lustre/llite/super.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Light Super operations
   5  *
   6  * This code is issued under the GNU General Public License.
   7  * See the file COPYING in this distribution
   8  *
   9  * Copryright (C) 1996 Peter J. Braam <braam@stelias.com>
  10  * Copryright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
  11  * Copryright (C) 1999 Seagate Technology Inc.
  12  * Copryright (C) 2001 Mountain View Data, Inc.
  13  * Copryright (C) 2002 Cluster File Systems, Inc.
  14  *
  15  */
  16
  17 #define DEBUG_SUBSYSTEM S_LLITE
  18
  19 #include <linux/module.h>
  20 #include <linux/lustre_lite.h>
  21 #include <linux/lustre_ha.h>
  22 #include <linux/lustre_dlm.h>
  23
  24 kmem_cache_t *ll_file_data_slab;
  25 extern struct address_space_operations ll_aops;
  26 extern struct address_space_operations ll_dir_aops;
  27 struct super_operations ll_super_operations;
  28
  29 extern void ll_recover(struct ptlrpc_client *);
  30 extern int ll_commitcbd_setup(struct ll_sb_info *);
  31 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  32
  33 static char *ll_read_opt(const char *opt, char *data)
  34 {
  35         char *value;
  36         char *retval;
  37         ENTRY;
  38
  39         CDEBUG(D_INFO, "option: %s, data %s\n", opt, data);
  40         if ( strncmp(opt, data, strlen(opt)) )
  41                 RETURN(NULL);
  42         if ( (value = strchr(data, '=')) == NULL )
  43                 RETURN(NULL);
  44
  45         value++;
  46         OBD_ALLOC(retval, strlen(value) + 1);
  47         if ( !retval ) {
  48                 CERROR("out of memory!\n");
  49                 RETURN(NULL);
  50         }
  51
  52         memcpy(retval, value, strlen(value)+1);
  53         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  54         RETURN(retval);
  55 }
  56
  57 static void ll_options(char *options, char **dev, char **vers)
  58 {
  59         char *this_char;
  60         ENTRY;
  61
  62         if (!options) {
  63                 EXIT;
  64                 return;
  65         }
  66
  67         for (this_char = strtok (options, ",");
  68              this_char != NULL;
  69              this_char = strtok (NULL, ",")) {
  70                 CDEBUG(D_INFO, "this_char %s\n", this_char);
  71                 if ( (!*dev && (*dev = ll_read_opt("device", this_char)))||
  72                      (!*vers && (*vers = ll_read_opt("version", this_char))) )
  73                         continue;
  74         }
  75         EXIT;
  76 }
  77
  78 static struct super_block * ll_read_super(struct super_block *sb,
  79                                           void *data, int silent)
  80 {
  81         struct inode *root = 0;
  82         struct ll_sb_info *sbi;
  83         char *device = NULL;
  84         char *version = NULL;
  85         int devno;
  86         int err;
  87         struct ll_fid rootfid;
  88         __u64 last_committed, last_rcvd;
  89         __u32 last_xid;
  90         struct ptlrpc_request *request = NULL;
  91
  92         ENTRY;
  93         MOD_INC_USE_COUNT;
  94
  95         OBD_ALLOC(sbi, sizeof(*sbi));
  96         if (!sbi) {
  97                 MOD_DEC_USE_COUNT;
  98                 RETURN(NULL);
  99         }
 100
 101         sb->u.generic_sbp = sbi;
 102
 103         ll_options(data, &device, &version);
 104
 105         if (!device) {
 106                 CERROR("no device\n");
 107                 GOTO(out_free, sb = NULL);
 108         }
 109
 110         devno = simple_strtoul(device, NULL, 0);
 111         if (devno >= MAX_OBD_DEVICES) {
 112                 CERROR("device of %s too high\n", device);
 113                 GOTO(out_free, sb = NULL);
 114         }
 115
 116         sbi->ll_conn.oc_dev = &obd_dev[devno];
 117         err = obd_connect(&sbi->ll_conn);
 118         if (err) {
 119                 CERROR("cannot connect to %s: rc = %d\n", device, err);
 120                 GOTO(out_free, sb = NULL);
 121         }
 122
 123         sbi->ll_namespace = ldlm_namespace_new(NULL, 1);
 124         if (sbi->ll_namespace == NULL) {
 125                 CERROR("failed to create local lock namespace\n");
 126                 GOTO(out_free, sb = NULL);
 127         }
 128
 129         ptlrpc_init_client(ptlrpc_connmgr, ll_recover,
 130                            MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
 131                            &sbi->ll_mds_client);
 132
 133         sbi->ll_mds_conn = ptlrpc_uuid_to_connection("mds");
 134         if (!sbi->ll_mds_conn) {
 135                 CERROR("cannot find MDS\n");
 136                 GOTO(out_disc, sb = NULL);
 137         }
 138
 139         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mds_conn);
 140         if (err) {
 141                 CERROR("cannot connect to MDS: rc = %d\n", err);
 142                 ptlrpc_put_connection(sbi->ll_mds_conn);
 143                 GOTO(out_disc, sb = NULL);
 144         }
 145
 146         sbi->ll_mds_conn->c_level = LUSTRE_CONN_FULL;
 147
 148         /* XXX: need to store the last_* values somewhere */
 149         err = mdc_connect(&sbi->ll_mds_client, sbi->ll_mds_conn,
 150                           &rootfid, &last_committed, &last_rcvd, &last_xid,
 151                           &request);
 152         if (err) {
 153                 CERROR("cannot mds_connect: rc = %d\n", err);
 154                 GOTO(out_disc, sb = NULL);
 155         }
 156         CERROR("rootfid %ld\n", (unsigned long)rootfid.id);
 157         sbi->ll_rootino = rootfid.id;
 158
 159         sb->s_maxbytes = 1ULL << 36;
 160         sb->s_blocksize = PAGE_SIZE;
 161         sb->s_blocksize_bits = (unsigned char)PAGE_SHIFT;
 162         sb->s_magic = LL_SUPER_MAGIC;
 163         sb->s_op = &ll_super_operations;
 164
 165         /* make root inode */
 166         err = mdc_getattr(&sbi->ll_mds_client, sbi->ll_mds_conn,
 167                           sbi->ll_rootino, S_IFDIR,
 168                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, &request);
 169         if (err) {
 170                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 171                 GOTO(out_req, sb = NULL);
 172         }
 173
 174         /* initialize committed transaction callback daemon */
 175         spin_lock_init(&sbi->ll_commitcbd_lock);
 176         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 177         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 178         sbi->ll_commitcbd_flags = 0;
 179         err = ll_commitcbd_setup(sbi);
 180         if (err) {
 181                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 182                 GOTO(out_req, sb = NULL);
 183         }
 184
 185         root = iget4(sb, sbi->ll_rootino, NULL,
 186                      lustre_msg_buf(request->rq_repmsg, 0));
 187         if (root) {
 188                 sb->s_root = d_alloc_root(root);
 189         } else {
 190                 CERROR("lustre_lite: bad iget4 for root\n");
 191                 GOTO(out_req, sb = NULL);
 192         }
 193
 194 out_req:
 195         ptlrpc_free_req(request);
 196         if (!sb) {
 197 out_disc:
 198                 obd_disconnect(&sbi->ll_conn);
 199 out_free:
 200                 MOD_DEC_USE_COUNT;
 201                 if (sbi->ll_namespace)
 202                         ldlm_namespace_free(sbi->ll_namespace);
 203                 OBD_FREE(sbi, sizeof(*sbi));
 204         }
 205         if (device)
 206                 OBD_FREE(device, strlen(device) + 1);
 207         if (version)
 208                 OBD_FREE(version, strlen(version) + 1);
 209
 210         RETURN(sb);
 211 } /* ll_read_super */
 212
 213 static void ll_put_super(struct super_block *sb)
 214 {
 215         struct ll_sb_info *sbi = sb->u.generic_sbp;
 216         ENTRY;
 217         ll_commitcbd_cleanup(sbi);
 218         obd_disconnect(&sbi->ll_conn);
 219         ldlm_namespace_free(sbi->ll_namespace);
 220         ptlrpc_put_connection(sbi->ll_mds_conn);
 221         ptlrpc_cleanup_client(&sbi->ll_mds_client);
 222         OBD_FREE(sb->u.generic_sbp, sizeof(*sbi));
 223         MOD_DEC_USE_COUNT;
 224         EXIT;
 225 } /* ll_put_super */
 226
 227
 228 extern inline struct obdo * ll_oa_from_inode(struct inode *inode, int valid);
 229 static void ll_delete_inode(struct inode *inode)
 230 {
 231         if (S_ISREG(inode->i_mode)) {
 232                 int err;
 233                 struct obdo *oa;
 234                 oa = ll_oa_from_inode(inode, OBD_MD_FLNOTOBD);
 235                 if (!oa) {
 236                         CERROR("no memory\n");
 237                 }
 238
 239                 err = obd_destroy(ll_i2obdconn(inode), oa);
 240                 CDEBUG(D_INODE, "obd destroy of %Ld error %d\n",
 241                        (unsigned long long)oa->o_id, err);
 242                 obdo_free(oa);
 243         }
 244
 245         clear_inode(inode);
 246 }
 247
 248 /* like inode_setattr, but doesn't mark the inode dirty */
 249 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 250 {
 251         unsigned int ia_valid = attr->ia_valid;
 252         int error = 0;
 253
 254         if ((ia_valid & ATTR_SIZE) && trunc ) {
 255                 error = vmtruncate(inode, attr->ia_size);
 256                 if (error)
 257                         goto out;
 258         } else if (ia_valid & ATTR_SIZE) {
 259                 inode->i_size = attr->ia_size;
 260         }
 261
 262         if (ia_valid & ATTR_UID)
 263                 inode->i_uid = attr->ia_uid;
 264         if (ia_valid & ATTR_GID)
 265                 inode->i_gid = attr->ia_gid;
 266         if (ia_valid & ATTR_ATIME)
 267                 inode->i_atime = attr->ia_atime;
 268         if (ia_valid & ATTR_MTIME)
 269                 inode->i_mtime = attr->ia_mtime;
 270         if (ia_valid & ATTR_CTIME)
 271                 inode->i_ctime = attr->ia_ctime;
 272         if (ia_valid & ATTR_MODE) {
 273                 inode->i_mode = attr->ia_mode;
 274                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 275                         inode->i_mode &= ~S_ISGID;
 276         }
 277 out:
 278         return error;
 279 }
 280
 281 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 282 {
 283         struct ptlrpc_request *request = NULL;
 284         struct ll_sb_info *sbi = ll_i2sbi(inode);
 285         int err;
 286
 287         ENTRY;
 288
 289         /* change incore inode */
 290         ll_attr2inode(inode, attr, do_trunc);
 291
 292         err = mdc_setattr(&sbi->ll_mds_client, sbi->ll_mds_conn, inode, attr,
 293                           &request);
 294         if (err)
 295                 CERROR("mdc_setattr fails (%d)\n", err);
 296
 297         ptlrpc_req_finished(request);
 298
 299         RETURN(err);
 300 }
 301
 302 int ll_setattr(struct dentry *de, struct iattr *attr)
 303 {
 304         return ll_inode_setattr(de->d_inode, attr, 1);
 305 }
 306
 307 static int ll_statfs(struct super_block *sb, struct statfs *buf)
 308 {
 309         struct statfs tmp;
 310         int err;
 311         ENTRY;
 312
 313         err = obd_statfs(&ll_s2sbi(sb)->ll_conn, &tmp);
 314         if (err) {
 315                 CERROR("obd_statfs fails (%d)\n", err);
 316                 RETURN(err);
 317         }
 318         memcpy(buf, &tmp, sizeof(*buf));
 319         CDEBUG(D_SUPER, "statfs returns avail %ld\n", tmp.f_bavail);
 320
 321         RETURN(err);
 322 }
 323
 324 static void inline ll_to_inode(struct inode *dst, struct mds_body *body)
 325 {
 326         struct ll_inode_info *ii =
 327                 (struct ll_inode_info *) &dst->u.generic_ip;
 328
 329         /* core attributes first */
 330         if ( body->valid & OBD_MD_FLID )
 331                 dst->i_ino = body->ino;
 332         if ( body->valid & OBD_MD_FLATIME )
 333                 dst->i_atime = body->atime;
 334         if ( body->valid & OBD_MD_FLMTIME )
 335                 dst->i_mtime = body->mtime;
 336         if ( body->valid & OBD_MD_FLCTIME )
 337                 dst->i_ctime = body->ctime;
 338         if ( body->valid & OBD_MD_FLSIZE )
 339                 dst->i_size = body->size;
 340         if ( body->valid & OBD_MD_FLMODE )
 341                 dst->i_mode = body->mode;
 342         if ( body->valid & OBD_MD_FLUID )
 343                 dst->i_uid = body->uid;
 344         if ( body->valid & OBD_MD_FLGID )
 345                 dst->i_gid = body->gid;
 346         if ( body->valid & OBD_MD_FLFLAGS )
 347                 dst->i_flags = body->flags;
 348         if ( body->valid & OBD_MD_FLNLINK )
 349                 dst->i_nlink = body->nlink;
 350         if ( body->valid & OBD_MD_FLGENER )
 351                 dst->i_generation = body->generation;
 352
 353         /* this will become more elaborate for striping etc */
 354         if (body->valid & OBD_MD_FLOBJID)
 355                 ii->lli_objid = body->objid;
 356 #if 0
 357
 358         if (obdo_has_inline(oa)) {
 359                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 360                     S_ISFIFO(inode->i_mode)) {
 361                         obd_rdev rdev = *((obd_rdev *)oa->o_inline);
 362                         CDEBUG(D_INODE,
 363                                "copying device %x from obdo to inode\n", rdev);
 364                         init_special_inode(inode, inode->i_mode, rdev);
 365                 } else {
 366                         CDEBUG(D_INFO, "copying inline from obdo to inode\n");
 367                         memcpy(oinfo->lli_inline, oa->o_inline, OBD_INLINESZ);
 368                 }
 369                 oinfo->lli_flags |= OBD_FL_INLINEDATA;
 370         }
 371 #endif
 372 } /* ll_to_inode */
 373
 374 static inline void ll_read_inode2(struct inode *inode, void *opaque)
 375 {
 376         struct mds_body *body = opaque;
 377
 378         ENTRY;
 379         ll_to_inode(inode, body);
 380
 381         /* OIDEBUG(inode); */
 382
 383         if (S_ISREG(inode->i_mode)) {
 384                 inode->i_op = &ll_file_inode_operations;
 385                 inode->i_fop = &ll_file_operations;
 386                 inode->i_mapping->a_ops = &ll_aops;
 387                 EXIT;
 388         } else if (S_ISDIR(inode->i_mode)) {
 389                 inode->i_op = &ll_dir_inode_operations;
 390                 inode->i_fop = &ll_dir_operations;
 391                 inode->i_mapping->a_ops = &ll_dir_aops;
 392                 EXIT;
 393         } else if (S_ISLNK(inode->i_mode)) {
 394                 inode->i_op = &ll_fast_symlink_inode_operations;
 395                 EXIT;
 396         } else {
 397                 init_special_inode(inode, inode->i_mode,
 398                                    ((int *)ll_i2info(inode)->lli_inline)[0]);
 399                 EXIT;
 400         }
 401
 402         return;
 403 }
 404
 405 /* exported operations */
 406 struct super_operations ll_super_operations =
 407 {
 408         read_inode2: ll_read_inode2,
 409         delete_inode: ll_delete_inode,
 410         put_super: ll_put_super,
 411         statfs: ll_statfs
 412 };
 413
 414 struct file_system_type lustre_lite_fs_type = {
 415         "lustre_lite", 0, ll_read_super, NULL
 416 };
 417
 418 static int __init init_lustre_lite(void)
 419 {
 420         printk(KERN_INFO "Lustre Lite 0.0.1, braam@clusterfs.com\n");
 421         ll_file_data_slab = kmem_cache_create("ll_file_data",
 422                                               sizeof(struct ll_file_data), 0,
 423                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 424         if (ll_file_data_slab == NULL)
 425                 return -ENOMEM;
 426         return register_filesystem(&lustre_lite_fs_type);
 427 }
 428
 429 static void __exit exit_lustre_lite(void)
 430 {
 431         unregister_filesystem(&lustre_lite_fs_type);
 432         kmem_cache_destroy(ll_file_data_slab);
 433 }
 434
 435 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 436 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 437 MODULE_LICENSE("GPL");
 438
 439 module_init(init_lustre_lite);
 440 module_exit(exit_lustre_lite);