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 */
30 struct 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 */
38 struct ceph_nfs_confh {
39 u64 ino, parent_ino;
40 u32 parent_name_hash;
41 } __attribute__ ((packed));
42
ceph_encode_fh(struct dentry * dentry,u32 * rawfh,int * max_len,int connectable)43 static int ceph_encode_fh(struct dentry *dentry, u32 *rawfh, int *max_len,
44 int connectable)
45 {
46 int type;
47 struct ceph_nfs_fh *fh = (void *)rawfh;
48 struct ceph_nfs_confh *cfh = (void *)rawfh;
49 struct dentry *parent;
50 struct inode *inode = dentry->d_inode;
51 int connected_handle_length = sizeof(*cfh)/4;
52 int handle_length = sizeof(*fh)/4;
53
54 /* don't re-export snaps */
55 if (ceph_snap(inode) != CEPH_NOSNAP)
56 return -EINVAL;
57
58 spin_lock(&dentry->d_lock);
59 parent = dentry->d_parent;
60 if (*max_len >= connected_handle_length) {
61 dout("encode_fh %p connectable\n", dentry);
62 cfh->ino = ceph_ino(dentry->d_inode);
63 cfh->parent_ino = ceph_ino(parent->d_inode);
64 cfh->parent_name_hash = ceph_dentry_hash(parent->d_inode,
65 dentry);
66 *max_len = connected_handle_length;
67 type = 2;
68 } else if (*max_len >= handle_length) {
69 if (connectable) {
70 *max_len = connected_handle_length;
71 type = 255;
72 } else {
73 dout("encode_fh %p\n", dentry);
74 fh->ino = ceph_ino(dentry->d_inode);
75 *max_len = handle_length;
76 type = 1;
77 }
78 } else {
79 *max_len = handle_length;
80 type = 255;
81 }
82 spin_unlock(&dentry->d_lock);
83 return type;
84 }
85
86 /*
87 * convert regular fh to dentry
88 *
89 * FIXME: we should try harder by querying the mds for the ino.
90 */
__fh_to_dentry(struct super_block * sb,struct ceph_nfs_fh * fh,int fh_len)91 static struct dentry *__fh_to_dentry(struct super_block *sb,
92 struct ceph_nfs_fh *fh, int fh_len)
93 {
94 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
95 struct inode *inode;
96 struct dentry *dentry;
97 struct ceph_vino vino;
98 int err;
99
100 if (fh_len < sizeof(*fh) / 4)
101 return ERR_PTR(-ESTALE);
102
103 dout("__fh_to_dentry %llx\n", fh->ino);
104 vino.ino = fh->ino;
105 vino.snap = CEPH_NOSNAP;
106 inode = ceph_find_inode(sb, vino);
107 if (!inode) {
108 struct ceph_mds_request *req;
109
110 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
111 USE_ANY_MDS);
112 if (IS_ERR(req))
113 return ERR_CAST(req);
114
115 req->r_ino1 = vino;
116 req->r_num_caps = 1;
117 err = ceph_mdsc_do_request(mdsc, NULL, req);
118 inode = req->r_target_inode;
119 if (inode)
120 ihold(inode);
121 ceph_mdsc_put_request(req);
122 if (!inode)
123 return ERR_PTR(-ESTALE);
124 }
125
126 dentry = d_obtain_alias(inode);
127 if (IS_ERR(dentry)) {
128 pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n",
129 fh->ino, inode);
130 iput(inode);
131 return dentry;
132 }
133 err = ceph_init_dentry(dentry);
134 if (err < 0) {
135 iput(inode);
136 return ERR_PTR(err);
137 }
138 dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry);
139 return dentry;
140 }
141
142 /*
143 * convert connectable fh to dentry
144 */
__cfh_to_dentry(struct super_block * sb,struct ceph_nfs_confh * cfh,int fh_len)145 static struct dentry *__cfh_to_dentry(struct super_block *sb,
146 struct ceph_nfs_confh *cfh, int fh_len)
147 {
148 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
149 struct inode *inode;
150 struct dentry *dentry;
151 struct ceph_vino vino;
152 int err;
153
154 if (fh_len < sizeof(*cfh) / 4)
155 return ERR_PTR(-ESTALE);
156
157 dout("__cfh_to_dentry %llx (%llx/%x)\n",
158 cfh->ino, cfh->parent_ino, cfh->parent_name_hash);
159
160 vino.ino = cfh->ino;
161 vino.snap = CEPH_NOSNAP;
162 inode = ceph_find_inode(sb, vino);
163 if (!inode) {
164 struct ceph_mds_request *req;
165
166 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH,
167 USE_ANY_MDS);
168 if (IS_ERR(req))
169 return ERR_CAST(req);
170
171 req->r_ino1 = vino;
172 req->r_ino2.ino = cfh->parent_ino;
173 req->r_ino2.snap = CEPH_NOSNAP;
174 req->r_path2 = kmalloc(16, GFP_NOFS);
175 snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash);
176 req->r_num_caps = 1;
177 err = ceph_mdsc_do_request(mdsc, NULL, req);
178 inode = req->r_target_inode;
179 if (inode)
180 ihold(inode);
181 ceph_mdsc_put_request(req);
182 if (!inode)
183 return ERR_PTR(err ? err : -ESTALE);
184 }
185
186 dentry = d_obtain_alias(inode);
187 if (IS_ERR(dentry)) {
188 pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n",
189 cfh->ino, inode);
190 iput(inode);
191 return dentry;
192 }
193 err = ceph_init_dentry(dentry);
194 if (err < 0) {
195 iput(inode);
196 return ERR_PTR(err);
197 }
198 dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry);
199 return dentry;
200 }
201
ceph_fh_to_dentry(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)202 static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid,
203 int fh_len, int fh_type)
204 {
205 if (fh_type == 1)
206 return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw,
207 fh_len);
208 else
209 return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw,
210 fh_len);
211 }
212
213 /*
214 * get parent, if possible.
215 *
216 * FIXME: we could do better by querying the mds to discover the
217 * parent.
218 */
ceph_fh_to_parent(struct super_block * sb,struct fid * fid,int fh_len,int fh_type)219 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
220 struct fid *fid,
221 int fh_len, int fh_type)
222 {
223 struct ceph_nfs_confh *cfh = (void *)fid->raw;
224 struct ceph_vino vino;
225 struct inode *inode;
226 struct dentry *dentry;
227 int err;
228
229 if (fh_type == 1)
230 return ERR_PTR(-ESTALE);
231 if (fh_len < sizeof(*cfh) / 4)
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
259 const 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 };
264