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, int fh_len) 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 if (fh_len < sizeof(*fh) / 4) 111 return ERR_PTR(-ESTALE); 112 113 dout("__fh_to_dentry %llx\n", fh->ino); 114 vino.ino = fh->ino; 115 vino.snap = CEPH_NOSNAP; 116 inode = ceph_find_inode(sb, vino); 117 if (!inode) { 118 struct ceph_mds_request *req; 119 120 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPINO, 121 USE_ANY_MDS); 122 if (IS_ERR(req)) 123 return ERR_CAST(req); 124 125 req->r_ino1 = vino; 126 req->r_num_caps = 1; 127 err = ceph_mdsc_do_request(mdsc, NULL, req); 128 inode = req->r_target_inode; 129 if (inode) 130 ihold(inode); 131 ceph_mdsc_put_request(req); 132 if (!inode) 133 return ERR_PTR(-ESTALE); 134 } 135 136 dentry = d_obtain_alias(inode); 137 if (IS_ERR(dentry)) { 138 pr_err("fh_to_dentry %llx -- inode %p but ENOMEM\n", 139 fh->ino, inode); 140 iput(inode); 141 return dentry; 142 } 143 err = ceph_init_dentry(dentry); 144 if (err < 0) { 145 iput(inode); 146 return ERR_PTR(err); 147 } 148 dout("__fh_to_dentry %llx %p dentry %p\n", fh->ino, inode, dentry); 149 return dentry; 150 } 151 152 /* 153 * convert connectable fh to dentry 154 */ 155 static struct dentry *__cfh_to_dentry(struct super_block *sb, 156 struct ceph_nfs_confh *cfh, int fh_len) 157 { 158 struct ceph_mds_client *mdsc = ceph_sb_to_client(sb)->mdsc; 159 struct inode *inode; 160 struct dentry *dentry; 161 struct ceph_vino vino; 162 int err; 163 164 if (fh_len < sizeof(*cfh) / 4) 165 return ERR_PTR(-ESTALE); 166 167 dout("__cfh_to_dentry %llx (%llx/%x)\n", 168 cfh->ino, cfh->parent_ino, cfh->parent_name_hash); 169 170 vino.ino = cfh->ino; 171 vino.snap = CEPH_NOSNAP; 172 inode = ceph_find_inode(sb, vino); 173 if (!inode) { 174 struct ceph_mds_request *req; 175 176 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_LOOKUPHASH, 177 USE_ANY_MDS); 178 if (IS_ERR(req)) 179 return ERR_CAST(req); 180 181 req->r_ino1 = vino; 182 req->r_ino2.ino = cfh->parent_ino; 183 req->r_ino2.snap = CEPH_NOSNAP; 184 req->r_path2 = kmalloc(16, GFP_NOFS); 185 snprintf(req->r_path2, 16, "%d", cfh->parent_name_hash); 186 req->r_num_caps = 1; 187 err = ceph_mdsc_do_request(mdsc, NULL, req); 188 inode = req->r_target_inode; 189 if (inode) 190 ihold(inode); 191 ceph_mdsc_put_request(req); 192 if (!inode) 193 return ERR_PTR(err ? err : -ESTALE); 194 } 195 196 dentry = d_obtain_alias(inode); 197 if (IS_ERR(dentry)) { 198 pr_err("cfh_to_dentry %llx -- inode %p but ENOMEM\n", 199 cfh->ino, inode); 200 iput(inode); 201 return dentry; 202 } 203 err = ceph_init_dentry(dentry); 204 if (err < 0) { 205 iput(inode); 206 return ERR_PTR(err); 207 } 208 dout("__cfh_to_dentry %llx %p dentry %p\n", cfh->ino, inode, dentry); 209 return dentry; 210 } 211 212 static struct dentry *ceph_fh_to_dentry(struct super_block *sb, struct fid *fid, 213 int fh_len, int fh_type) 214 { 215 if (fh_type == 1) 216 return __fh_to_dentry(sb, (struct ceph_nfs_fh *)fid->raw, 217 fh_len); 218 else 219 return __cfh_to_dentry(sb, (struct ceph_nfs_confh *)fid->raw, 220 fh_len); 221 } 222 223 /* 224 * get parent, if possible. 225 * 226 * FIXME: we could do better by querying the mds to discover the 227 * parent. 228 */ 229 static struct dentry *ceph_fh_to_parent(struct super_block *sb, 230 struct fid *fid, 231 int fh_len, int fh_type) 232 { 233 struct ceph_nfs_confh *cfh = (void *)fid->raw; 234 struct ceph_vino vino; 235 struct inode *inode; 236 struct dentry *dentry; 237 int err; 238 239 if (fh_type == 1) 240 return ERR_PTR(-ESTALE); 241 if (fh_len < sizeof(*cfh) / 4) 242 return ERR_PTR(-ESTALE); 243 244 pr_debug("fh_to_parent %llx/%d\n", cfh->parent_ino, 245 cfh->parent_name_hash); 246 247 vino.ino = cfh->ino; 248 vino.snap = CEPH_NOSNAP; 249 inode = ceph_find_inode(sb, vino); 250 if (!inode) 251 return ERR_PTR(-ESTALE); 252 253 dentry = d_obtain_alias(inode); 254 if (IS_ERR(dentry)) { 255 pr_err("fh_to_parent %llx -- inode %p but ENOMEM\n", 256 cfh->ino, inode); 257 iput(inode); 258 return dentry; 259 } 260 err = ceph_init_dentry(dentry); 261 if (err < 0) { 262 iput(inode); 263 return ERR_PTR(err); 264 } 265 dout("fh_to_parent %llx %p dentry %p\n", cfh->ino, inode, dentry); 266 return dentry; 267 } 268 269 const struct export_operations ceph_export_ops = { 270 .encode_fh = ceph_encode_fh, 271 .fh_to_dentry = ceph_fh_to_dentry, 272 .fh_to_parent = ceph_fh_to_parent, 273 }; 274