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) 2002 Cluster File Systems, Inc.
  10  */
  11
  12 #define DEBUG_SUBSYSTEM S_LLITE
  13
  14 #include <linux/module.h>
  15 #include <linux/random.h>
  16 #include <linux/lustre_lite.h>
  17 #include <linux/lustre_ha.h>
  18 #include <linux/obd_lov.h>
  19 #include <linux/lustre_dlm.h>
  20 #include <linux/init.h>
  21
  22 kmem_cache_t *ll_file_data_slab;
  23 extern struct address_space_operations ll_aops;
  24 extern struct address_space_operations ll_dir_aops;
  25 struct super_operations ll_super_operations;
  26
  27 extern int ll_recover(struct recovd_data *, int);
  28 extern int ll_commitcbd_setup(struct ll_sb_info *);
  29 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  30
  31 static char *ll_read_opt(const char *opt, char *data)
  32 {
  33         char *value;
  34         char *retval;
  35         ENTRY;
  36
  37         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  38         if ( strncmp(opt, data, strlen(opt)) )
  39                 RETURN(NULL);
  40         if ( (value = strchr(data, '=')) == NULL )
  41                 RETURN(NULL);
  42
  43         value++;
  44         OBD_ALLOC(retval, strlen(value) + 1);
  45         if ( !retval ) {
  46                 CERROR("out of memory!\n");
  47                 RETURN(NULL);
  48         }
  49
  50         memcpy(retval, value, strlen(value)+1);
  51         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  52         RETURN(retval);
  53 }
  54
  55 static int ll_set_opt(const char *opt, char *data, int fl)
  56 {
  57         ENTRY;
  58
  59         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  60         if ( strncmp(opt, data, strlen(opt)) )
  61                 RETURN(0);
  62         else
  63                 RETURN(fl);
  64 }
  65
  66 static void ll_options(char *options, char **ost, char **mds, int *flags)
  67 {
  68         char *this_char;
  69         ENTRY;
  70
  71         if (!options) {
  72                 EXIT;
  73                 return;
  74         }
  75
  76         for (this_char = strtok (options, ",");
  77              this_char != NULL;
  78              this_char = strtok (NULL, ",")) {
  79                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
  80                 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
  81                      (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
  82                      (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
  83                       ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
  84                         continue;
  85         }
  86         EXIT;
  87 }
  88
  89 #ifndef log2
  90 #define log2(n) ffz(~(n))
  91 #endif
  92 inline int ll_mds_easize(struct super_block *sb)
  93 {
  94         struct client_obd *mdc = sbi2mdc(ll_s2sbi(sb));
  95         return mdc->cl_max_mds_easize;
  96 }
  97
  98 inline int ll_lov_easize(struct lov_stripe_md *md)
  99 {
 100         return sizeof(*md) + md->lmd_stripe_count *
 101                 sizeof(struct lov_oinfo);
 102 }
 103
 104 static struct super_block * ll_read_super(struct super_block *sb,
 105                                           void *data, int silent)
 106 {
 107         struct inode *root = 0;
 108         struct obd_device *obd;
 109         struct ll_sb_info *sbi;
 110         char *osc = NULL;
 111         char *mdc = NULL;
 112         int err;
 113         struct ll_fid rootfid;
 114         struct obd_statfs osfs;
 115         __u64 last_committed;
 116         __u64 last_xid;
 117         struct ptlrpc_request *request = NULL;
 118         struct ll_inode_md md;
 119         class_uuid_t uuid;
 120
 121         ENTRY;
 122         MOD_INC_USE_COUNT;
 123
 124         OBD_ALLOC(sbi, sizeof(*sbi));
 125         if (!sbi) {
 126                 MOD_DEC_USE_COUNT;
 127                 RETURN(NULL);
 128         }
 129
 130         generate_random_uuid(uuid);
 131         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 132
 133         sb->u.generic_sbp = sbi;
 134
 135         ll_options(data, &osc, &mdc, &sbi->ll_flags);
 136
 137         if (!osc) {
 138                 CERROR("no osc\n");
 139                 GOTO(out_free, sb = NULL);
 140         }
 141
 142         if (!mdc) {
 143                 CERROR("no mdc\n");
 144                 GOTO(out_free, sb = NULL);
 145         }
 146
 147         obd = class_uuid2obd(mdc);
 148         if (!obd) {
 149                 CERROR("MDC %s: not setup or attached\n", mdc);
 150                 GOTO(out_free, sb = NULL);
 151         }
 152
 153         err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid);
 154         if (err) {
 155                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 156                 GOTO(out_free, sb = NULL);
 157         }
 158
 159 #warning Peter: is this the right place to raise the connection level?
 160         sbi2mdc(sbi)->cl_import.imp_connection->c_level = LUSTRE_CONN_FULL;
 161
 162         obd = class_uuid2obd(osc);
 163         if (!obd) {
 164                 CERROR("OSC %s: not setup or attached\n", osc);
 165                 GOTO(out_mdc, sb = NULL);
 166         }
 167         err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid);
 168         if (err) {
 169                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 170                 GOTO(out_mdc, sb = NULL);
 171         }
 172
 173         /* XXX: need to store the last_* values somewhere */
 174         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
 175                             &last_xid, &request);
 176         ptlrpc_req_finished(request);
 177         if (err) {
 178                 CERROR("cannot mds_connect: rc = %d\n", err);
 179                 GOTO(out_mdc, sb = NULL);
 180         }
 181         CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
 182         sbi->ll_rootino = rootfid.id;
 183
 184         memset(&osfs, 0, sizeof(osfs));
 185         request = NULL;
 186         err = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 187         ptlrpc_req_finished(request);
 188         sb->s_blocksize = osfs.os_bsize;
 189         sb->s_blocksize_bits = log2(osfs.os_bsize);
 190         sb->s_magic = LL_SUPER_MAGIC;
 191         sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
 192
 193         sb->s_op = &ll_super_operations;
 194
 195         /* make root inode */
 196         request = NULL;
 197         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
 198                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 199         if (err) {
 200                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 201                 GOTO(out_request, sb = NULL);
 202         }
 203
 204         /* initialize committed transaction callback daemon */
 205         spin_lock_init(&sbi->ll_commitcbd_lock);
 206         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 207         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 208         sbi->ll_commitcbd_flags = 0;
 209         err = ll_commitcbd_setup(sbi);
 210         if (err) {
 211                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 212                 GOTO(out_request, sb = NULL);
 213         }
 214
 215         md.body = lustre_msg_buf(request->rq_repmsg, 0);
 216         md.md = NULL;
 217         root = iget4(sb, sbi->ll_rootino, NULL, &md);
 218
 219         if (root) {
 220                 sb->s_root = d_alloc_root(root);
 221         } else {
 222                 CERROR("lustre_lite: bad iget4 for root\n");
 223                 GOTO(out_cdb, sb = NULL);
 224         }
 225
 226         ptlrpc_req_finished(request);
 227
 228 out_dev:
 229         if (mdc)
 230                 OBD_FREE(mdc, strlen(mdc) + 1);
 231         if (osc)
 232                 OBD_FREE(osc, strlen(osc) + 1);
 233
 234         RETURN(sb);
 235
 236 out_cdb:
 237         ll_commitcbd_cleanup(sbi);
 238 out_request:
 239         ptlrpc_req_finished(request);
 240         obd_disconnect(&sbi->ll_osc_conn);
 241 out_mdc:
 242         obd_disconnect(&sbi->ll_mdc_conn);
 243 out_free:
 244         OBD_FREE(sbi, sizeof(*sbi));
 245
 246         MOD_DEC_USE_COUNT;
 247         goto out_dev;
 248 } /* ll_read_super */
 249
 250 static void ll_put_super(struct super_block *sb)
 251 {
 252         struct ll_sb_info *sbi = ll_s2sbi(sb);
 253         ENTRY;
 254         ll_commitcbd_cleanup(sbi);
 255         obd_disconnect(&sbi->ll_osc_conn);
 256         obd_disconnect(&sbi->ll_mdc_conn);
 257         OBD_FREE(sbi, sizeof(*sbi));
 258
 259         MOD_DEC_USE_COUNT;
 260         EXIT;
 261 } /* ll_put_super */
 262
 263 static void ll_clear_inode(struct inode *inode)
 264 {
 265         if (atomic_read(&inode->i_count) == 0) {
 266                 struct ll_inode_info *lli = ll_i2info(inode);
 267                 struct lov_stripe_md *md = lli->lli_smd;
 268                 char *symlink_name = lli->lli_symlink_name;
 269
 270                 if (md) {
 271                         int size = ll_lov_easize(md);
 272                         OBD_FREE(md, size);
 273                         lli->lli_smd = NULL;
 274                 }
 275                 if (symlink_name) {
 276                         OBD_FREE(symlink_name, strlen(symlink_name) + 1);
 277                         lli->lli_symlink_name = NULL;
 278                 }
 279         }
 280 }
 281
 282 static void ll_delete_inode(struct inode *inode)
 283 {
 284         if (S_ISREG(inode->i_mode)) {
 285                 int err;
 286                 struct obdo *oa;
 287                 struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
 288
 289                 if (!md)
 290                         GOTO(out, -EINVAL);
 291
 292                 if (md->lmd_object_id == 0) {
 293                         CERROR("This really happens\n");
 294                         /* No obdo was ever created */
 295                         GOTO(out, 0);
 296                 }
 297
 298                 oa = obdo_alloc();
 299                 if (oa == NULL)
 300                         GOTO(out, -ENOMEM);
 301
 302                 oa->o_id = md->lmd_object_id;
 303                 oa->o_easize = md->lmd_mds_easize;
 304                 oa->o_mode = inode->i_mode;
 305                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLMODE;
 306
 307                 err = obd_destroy(ll_i2obdconn(inode), oa, md);
 308                 obdo_free(oa);
 309                 CDEBUG(D_SUPER, "obd destroy of "LPD64" error %d\n",
 310                        md->lmd_object_id, err);
 311         }
 312 out:
 313         clear_inode(inode);
 314 }
 315
 316 /* like inode_setattr, but doesn't mark the inode dirty */
 317 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 318 {
 319         unsigned int ia_valid = attr->ia_valid;
 320         int error = 0;
 321
 322         if ((ia_valid & ATTR_SIZE) && trunc) {
 323                 error = vmtruncate(inode, attr->ia_size);
 324                 if (error)
 325                         goto out;
 326         } else if (ia_valid & ATTR_SIZE)
 327                 inode->i_size = attr->ia_size;
 328
 329         if (ia_valid & ATTR_UID)
 330                 inode->i_uid = attr->ia_uid;
 331         if (ia_valid & ATTR_GID)
 332                 inode->i_gid = attr->ia_gid;
 333         if (ia_valid & ATTR_ATIME)
 334                 inode->i_atime = attr->ia_atime;
 335         if (ia_valid & ATTR_MTIME)
 336                 inode->i_mtime = attr->ia_mtime;
 337         if (ia_valid & ATTR_CTIME)
 338                 inode->i_ctime = attr->ia_ctime;
 339         if (ia_valid & ATTR_MODE) {
 340                 inode->i_mode = attr->ia_mode;
 341                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 342                         inode->i_mode &= ~S_ISGID;
 343         }
 344 out:
 345         return error;
 346 }
 347
 348 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 349 {
 350         struct ptlrpc_request *request = NULL;
 351         struct ll_sb_info *sbi = ll_i2sbi(inode);
 352         int err;
 353
 354         ENTRY;
 355
 356         /* change incore inode */
 357         ll_attr2inode(inode, attr, do_trunc);
 358
 359         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 360         if (err)
 361                 CERROR("mdc_setattr fails (%d)\n", err);
 362
 363         ptlrpc_req_finished(request);
 364
 365         RETURN(err);
 366 }
 367
 368 int ll_setattr(struct dentry *de, struct iattr *attr)
 369 {
 370         int rc = inode_change_ok(de->d_inode, attr);
 371
 372         if (rc)
 373                 return rc;
 374
 375         return ll_inode_setattr(de->d_inode, attr, 1);
 376 }
 377
 378 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 379 {
 380         struct ptlrpc_request *request = NULL;
 381         struct ll_sb_info *sbi = ll_s2sbi(sb);
 382         struct obd_statfs osfs;
 383         int rc;
 384         ENTRY;
 385
 386         memset(sfs, 0, sizeof(*sfs));
 387         rc = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 388         statfs_unpack(sfs, &osfs);
 389         ptlrpc_req_finished(request);
 390         if (rc)
 391                 CERROR("mdc_statfs fails: rc = %d\n", rc);
 392         else
 393                 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
 394                        " objects "LPU64"/"LPU64"\n",
 395                        osfs.os_bavail, osfs.os_blocks,
 396                        osfs.os_ffree, osfs.os_files);
 397
 398         /* temporary until mds_statfs returns statfs info for all OSTs */
 399         if (!rc) {
 400                 struct statfs obd_sfs;
 401
 402                 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
 403                 statfs_unpack(&obd_sfs, &osfs);
 404                 if (rc) {
 405                         CERROR("obd_statfs fails: rc = %d\n", rc);
 406                         GOTO(out, rc);
 407                 }
 408                 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
 409                        " objects "LPU64"/"LPU64"\n",
 410                        osfs.os_bavail, osfs.os_blocks,
 411                        osfs.os_ffree, osfs.os_files);
 412
 413                 sfs->f_bfree = obd_sfs.f_bfree;
 414                 sfs->f_bavail = obd_sfs.f_bavail;
 415                 sfs->f_blocks = obd_sfs.f_blocks;
 416                 if (obd_sfs.f_ffree < sfs->f_ffree)
 417                         sfs->f_ffree = obd_sfs.f_ffree;
 418         }
 419
 420 out:
 421         RETURN(rc);
 422 }
 423
 424 static void ll_read_inode2(struct inode *inode, void *opaque)
 425 {
 426         struct ll_inode_md *md = opaque;
 427         struct mds_body *body = md->body;
 428         struct ll_inode_info *lli = ll_i2info(inode);
 429         ENTRY;
 430
 431         sema_init(&lli->lli_open_sem, 1);
 432
 433         /* core attributes first */
 434         if (body->valid & OBD_MD_FLID)
 435                 inode->i_ino = body->ino;
 436         if (body->valid & OBD_MD_FLATIME)
 437                 inode->i_atime = body->atime;
 438         if (body->valid & OBD_MD_FLMTIME)
 439                 inode->i_mtime = body->mtime;
 440         if (body->valid & OBD_MD_FLCTIME)
 441                 inode->i_ctime = body->ctime;
 442         if (body->valid & OBD_MD_FLMODE)
 443                 inode->i_mode = body->mode;
 444         if (body->valid & OBD_MD_FLUID)
 445                 inode->i_uid = body->uid;
 446         if (body->valid & OBD_MD_FLGID)
 447                 inode->i_gid = body->gid;
 448         if (body->valid & OBD_MD_FLFLAGS)
 449                 inode->i_flags = body->flags;
 450         if (body->valid & OBD_MD_FLNLINK)
 451                 inode->i_nlink = body->nlink;
 452         if (body->valid & OBD_MD_FLGENER)
 453                 inode->i_generation = body->generation;
 454         if (body->valid & OBD_MD_FLRDEV)
 455                 inode->i_rdev = body->extra;
 456         if (body->valid & OBD_MD_FLSIZE)
 457                 inode->i_size = body->size;
 458
 459         //if (body->valid & OBD_MD_FLEASIZE)
 460         if (md && md->md) {
 461                 struct lov_mds_md *smd = md->md;
 462                 int size;
 463                 if (md->md->lmd_easize != ll_mds_easize(inode->i_sb)) {
 464                         CERROR("Striping metadata size error %ld\n",
 465                                inode->i_ino);
 466                         LBUG();
 467                 }
 468                 size = ll_lov_easize(lli->lli_smd);
 469                 OBD_ALLOC(lli->lli_smd, size);
 470                 if (!lli->lli_smd) {
 471                         CERROR("No memory for %d\n", size);
 472                         LBUG();
 473                 }
 474                 lov_unpackmd(lli->lli_smd, smd);
 475         } else {
 476                 lli->lli_smd = NULL;
 477         }
 478
 479         /* Get the authoritative file size */
 480         if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
 481                 int rc;
 482
 483                 rc = ll_file_size(inode, lli->lli_smd);
 484                 if (rc) {
 485                         CERROR("ll_file_size: %d\n", rc);
 486                         /* FIXME: need to somehow prevent inode creation */
 487                         LBUG();
 488                 }
 489         }
 490
 491         /* OIDEBUG(inode); */
 492
 493         if (S_ISREG(inode->i_mode)) {
 494                 inode->i_op = &ll_file_inode_operations;
 495                 inode->i_fop = &ll_file_operations;
 496                 inode->i_mapping->a_ops = &ll_aops;
 497                 EXIT;
 498         } else if (S_ISDIR(inode->i_mode)) {
 499                 inode->i_op = &ll_dir_inode_operations;
 500                 inode->i_fop = &ll_dir_operations;
 501                 inode->i_mapping->a_ops = &ll_dir_aops;
 502                 EXIT;
 503         } else if (S_ISLNK(inode->i_mode)) {
 504                 inode->i_op = &ll_fast_symlink_inode_operations;
 505                 EXIT;
 506         } else {
 507                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 508                 EXIT;
 509         }
 510
 511         return;
 512 }
 513
 514 /* exported operations */
 515 struct super_operations ll_super_operations =
 516 {
 517         read_inode2: ll_read_inode2,
 518         clear_inode: ll_clear_inode,
 519         delete_inode: ll_delete_inode,
 520         put_super: ll_put_super,
 521         statfs: ll_statfs
 522 };
 523
 524 struct file_system_type lustre_lite_fs_type = {
 525         "lustre_lite", 0, ll_read_super, NULL
 526 };
 527
 528 static int __init init_lustre_lite(void)
 529 {
 530         printk(KERN_INFO "Lustre Lite 0.0.1, info@clusterfs.com\n");
 531         ll_file_data_slab = kmem_cache_create("ll_file_data",
 532                                               sizeof(struct ll_file_data), 0,
 533                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 534         if (ll_file_data_slab == NULL)
 535                 return -ENOMEM;
 536         return register_filesystem(&lustre_lite_fs_type);
 537 }
 538
 539 static void __exit exit_lustre_lite(void)
 540 {
 541         unregister_filesystem(&lustre_lite_fs_type);
 542         kmem_cache_destroy(ll_file_data_slab);
 543 }
 544
 545 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 546 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 547 MODULE_LICENSE("GPL");
 548
 549 module_init(init_lustre_lite);
 550 module_exit(exit_lustre_lite);