xref: /openbmc/linux/fs/ceph/export.c (revision 54cbac81)
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;
60 	struct dentry *parent;
61 
62 	/* don't re-export snaps */
63 	if (ceph_snap(inode) != CEPH_NOSNAP)
64 		return -EINVAL;
65 
66 	dentry = d_find_alias(inode);
67 
68 	/* if we found an alias, generate a connectable fh */
69 	if (*max_len >= connected_handle_length && dentry) {
70 		dout("encode_fh %p connectable\n", dentry);
71 		spin_lock(&dentry->d_lock);
72 		parent = dentry->d_parent;
73 		cfh->ino = ceph_ino(inode);
74 		cfh->parent_ino = ceph_ino(parent->d_inode);
75 		cfh->parent_name_hash = ceph_dentry_hash(parent->d_inode,
76 							 dentry);
77 		*max_len = connected_handle_length;
78 		type = 2;
79 		spin_unlock(&dentry->d_lock);
80 	} else if (*max_len >= handle_length) {
81 		if (parent_inode) {
82 			/* nfsd wants connectable */
83 			*max_len = connected_handle_length;
84 			type = 255;
85 		} else {
86 			dout("encode_fh %p\n", dentry);
87 			fh->ino = ceph_ino(inode);
88 			*max_len = handle_length;
89 			type = 1;
90 		}
91 	} else {
92 		*max_len = handle_length;
93 		type = 255;
94 	}
95 	if (dentry)
96 		dput(dentry);
97 	return type;
98 }
99 
100 /*
101  * convert regular fh to dentry
102  *
103  * FIXME: we should try harder by querying the mds for the ino.
104  */
105 static struct dentry *__fh_to_dentry(struct super_block *sb,
106 				     struct ceph_nfs_fh *fh, int fh_len)
107 {
108 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
109 	struct inode *inode;
110 	struct dentry *dentry;
111 	struct ceph_vino vino;
112 	int err;
113 
114 	if (fh_len < sizeof(*fh) / 4)
115 		return ERR_PTR(-ESTALE);
116 
117 	dout("__fh_to_dentry %llx\n", fh->ino);
118 	vino.ino = fh->ino;
119 	vino.snap = CEPH_NOSNAP;
120 	inode = ceph_find_inode(sb, vino);
121 	if (!inode) {
122 		struct ceph_mds_request *req;
123 
124 		req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO,
125 					       USE_ANY_MDS);
126 		if (IS_ERR(req))
127 			return ERR_CAST(req);
128 
129 		req->r_ino1 = vino;
130 		req->r_num_caps = 1;
131 		err = ceph_mdsc_do_request(mdsc, NULL, req);
132 		inode = req->r_target_inode;
133 		if (inode)
134 			ihold(inode);
135 		ceph_mdsc_put_request(req);
136 		if (!inode)
137 			return ERR_PTR(-ESTALE);
138 	}
139 
140 	dentry = d_obtain_alias(inode);
141 	if (IS_ERR(dentry)) {
142 		pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n",
143 		       fh->ino, inode);
144 		iput(inode);
145 		return dentry;
146 	}
147 	err = ceph_init_dentry(dentry);
148 	if (err < 0) {
149 		iput(inode);
150 		return ERR_PTR(err);
151 	}
152 	dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry);
153 	return dentry;
154 }
155 
156 /*
157  * convert connectable fh to dentry
158  */
159 static struct dentry *__cfh_to_dentry(struct super_block *sb,
160 				      struct ceph_nfs_confh *cfh, int fh_len)
161 {
162 	struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc;
163 	struct inode *inode;
164 	struct dentry *dentry;
165 	struct ceph_vino vino;
166 	int err;
167 
168 	if (fh_len < sizeof(*cfh) / 4)
169 		return ERR_PTR(-ESTALE);
170 
171 	dout("__cfh_to_dentry %llx (%llx/%x)\n",
172 	     cfh->ino, cfh->parent_ino, cfh->parent_name_hash);
173 
174 	vino.ino = cfh->ino;
175 	vino.snap = CEPH_NOSNAP;
176 	inode = ceph_find_inode(sb, vino);
177 	if (!inode) {
178 		struct ceph_mds_request *req;
179 
180 		req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH,
181 					       USE_ANY_MDS);
182 		if (IS_ERR(req))
183 			return ERR_CAST(req);
184 
185 		req->r_ino1 = vino;
186 		req->r_ino2.ino = cfh->parent_ino;
187 		req->r_ino2.snap = CEPH_NOSNAP;
188 		req->r_path2 = kmalloc(16, GFP_NOFS);
189 		snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash);
190 		req->r_num_caps = 1;
191 		err = ceph_mdsc_do_request(mdsc, NULL, req);
192 		inode = req->r_target_inode;
193 		if (inode)
194 			ihold(inode);
195 		ceph_mdsc_put_request(req);
196 		if (!inode)
197 			return ERR_PTR(err ? err : -ESTALE);
198 	}
199 
200 	dentry = d_obtain_alias(inode);
201 	if (IS_ERR(dentry)) {
202 		pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n",
203 		       cfh->ino, inode);
204 		iput(inode);
205 		return dentry;
206 	}
207 	err = ceph_init_dentry(dentry);
208 	if (err < 0) {
209 		iput(inode);
210 		return ERR_PTR(err);
211 	}
212 	dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry);
213 	return dentry;
214 }
215 
216 static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid,
217 					int fh_len, int fh_type)
218 {
219 	if (fh_type == 1)
220 		return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw,
221 								fh_len);
222 	else
223 		return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw,
224 								fh_len);
225 }
226 
227 /*
228  * get parent, if possible.
229  *
230  * FIXME: we could do better by querying the mds to discover the
231  * parent.
232  */
233 static struct dentry *ceph_fh_to_parent(struct super_block *sb,
234 					 struct fid *fid,
235 					int fh_len, int fh_type)
236 {
237 	struct ceph_nfs_confh *cfh = (void *)fid->raw;
238 	struct ceph_vino vino;
239 	struct inode *inode;
240 	struct dentry *dentry;
241 	int err;
242 
243 	if (fh_type == 1)
244 		return ERR_PTR(-ESTALE);
245 	if (fh_len < sizeof(*cfh) / 4)
246 		return ERR_PTR(-ESTALE);
247 
248 	pr_debug("fh_to_parent %llx/%d\n", cfh->parent_ino,
249 		 cfh->parent_name_hash);
250 
251 	vino.ino = cfh->ino;
252 	vino.snap = CEPH_NOSNAP;
253 	inode = ceph_find_inode(sb, vino);
254 	if (!inode)
255 		return ERR_PTR(-ESTALE);
256 
257 	dentry = d_obtain_alias(inode);
258 	if (IS_ERR(dentry)) {
259 		pr_err("fh_to_parent %llx -- inode %p but ENOMEM\n",
260 		       cfh->ino, inode);
261 		iput(inode);
262 		return dentry;
263 	}
264 	err = ceph_init_dentry(dentry);
265 	if (err < 0) {
266 		iput(inode);
267 		return ERR_PTR(err);
268 	}
269 	dout("fh_to_parent %llx %p dentry %p\n", cfh->ino, inode, dentry);
270 	return dentry;
271 }
272 
273 const struct export_operations ceph_export_ops = {
274 	.encode_fh = ceph_encode_fh,
275 	.fh_to_dentry = ceph_fh_to_dentry,
276 	.fh_to_parent = ceph_fh_to_parent,
277 };
278