xref: /openbmc/linux/fs/ceph/export.c (revision 600a711c)
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 
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  */
51 static 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  */
101 static 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  */
152 static 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 
206 static 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  */
221 static 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 
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