xref: /openbmc/linux/fs/ceph/export.c (revision 63dc02bd)
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 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  */
91 static struct dentry *__fh_to_dentry(struct super_block *sb,
92 				     struct ceph_nfs_fh *fh)
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 	dout("__fh_to_dentry %llx\n", fh->ino);
101 	vino.ino = fh->ino;
102 	vino.snap = CEPH_NOSNAP;
103 	inode = ceph_find_inode(sb, vino);
104 	if (!inode) {
105 		struct ceph_mds_request *req;
106 
107 		req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
108 					       USE_ANY_MDS);
109 		if (IS_ERR(req))
110 			return ERR_CAST(req);
111 
112 		req->r_ino1 = vino;
113 		req->r_num_caps = 1;
114 		err = ceph_mdsc_do_request(mdsc, NULL, req);
115 		inode = req->r_target_inode;
116 		if (inode)
117 			ihold(inode);
118 		ceph_mdsc_put_request(req);
119 		if (!inode)
120 			return ERR_PTR(-ESTALE);
121 	}
122 
123 	dentry = d_obtain_alias(inode);
124 	if (IS_ERR(dentry)) {
125 		pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n",
126 		       fh->ino, inode);
127 		iput(inode);
128 		return dentry;
129 	}
130 	err = ceph_init_dentry(dentry);
131 	if (err < 0) {
132 		iput(inode);
133 		return ERR_PTR(err);
134 	}
135 	dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry);
136 	return dentry;
137 }
138 
139 /*
140  * convert connectable fh to dentry
141  */
142 static struct dentry *__cfh_to_dentry(struct super_block *sb,
143 				      struct ceph_nfs_confh *cfh)
144 {
145 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
146 	struct inode *inode;
147 	struct dentry *dentry;
148 	struct ceph_vino vino;
149 	int err;
150 
151 	dout("__cfh_to_dentry %llx (%llx/%x)\n",
152 	     cfh->ino, cfh->parent_ino, cfh->parent_name_hash);
153 
154 	vino.ino = cfh->ino;
155 	vino.snap = CEPH_NOSNAP;
156 	inode = ceph_find_inode(sb, vino);
157 	if (!inode) {
158 		struct ceph_mds_request *req;
159 
160 		req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH,
161 					       USE_ANY_MDS);
162 		if (IS_ERR(req))
163 			return ERR_CAST(req);
164 
165 		req->r_ino1 = vino;
166 		req->r_ino2.ino = cfh->parent_ino;
167 		req->r_ino2.snap = CEPH_NOSNAP;
168 		req->r_path2 = kmalloc(16, GFP_NOFS);
169 		snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash);
170 		req->r_num_caps = 1;
171 		err = ceph_mdsc_do_request(mdsc, NULL, req);
172 		inode = req->r_target_inode;
173 		if (inode)
174 			ihold(inode);
175 		ceph_mdsc_put_request(req);
176 		if (!inode)
177 			return ERR_PTR(err ? err : -ESTALE);
178 	}
179 
180 	dentry = d_obtain_alias(inode);
181 	if (IS_ERR(dentry)) {
182 		pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n",
183 		       cfh->ino, inode);
184 		iput(inode);
185 		return dentry;
186 	}
187 	err = ceph_init_dentry(dentry);
188 	if (err < 0) {
189 		iput(inode);
190 		return ERR_PTR(err);
191 	}
192 	dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry);
193 	return dentry;
194 }
195 
196 static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid,
197 					int fh_len, int fh_type)
198 {
199 	if (fh_type == 1)
200 		return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw);
201 	else
202 		return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw);
203 }
204 
205 /*
206  * get parent, if possible.
207  *
208  * FIXME: we could do better by querying the mds to discover the
209  * parent.
210  */
211 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
212 					 struct fid *fid,
213 					int fh_len, int fh_type)
214 {
215 	struct ceph_nfs_confh *cfh = (void *)fid->raw;
216 	struct ceph_vino vino;
217 	struct inode *inode;
218 	struct dentry *dentry;
219 	int err;
220 
221 	if (fh_type == 1)
222 		return ERR_PTR(-ESTALE);
223 
224 	pr_debug("fh_to_parent %llx/%d\n", cfh->parent_ino,
225 		 cfh->parent_name_hash);
226 
227 	vino.ino = cfh->ino;
228 	vino.snap = CEPH_NOSNAP;
229 	inode = ceph_find_inode(sb, vino);
230 	if (!inode)
231 		return ERR_PTR(-ESTALE);
232 
233 	dentry = d_obtain_alias(inode);
234 	if (IS_ERR(dentry)) {
235 		pr_err("fh_to_parent %llx -- inode %p but ENOMEM\n",
236 		       cfh->ino, inode);
237 		iput(inode);
238 		return dentry;
239 	}
240 	err = ceph_init_dentry(dentry);
241 	if (err < 0) {
242 		iput(inode);
243 		return ERR_PTR(err);
244 	}
245 	dout("fh_to_parent %llx %p dentry %p\n", cfh->ino, inode, dentry);
246 	return dentry;
247 }
248 
249 const struct export_operations ceph_export_ops = {
250 	.encode_fh = ceph_encode_fh,
251 	.fh_to_dentry = ceph_fh_to_dentry,
252 	.fh_to_parent = ceph_fh_to_parent,
253 };
254