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