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 int 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         struct ll_inode_md md;
  92
  93         ENTRY;
  94         MOD_INC_USE_COUNT;
  95
  96         OBD_ALLOC(sbi, sizeof(*sbi));
  97         if (!sbi) {
  98                 MOD_DEC_USE_COUNT;
  99                 RETURN(NULL);
 100         }
 101
 102         sb->u.generic_sbp = sbi;
 103
 104         ll_options(data, &device, &version);
 105
 106         if (!device) {
 107                 CERROR("no device\n");
 108                 GOTO(out_free, sb = NULL);
 109         }
 110
 111         devno = simple_strtoul(device, NULL, 0);
 112         if (devno >= MAX_OBD_DEVICES) {
 113                 CERROR("device of %s too high\n", device);
 114                 GOTO(out_free, sb = NULL);
 115         }
 116
 117         sbi->ll_conn.oc_dev = &obd_dev[devno];
 118         err = obd_connect(&sbi->ll_conn);
 119         if (err) {
 120                 CERROR("cannot connect to %s: rc = %d\n", device, err);
 121                 GOTO(out_free, sb = NULL);
 122         }
 123
 124         sbi->ll_namespace = ldlm_namespace_new(NULL, 1);
 125         if (sbi->ll_namespace == NULL) {
 126                 CERROR("failed to create local lock namespace\n");
 127                 GOTO(out_obd, sb = NULL);
 128         }
 129
 130         ptlrpc_init_client(ptlrpc_connmgr, ll_recover,
 131                            MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
 132                            &sbi->ll_mds_client);
 133
 134         sbi->ll_mds_client.cli_data = sbi;
 135         sbi->ll_mds_client.cli_name = "mdc";
 136         sbi->ll_mds_conn = ptlrpc_uuid_to_connection("mds");
 137         if (!sbi->ll_mds_conn) {
 138                 CERROR("cannot find MDS\n");
 139                 GOTO(out_ldlm, sb = NULL);
 140         }
 141
 142         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mds_conn);
 143         if (err) {
 144                 CERROR("cannot connect to MDS: rc = %d\n", err);
 145                 GOTO(out_rpc, sb = NULL);
 146         }
 147
 148         sbi->ll_mds_conn->c_level = LUSTRE_CONN_FULL;
 149
 150         /* XXX: need to store the last_* values somewhere */
 151         err = mdc_connect(&sbi->ll_mds_client, sbi->ll_mds_conn,
 152                           &rootfid, &last_committed, &last_rcvd, &last_xid,
 153                           &request);
 154         if (err) {
 155                 CERROR("cannot mds_connect: rc = %d\n", err);
 156                 GOTO(out_disc, sb = NULL);
 157         }
 158         CDEBUG(D_SUPER, "rootfid %ld\n", (unsigned long)rootfid.id);
 159         sbi->ll_rootino = rootfid.id;
 160
 161         sb->s_maxbytes = 1ULL << 36;
 162         /* XXX get this with a get_info call (like we have in OBDFS),
 163            this info call should return the blocksize of the MDS */
 164         sb->s_blocksize = 4096;
 165         sb->s_blocksize_bits = 12;
 166         sb->s_magic = LL_SUPER_MAGIC;
 167         sb->s_op = &ll_super_operations;
 168
 169         /* make root inode */
 170         err = mdc_getattr(&sbi->ll_mds_client, sbi->ll_mds_conn,
 171                           sbi->ll_rootino, S_IFDIR,
 172                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 173         if (err) {
 174                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 175                 GOTO(out_mdc, sb = NULL);
 176         }
 177
 178         /* initialize committed transaction callback daemon */
 179         spin_lock_init(&sbi->ll_commitcbd_lock);
 180         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 181         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 182         sbi->ll_commitcbd_flags = 0;
 183         err = ll_commitcbd_setup(sbi);
 184         if (err) {
 185                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 186                 GOTO(out_mdc, sb = NULL);
 187         }
 188
 189         md.body = lustre_msg_buf(request->rq_repmsg, 0);
 190         md.obdo = NULL;
 191         root = iget4(sb, sbi->ll_rootino, NULL, &md);
 192
 193         if (root) {
 194                 sb->s_root = d_alloc_root(root);
 195         } else {
 196                 CERROR("lustre_lite: bad iget4 for root\n");
 197                 GOTO(out_cdb, sb = NULL);
 198         }
 199
 200         ptlrpc_free_req(request);
 201
 202 out_dev:
 203         if (device)
 204                 OBD_FREE(device, strlen(device) + 1);
 205         if (version)
 206                 OBD_FREE(version, strlen(version) + 1);
 207
 208         RETURN(sb);
 209
 210 out_cdb:
 211         ll_commitcbd_cleanup(sbi);
 212 out_mdc:
 213         ptlrpc_cleanup_client(&sbi->ll_mds_client);
 214 out_disc:
 215         ptlrpc_free_req(request);
 216 out_rpc:
 217         ptlrpc_put_connection(sbi->ll_mds_conn);
 218 out_ldlm:
 219         ldlm_namespace_free(sbi->ll_namespace);
 220 out_obd:
 221         obd_disconnect(&sbi->ll_conn);
 222 out_free:
 223         OBD_FREE(sbi, sizeof(*sbi));
 224
 225         MOD_DEC_USE_COUNT;
 226         goto out_dev;
 227 } /* ll_read_super */
 228
 229 static void ll_put_super(struct super_block *sb)
 230 {
 231         struct ll_sb_info *sbi = ll_s2sbi(sb);
 232         ENTRY;
 233         ll_commitcbd_cleanup(sbi);
 234         ptlrpc_cleanup_client(&sbi->ll_mds_client);
 235         ptlrpc_put_connection(sbi->ll_mds_conn);
 236         ldlm_namespace_free(sbi->ll_namespace);
 237         obd_disconnect(&sbi->ll_conn);
 238         OBD_FREE(sbi, sizeof(*sbi));
 239         MOD_DEC_USE_COUNT;
 240         EXIT;
 241 } /* ll_put_super */
 242
 243 static void ll_clear_inode(struct inode *inode)
 244 {
 245         if (atomic_read(&inode->i_count) == 0) {
 246                 struct obdo *oa = ll_i2info(inode)->lli_obdo;
 247                 if (oa) {
 248                         obdo_free(oa);
 249                         ll_i2info(inode)->lli_obdo = NULL;
 250                 }
 251                 if (ll_i2info(inode)->lli_symlink_name) {
 252                         OBD_FREE(ll_i2info(inode)->lli_symlink_name,
 253                                  strlen(ll_i2info(inode)->lli_symlink_name)+ 1);
 254                         ll_i2info(inode)->lli_symlink_name = NULL;
 255                 }
 256         }
 257 }
 258
 259 static void ll_delete_inode(struct inode *inode)
 260 {
 261         if (S_ISREG(inode->i_mode)) {
 262                 int err;
 263                 struct obdo *oa;
 264                 oa = ll_i2info(inode)->lli_obdo;
 265
 266                 if (!oa) {
 267                         CERROR("no memory\n");
 268                         GOTO(out, -ENOMEM);
 269                 }
 270
 271                 err = obd_destroy(ll_i2obdconn(inode), oa);
 272                 CDEBUG(D_INODE, "obd destroy of %Ld error %d\n",
 273                        (unsigned long long)oa->o_id, err);
 274                 obdo_free(oa);
 275         }
 276 out:
 277         clear_inode(inode);
 278 }
 279
 280 /* like inode_setattr, but doesn't mark the inode dirty */
 281 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 282 {
 283         unsigned int ia_valid = attr->ia_valid;
 284         int error = 0;
 285
 286         if ((ia_valid & ATTR_SIZE) && trunc ) {
 287                 error = vmtruncate(inode, attr->ia_size);
 288                 if (error)
 289                         goto out;
 290         } else if (ia_valid & ATTR_SIZE) {
 291                 inode->i_size = attr->ia_size;
 292         }
 293
 294         if (ia_valid & ATTR_UID)
 295                 inode->i_uid = attr->ia_uid;
 296         if (ia_valid & ATTR_GID)
 297                 inode->i_gid = attr->ia_gid;
 298         if (ia_valid & ATTR_ATIME)
 299                 inode->i_atime = attr->ia_atime;
 300         if (ia_valid & ATTR_MTIME)
 301                 inode->i_mtime = attr->ia_mtime;
 302         if (ia_valid & ATTR_CTIME)
 303                 inode->i_ctime = attr->ia_ctime;
 304         if (ia_valid & ATTR_MODE) {
 305                 inode->i_mode = attr->ia_mode;
 306                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 307                         inode->i_mode &= ~S_ISGID;
 308         }
 309 out:
 310         return error;
 311 }
 312
 313 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 314 {
 315         struct ptlrpc_request *request = NULL;
 316         struct ll_sb_info *sbi = ll_i2sbi(inode);
 317         int err;
 318
 319         ENTRY;
 320
 321         /* change incore inode */
 322         ll_attr2inode(inode, attr, do_trunc);
 323
 324         err = mdc_setattr(&sbi->ll_mds_client, sbi->ll_mds_conn, inode, attr,
 325                           &request);
 326         if (err)
 327                 CERROR("mdc_setattr fails (%d)\n", err);
 328
 329         ptlrpc_req_finished(request);
 330
 331         RETURN(err);
 332 }
 333
 334 int ll_setattr(struct dentry *de, struct iattr *attr)
 335 {
 336         return ll_inode_setattr(de->d_inode, attr, 1);
 337 }
 338
 339 static int ll_statfs(struct super_block *sb, struct statfs *buf)
 340 {
 341         struct statfs tmp;
 342         int err;
 343         ENTRY;
 344
 345         err = obd_statfs(&ll_s2sbi(sb)->ll_conn, &tmp);
 346         if (err) {
 347                 CERROR("obd_statfs fails (%d)\n", err);
 348                 RETURN(err);
 349         }
 350         memcpy(buf, &tmp, sizeof(*buf));
 351 #if 0
 352         err = mdc_statfs(&sbi->ll_mds_client, sbi->ll_mds_conn, &tmp,
 353                           &request);
 354         if (err) {
 355                 CERROR("obd_statfs fails (%d)\n", err);
 356                 RETURN(err);
 357         }
 358         if (tmp.f_files < buf->f_files)
 359                 buf->f_files = tmp.f_files;
 360         if (tmp.f_ffree < buf->f_ffree)
 361                 buf->f_ffree = tmp.f_ffree;
 362         buf->f_namelen = tmp.f_namelen;
 363 #endif
 364         CDEBUG(D_SUPER, "statfs returns avail %ld\n", tmp.f_bavail);
 365
 366         RETURN(err);
 367 }
 368
 369 static void inline ll_to_inode(struct inode *dst, struct ll_inode_md *md)
 370 {
 371         struct mds_body *body = md->body;
 372         struct ll_inode_info *ii = ll_i2info(dst);
 373
 374         /* core attributes first */
 375         if ( body->valid & OBD_MD_FLID )
 376                 dst->i_ino = body->ino;
 377         if ( body->valid & OBD_MD_FLATIME )
 378                 dst->i_atime = body->atime;
 379         if ( body->valid & OBD_MD_FLMTIME )
 380                 dst->i_mtime = body->mtime;
 381         if ( body->valid & OBD_MD_FLCTIME )
 382                 dst->i_ctime = body->ctime;
 383         if ( body->valid & OBD_MD_FLSIZE )
 384                 dst->i_size = body->size;
 385         if ( body->valid & OBD_MD_FLMODE )
 386                 dst->i_mode = body->mode;
 387         if ( body->valid & OBD_MD_FLUID )
 388                 dst->i_uid = body->uid;
 389         if ( body->valid & OBD_MD_FLGID )
 390                 dst->i_gid = body->gid;
 391         if ( body->valid & OBD_MD_FLFLAGS )
 392                 dst->i_flags = body->flags;
 393         if ( body->valid & OBD_MD_FLNLINK )
 394                 dst->i_nlink = body->nlink;
 395         if ( body->valid & OBD_MD_FLGENER )
 396                 dst->i_generation = body->generation;
 397
 398         /* this will become more elaborate for striping etc */
 399         if (md->obdo != NULL) {
 400                 ii->lli_obdo = obdo_alloc();
 401                 memcpy(ii->lli_obdo, md->obdo, sizeof(*md->obdo));
 402         }
 403 #if 0
 404
 405         if (obdo_has_inline(oa)) {
 406                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 407                     S_ISFIFO(inode->i_mode)) {
 408                         obd_rdev rdev = *((obd_rdev *)oa->o_inline);
 409                         CDEBUG(D_INODE,
 410                                "copying device %x from obdo to inode\n", rdev);
 411                         init_special_inode(inode, inode->i_mode, rdev);
 412                 } else {
 413                         CDEBUG(D_INFO, "copying inline from obdo to inode\n");
 414                         memcpy(oinfo->lli_inline, oa->o_inline, OBD_INLINESZ);
 415                 }
 416                 oinfo->lli_flags |= OBD_FL_INLINEDATA;
 417         }
 418 #endif
 419 } /* ll_to_inode */
 420
 421 static inline void ll_read_inode2(struct inode *inode, void *opaque)
 422 {
 423         struct ll_inode_md *md = opaque;
 424
 425         ENTRY;
 426         ll_to_inode(inode, md);
 427
 428         /* OIDEBUG(inode); */
 429
 430         if (S_ISREG(inode->i_mode)) {
 431                 inode->i_op = &ll_file_inode_operations;
 432                 inode->i_fop = &ll_file_operations;
 433                 inode->i_mapping->a_ops = &ll_aops;
 434                 EXIT;
 435         } else if (S_ISDIR(inode->i_mode)) {
 436                 inode->i_op = &ll_dir_inode_operations;
 437                 inode->i_fop = &ll_dir_operations;
 438                 inode->i_mapping->a_ops = &ll_dir_aops;
 439                 EXIT;
 440         } else if (S_ISLNK(inode->i_mode)) {
 441                 inode->i_op = &ll_fast_symlink_inode_operations;
 442                 EXIT;
 443         } else {
 444                 init_special_inode(inode, inode->i_mode,
 445                                    ((int *)ll_i2info(inode)->lli_inline)[0]);
 446                 EXIT;
 447         }
 448
 449         return;
 450 }
 451
 452 /* exported operations */
 453 struct super_operations ll_super_operations =
 454 {
 455         read_inode2: ll_read_inode2,
 456         clear_inode: ll_clear_inode,
 457         delete_inode: ll_delete_inode,
 458         put_super: ll_put_super,
 459         statfs: ll_statfs
 460 };
 461
 462 struct file_system_type lustre_lite_fs_type = {
 463         "lustre_lite", 0, ll_read_super, NULL
 464 };
 465
 466 static int __init init_lustre_lite(void)
 467 {
 468         printk(KERN_INFO "Lustre Lite 0.0.1, braam@clusterfs.com\n");
 469         ll_file_data_slab = kmem_cache_create("ll_file_data",
 470                                               sizeof(struct ll_file_data), 0,
 471                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 472         if (ll_file_data_slab == NULL)
 473                 return -ENOMEM;
 474         return register_filesystem(&lustre_lite_fs_type);
 475 }
 476
 477 static void __exit exit_lustre_lite(void)
 478 {
 479         unregister_filesystem(&lustre_lite_fs_type);
 480         kmem_cache_destroy(ll_file_data_slab);
 481 }
 482
 483 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 484 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 485 MODULE_LICENSE("GPL");
 486
 487 module_init(init_lustre_lite);
 488 module_exit(exit_lustre_lite);