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 **ost, char **mds)
  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 ( (!*ost && (*ost = ll_read_opt("ost", this_char)))||
  72                      (!*mds && (*mds = ll_read_opt("mds", 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 *ost = NULL;
  84         char *mds = 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, &ost, &mds);
 105
 106         if (!ost) {
 107                 CERROR("no ost\n");
 108                 GOTO(out_free, sb = NULL);
 109         }
 110
 111         if (!mds) {
 112                 CERROR("no mds\n");
 113                 GOTO(out_free, sb = NULL);
 114         }
 115
 116         devno = simple_strtoul(mds, NULL, 0);
 117         if (devno >= MAX_OBD_DEVICES) {
 118                 CERROR("devno of %s too high\n", mds);
 119                 GOTO(out_free, sb = NULL);
 120         }
 121
 122         /* First the MDS since an LOV requires an the MDC connection
 123            to find its descriptor */
 124 #if 0
 125         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mds_conn);
 126         if (err) {
 127                 CERROR("cannot connect to MDS: rc = %d\n", err);
 128                 GOTO(out_rpc, sb = NULL);
 129         }
 130 #endif
 131
 132         err = obd_connect(&sbi->ll_mdc_conn, &obd_dev[devno]);
 133         if (err) {
 134                 CERROR("cannot connect to %s: rc = %d\n", mds, err);
 135                 GOTO(out_free, sb = NULL);
 136         }
 137         sbi2mdc(sbi)->mdc_conn->c_level = LUSTRE_CONN_FULL;
 138
 139         /* now the OST, which could be an LOV */
 140         devno = simple_strtoul(ost, NULL, 0);
 141         if (devno >= MAX_OBD_DEVICES) {
 142                 CERROR("devno of %s too high\n", ost);
 143                 GOTO(out_free, sb = NULL);
 144         }
 145         err = obd_connect(&sbi->ll_osc_conn, &obd_dev[devno]);
 146         if (err) {
 147                 CERROR("cannot connect to %s: rc = %d\n", ost, err);
 148                 GOTO(out_free, sb = NULL);
 149         }
 150
 151         /* XXX: need to store the last_* values somewhere */
 152         err = mdc_getstatus(&sbi->ll_mdc_conn,
 153                             &rootfid, &last_committed,
 154                             &last_rcvd, &last_xid,
 155                             &request);
 156         if (err) {
 157                 CERROR("cannot mds_connect: rc = %d\n", err);
 158                 GOTO(out_disc, sb = NULL);
 159         }
 160         CDEBUG(D_SUPER, "rootfid %ld\n", (unsigned long)rootfid.id);
 161         sbi->ll_rootino = rootfid.id;
 162
 163         sb->s_maxbytes = 1ULL << 36;
 164         /* XXX get this with a get_info call (like we have in OBDFS),
 165            this info call should return the blocksize of the MDS (statfs?) */
 166         sb->s_blocksize = 4096;
 167         sb->s_blocksize_bits = 12;
 168         sb->s_magic = LL_SUPER_MAGIC;
 169         sb->s_op = &ll_super_operations;
 170
 171         /* make root inode */
 172         err = mdc_getattr(&sbi->ll_mdc_conn,
 173                           sbi->ll_rootino, S_IFDIR,
 174                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 175         if (err) {
 176                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 177                 GOTO(out_mdc, sb = NULL);
 178         }
 179
 180         /* initialize committed transaction callback daemon */
 181         spin_lock_init(&sbi->ll_commitcbd_lock);
 182         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 183         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 184         sbi->ll_commitcbd_flags = 0;
 185         err = ll_commitcbd_setup(sbi);
 186         if (err) {
 187                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 188                 GOTO(out_mdc, sb = NULL);
 189         }
 190
 191         md.body = lustre_msg_buf(request->rq_repmsg, 0);
 192         md.obdo = NULL;
 193         root = iget4(sb, sbi->ll_rootino, NULL, &md);
 194
 195         if (root) {
 196                 sb->s_root = d_alloc_root(root);
 197         } else {
 198                 CERROR("lustre_lite: bad iget4 for root\n");
 199                 GOTO(out_cdb, sb = NULL);
 200         }
 201
 202         ptlrpc_free_req(request);
 203
 204 out_dev:
 205         if (mds)
 206                 OBD_FREE(mds, strlen(mds) + 1);
 207         if (ost)
 208                 OBD_FREE(ost, strlen(ost) + 1);
 209
 210         RETURN(sb);
 211
 212 out_cdb:
 213         ll_commitcbd_cleanup(sbi);
 214 out_mdc:
 215         obd_disconnect(&sbi->ll_mdc_conn);
 216         obd_disconnect(&sbi->ll_osc_conn);
 217 out_disc:
 218         ptlrpc_free_req(request);
 219         obd_disconnect(&sbi->ll_osc_conn);
 220 out_free:
 221         OBD_FREE(sbi, sizeof(*sbi));
 222
 223         MOD_DEC_USE_COUNT;
 224         goto out_dev;
 225 } /* ll_read_super */
 226
 227 static void ll_put_super(struct super_block *sb)
 228 {
 229         struct ll_sb_info *sbi = ll_s2sbi(sb);
 230         ENTRY;
 231         ll_commitcbd_cleanup(sbi);
 232         obd_disconnect(&sbi->ll_osc_conn);
 233         obd_disconnect(&sbi->ll_mdc_conn);
 234         OBD_FREE(sbi, sizeof(*sbi));
 235
 236         MOD_DEC_USE_COUNT;
 237         EXIT;
 238 } /* ll_put_super */
 239
 240 static void ll_clear_inode(struct inode *inode)
 241 {
 242         if (atomic_read(&inode->i_count) == 0) {
 243                 struct obdo *oa = ll_i2info(inode)->lli_obdo;
 244                 if (oa) {
 245                         obdo_free(oa);
 246                         ll_i2info(inode)->lli_obdo = NULL;
 247                 }
 248                 if (ll_i2info(inode)->lli_symlink_name) {
 249                         OBD_FREE(ll_i2info(inode)->lli_symlink_name,
 250                                  strlen(ll_i2info(inode)->lli_symlink_name)+ 1);
 251                         ll_i2info(inode)->lli_symlink_name = NULL;
 252                 }
 253         }
 254 }
 255
 256 static void ll_delete_inode(struct inode *inode)
 257 {
 258         if (S_ISREG(inode->i_mode)) {
 259                 int err;
 260                 struct obdo *oa;
 261                 oa = ll_i2info(inode)->lli_obdo;
 262
 263                 if (!oa)
 264                         GOTO(out, -EINVAL);
 265
 266                 if (oa->o_id == 0)
 267                         /* No obdo was ever created */
 268                         GOTO(out, 0);
 269
 270                 err = obd_destroy(ll_i2obdconn(inode), oa);
 271                 CDEBUG(D_INODE, "obd destroy of %Ld error %d\n",
 272                        (unsigned long long)oa->o_id, err);
 273         }
 274 out:
 275         clear_inode(inode);
 276 }
 277
 278 /* like inode_setattr, but doesn't mark the inode dirty */
 279 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 280 {
 281         unsigned int ia_valid = attr->ia_valid;
 282         int error = 0;
 283
 284         if ((ia_valid & ATTR_SIZE) && trunc) {
 285                 error = vmtruncate(inode, attr->ia_size);
 286                 if (error)
 287                         goto out;
 288         } else if (ia_valid & ATTR_SIZE)
 289                 inode->i_size = attr->ia_size;
 290
 291         if (ia_valid & ATTR_UID)
 292                 inode->i_uid = attr->ia_uid;
 293         if (ia_valid & ATTR_GID)
 294                 inode->i_gid = attr->ia_gid;
 295         if (ia_valid & ATTR_ATIME)
 296                 inode->i_atime = attr->ia_atime;
 297         if (ia_valid & ATTR_MTIME)
 298                 inode->i_mtime = attr->ia_mtime;
 299         if (ia_valid & ATTR_CTIME)
 300                 inode->i_ctime = attr->ia_ctime;
 301         if (ia_valid & ATTR_MODE) {
 302                 inode->i_mode = attr->ia_mode;
 303                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 304                         inode->i_mode &= ~S_ISGID;
 305         }
 306 out:
 307         return error;
 308 }
 309
 310 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 311 {
 312         struct ptlrpc_request *request = NULL;
 313         struct ll_sb_info *sbi = ll_i2sbi(inode);
 314         int err;
 315
 316         ENTRY;
 317
 318         /* change incore inode */
 319         ll_attr2inode(inode, attr, do_trunc);
 320
 321         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr,
 322                           &request);
 323         if (err)
 324                 CERROR("mdc_setattr fails (%d)\n", err);
 325
 326         ptlrpc_req_finished(request);
 327
 328         RETURN(err);
 329 }
 330
 331 int ll_setattr(struct dentry *de, struct iattr *attr)
 332 {
 333         int rc = inode_change_ok(de->d_inode, attr);
 334
 335         if (rc)
 336                 return rc;
 337
 338         return ll_inode_setattr(de->d_inode, attr, 1);
 339 }
 340
 341 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 342 {
 343         struct ptlrpc_request *request = NULL;
 344         struct ll_sb_info *sbi = ll_s2sbi(sb);
 345         int err;
 346         ENTRY;
 347
 348         memset(sfs, 0, sizeof(*sfs));
 349         err = mdc_statfs(&sbi->ll_mdc_conn, sfs, &request);
 350         if (err)
 351                 CERROR("obd_statfs fails (%d)\n", err);
 352         else
 353                 CDEBUG(D_SUPER,
 354                        "statfs returns blocks %ld/%ld objects %ld/%ld\n",
 355                        sfs->f_bavail, sfs->f_blocks, sfs->f_files,sfs->f_ffree);
 356
 357         ptlrpc_req_finished(request);
 358
 359         RETURN(err);
 360 }
 361
 362 static void ll_to_inode(struct inode *dst, struct ll_inode_md *md)
 363 {
 364         struct mds_body *body = md->body;
 365         struct ll_inode_info *ii = ll_i2info(dst);
 366
 367         /* core attributes first */
 368         if (body->valid & OBD_MD_FLID)
 369                 dst->i_ino = body->ino;
 370         if (body->valid & OBD_MD_FLATIME)
 371                 dst->i_atime = body->atime;
 372         if (body->valid & OBD_MD_FLMTIME)
 373                 dst->i_mtime = body->mtime;
 374         if (body->valid & OBD_MD_FLCTIME)
 375                 dst->i_ctime = body->ctime;
 376         if (body->valid & OBD_MD_FLSIZE)
 377                 dst->i_size = body->size;
 378         if (body->valid & OBD_MD_FLMODE)
 379                 dst->i_mode = body->mode;
 380         if (body->valid & OBD_MD_FLUID)
 381                 dst->i_uid = body->uid;
 382         if (body->valid & OBD_MD_FLGID)
 383                 dst->i_gid = body->gid;
 384         if (body->valid & OBD_MD_FLFLAGS)
 385                 dst->i_flags = body->flags;
 386         if (body->valid & OBD_MD_FLNLINK)
 387                 dst->i_nlink = body->nlink;
 388         if (body->valid & OBD_MD_FLGENER)
 389                 dst->i_generation = body->generation;
 390
 391         /* this will become more elaborate for striping etc */
 392         if (md->obdo != NULL && md->obdo->o_valid != 0) {
 393                 ii->lli_obdo = obdo_alloc();
 394                 memcpy(ii->lli_obdo, md->obdo, sizeof(*md->obdo));
 395         }
 396 #if 0
 397
 398         if (obdo_has_inline(oa)) {
 399                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 400                     S_ISFIFO(inode->i_mode)) {
 401                         obd_rdev rdev = *((obd_rdev *)oa->o_inline);
 402                         CDEBUG(D_INODE,
 403                                "copying device %x from obdo to inode\n", rdev);
 404                         init_special_inode(inode, inode->i_mode, rdev);
 405                 } else {
 406                         CDEBUG(D_INFO, "copying inline from obdo to inode\n");
 407                         memcpy(oinfo->lli_inline, oa->o_inline, OBD_INLINESZ);
 408                 }
 409                 oinfo->lli_flags |= OBD_FL_INLINEDATA;
 410         }
 411 #endif
 412 } /* ll_to_inode */
 413
 414 static void ll_read_inode2(struct inode *inode, void *opaque)
 415 {
 416         struct ll_inode_md *md = opaque;
 417
 418         ENTRY;
 419         ll_to_inode(inode, md);
 420
 421         /* OIDEBUG(inode); */
 422
 423         if (S_ISREG(inode->i_mode)) {
 424                 inode->i_op = &ll_file_inode_operations;
 425                 inode->i_fop = &ll_file_operations;
 426                 inode->i_mapping->a_ops = &ll_aops;
 427                 EXIT;
 428         } else if (S_ISDIR(inode->i_mode)) {
 429                 inode->i_op = &ll_dir_inode_operations;
 430                 inode->i_fop = &ll_dir_operations;
 431                 inode->i_mapping->a_ops = &ll_dir_aops;
 432                 EXIT;
 433         } else if (S_ISLNK(inode->i_mode)) {
 434                 inode->i_op = &ll_fast_symlink_inode_operations;
 435                 EXIT;
 436         } else {
 437                 init_special_inode(inode, inode->i_mode,
 438                                    ((int *)ll_i2info(inode)->lli_inline)[0]);
 439                 EXIT;
 440         }
 441
 442         return;
 443 }
 444
 445 /* exported operations */
 446 struct super_operations ll_super_operations =
 447 {
 448         read_inode2: ll_read_inode2,
 449         clear_inode: ll_clear_inode,
 450         delete_inode: ll_delete_inode,
 451         put_super: ll_put_super,
 452         statfs: ll_statfs
 453 };
 454
 455 struct file_system_type lustre_lite_fs_type = {
 456         "lustre_lite", 0, ll_read_super, NULL
 457 };
 458
 459 static int __init init_lustre_lite(void)
 460 {
 461         printk(KERN_INFO "Lustre Lite 0.0.1, braam@clusterfs.com\n");
 462         ll_file_data_slab = kmem_cache_create("ll_file_data",
 463                                               sizeof(struct ll_file_data), 0,
 464                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 465         if (ll_file_data_slab == NULL)
 466                 return -ENOMEM;
 467         return register_filesystem(&lustre_lite_fs_type);
 468 }
 469
 470 static void __exit exit_lustre_lite(void)
 471 {
 472         unregister_filesystem(&lustre_lite_fs_type);
 473         kmem_cache_destroy(ll_file_data_slab);
 474 }
 475
 476 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 477 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 478 MODULE_LICENSE("GPL");
 479
 480 module_init(init_lustre_lite);
 481 module_exit(exit_lustre_lite);