142 iput(inode);
143 return ERR_PTR(err);
144 }
145 dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry);
146 return dentry;
147}
148
149/*
150 * convert connectable fh to dentry
151 */
152static struct dentry *__cfh_to_dentry(struct super_block *sb,
153 struct ceph_nfs_confh *cfh)
154{
155 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
156 struct inode *inode;
157 struct dentry *dentry;
158 struct ceph_vino vino;
159 int err;
160
161 dout("__cfh_to_dentry %llx (%llx/%x)\n",
162 cfh->ino, cfh->parent_ino, cfh->parent_name_hash);
163
164 vino.ino = cfh->ino;
165 vino.snap = CEPH_NOSNAP;
166 inode = ceph_find_inode(sb, vino);
167 if (!inode) {
168 struct ceph_mds_request *req;
169
170 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH,
171 USE_ANY_MDS);
172 if (IS_ERR(req))
173 return ERR_CAST(req);
174
175 req->r_ino1 = vino;
176 req->r_ino2.ino = cfh->parent_ino;
177 req->r_ino2.snap = CEPH_NOSNAP;
178 req->r_path2 = kmalloc(16, GFP_NOFS);
179 snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash);
180 req->r_num_caps = 1;
181 err = ceph_mdsc_do_request(mdsc, NULL, req);
182 inode = req->r_target_inode;
183 if (inode)
184 ihold(inode);
185 ceph_mdsc_put_request(req);
186 if (!inode)
187 return ERR_PTR(err ? err : -ESTALE);
188 }
189
190 dentry = d_obtain_alias(inode);
191 if (IS_ERR(dentry)) {
192 pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n",
193 cfh->ino, inode);
194 iput(inode);
195 return dentry;
196 }
197 err = ceph_init_dentry(dentry);
198 if (err < 0) {
199 iput(inode);
200 return ERR_PTR(err);
201 }
202 dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry);
203 return dentry;
204}
205
206static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid,
207 int fh_len, int fh_type)
208{
209 if (fh_type == 1)
210 return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw);
211 else
212 return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw);
213}
214
215/*
216 * get parent, if possible.
217 *
218 * FIXME: we could do better by querying the mds to discover the
219 * parent.
220 */
221static struct dentry *ceph_fh_to_parent(struct super_block *sb,
222 struct fid *fid,
223 int fh_len, int fh_type)
224{
225 struct ceph_nfs_confh *cfh = (void *)fid->raw;
226 struct ceph_vino vino;
227 struct inode *inode;
228 struct dentry *dentry;
229 int err;
230
231 if (fh_type == 1)
232 return ERR_PTR(-ESTALE);
233
234 pr_debug("fh_to_parent %llx/%d\n", cfh->parent_ino,
235 cfh->parent_name_hash);
236
237 vino.ino = cfh->ino;
238 vino.snap = CEPH_NOSNAP;
239 inode = ceph_find_inode(sb, vino);
240 if (!inode)
241 return ERR_PTR(-ESTALE);
242
243 dentry = d_obtain_alias(inode);
244 if (IS_ERR(dentry)) {
245 pr_err("fh_to_parent %llx -- inode %p but ENOMEM\n",
246 cfh->ino, inode);
247 iput(inode);
248 return dentry;
249 }
250 err = ceph_init_dentry(dentry);
251 if (err < 0) {
252 iput(inode);
253 return ERR_PTR(err);
254 }
255 dout("fh_to_parent %llx %p dentry %p\n", cfh->ino, inode, dentry);
256 return dentry;
257}
258
259const struct export_operations ceph_export_ops = {
260 .encode_fh = ceph_encode_fh,
261 .fh_to_dentry = ceph_fh_to_dentry,
262 .fh_to_parent = ceph_fh_to_parent,
263};