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 ptlrpc_client *);
  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 void ll_options(char *options, char **ost, char **mds)
  56 {
  57         char *this_char;
  58         ENTRY;
  59
  60         if (!options) {
  61                 EXIT;
  62                 return;
  63         }
  64
  65         for (this_char = strtok (options, ",");
  66              this_char != NULL;
  67              this_char = strtok (NULL, ",")) {
  68                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
  69                 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
  70                      (!*mds && (*mds = ll_read_opt("mdc", this_char))) )
  71                         continue;
  72         }
  73         EXIT;
  74 }
  75
  76 #ifndef log2
  77 #define log2(n) ffz(~(n))
  78 #endif
  79
  80 static struct super_block * ll_read_super(struct super_block *sb,
  81                                           void *data, int silent)
  82 {
  83         struct inode *root = 0;
  84         struct obd_device *obd;
  85         struct ll_sb_info *sbi;
  86         char *osc = NULL;
  87         char *mdc = NULL;
  88         int err;
  89         struct ll_fid rootfid;
  90         struct statfs sfs;
  91         __u64 last_committed;
  92         __u32 last_xid;
  93         struct ptlrpc_request *request = NULL;
  94         struct ll_inode_md md;
  95         class_uuid_t uuid;
  96
  97         ENTRY;
  98         MOD_INC_USE_COUNT;
  99
 100         OBD_ALLOC(sbi, sizeof(*sbi));
 101         if (!sbi) {
 102                 MOD_DEC_USE_COUNT;
 103                 RETURN(NULL);
 104         }
 105
 106         generate_random_uuid(uuid);
 107         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 108
 109         sb->u.generic_sbp = sbi;
 110
 111         ll_options(data, &osc, &mdc);
 112
 113         if (!osc) {
 114                 CERROR("no osc\n");
 115                 GOTO(out_free, sb = NULL);
 116         }
 117
 118         if (!mdc) {
 119                 CERROR("no mdc\n");
 120                 GOTO(out_free, sb = NULL);
 121         }
 122
 123         obd = class_uuid2obd(mdc);
 124         if (!obd) {
 125                 CERROR("MDC %s: not setup or attached\n", mdc);
 126                 GOTO(out_free, sb = NULL);
 127         }
 128
 129 #if 0
 130         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mdc_conn);
 131         if (err) {
 132                 CERROR("cannot connect to MDC: rc = %d\n", err);
 133                 GOTO(out_rpc, sb = NULL);
 134         }
 135 #endif
 136
 137         err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid);
 138         if (err) {
 139                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 140                 GOTO(out_free, sb = NULL);
 141         }
 142         sbi2mdc(sbi)->cl_conn->c_level = LUSTRE_CONN_FULL;
 143
 144         obd = class_uuid2obd(osc);
 145         if (!obd) {
 146                 CERROR("OSC %s: not setup or attached\n", osc);
 147                 GOTO(out_mdc, sb = NULL);
 148         }
 149         err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid);
 150         if (err) {
 151                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 152                 GOTO(out_mdc, sb = NULL);
 153         }
 154
 155         /* XXX: need to store the last_* values somewhere */
 156         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
 157                             &last_xid, &request);
 158         ptlrpc_req_finished(request);
 159         if (err) {
 160                 CERROR("cannot mds_connect: rc = %d\n", err);
 161                 GOTO(out_mdc, sb = NULL);
 162         }
 163         CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
 164         sbi->ll_rootino = rootfid.id;
 165
 166         memset(&sfs, 0, sizeof(sfs));
 167         request = NULL;
 168         err = mdc_statfs(&sbi->ll_mdc_conn, &sfs, &request);
 169         ptlrpc_req_finished(request);
 170         sb->s_blocksize = sfs.f_bsize;
 171         sb->s_blocksize_bits = log2(sfs.f_bsize);
 172         sb->s_magic = LL_SUPER_MAGIC;
 173         sb->s_maxbytes = (1ULL << (32 + 9)) - sfs.f_bsize;
 174
 175         sb->s_op = &ll_super_operations;
 176
 177         /* make root inode */
 178         request = NULL;
 179         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
 180                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 181         if (err) {
 182                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 183                 GOTO(out_request, sb = NULL);
 184         }
 185
 186         /* initialize committed transaction callback daemon */
 187         spin_lock_init(&sbi->ll_commitcbd_lock);
 188         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 189         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 190         sbi->ll_commitcbd_flags = 0;
 191         err = ll_commitcbd_setup(sbi);
 192         if (err) {
 193                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 194                 GOTO(out_request, sb = NULL);
 195         }
 196
 197         md.body = lustre_msg_buf(request->rq_repmsg, 0);
 198         md.md = NULL;
 199         root = iget4(sb, sbi->ll_rootino, NULL, &md);
 200
 201         if (root) {
 202                 sb->s_root = d_alloc_root(root);
 203         } else {
 204                 CERROR("lustre_lite: bad iget4 for root\n");
 205                 GOTO(out_cdb, sb = NULL);
 206         }
 207
 208         ptlrpc_req_finished(request);
 209
 210 out_dev:
 211         if (mdc)
 212                 OBD_FREE(mdc, strlen(mdc) + 1);
 213         if (osc)
 214                 OBD_FREE(osc, strlen(osc) + 1);
 215
 216         RETURN(sb);
 217
 218 out_cdb:
 219         ll_commitcbd_cleanup(sbi);
 220 out_request:
 221         ptlrpc_req_finished(request);
 222         obd_disconnect(&sbi->ll_osc_conn);
 223 out_mdc:
 224         obd_disconnect(&sbi->ll_mdc_conn);
 225 out_free:
 226         OBD_FREE(sbi, sizeof(*sbi));
 227
 228         MOD_DEC_USE_COUNT;
 229         goto out_dev;
 230 } /* ll_read_super */
 231
 232 static void ll_put_super(struct super_block *sb)
 233 {
 234         struct ll_sb_info *sbi = ll_s2sbi(sb);
 235         ENTRY;
 236         ll_commitcbd_cleanup(sbi);
 237         obd_disconnect(&sbi->ll_osc_conn);
 238         obd_disconnect(&sbi->ll_mdc_conn);
 239         OBD_FREE(sbi, sizeof(*sbi));
 240
 241         MOD_DEC_USE_COUNT;
 242         EXIT;
 243 } /* ll_put_super */
 244
 245 static void ll_clear_inode(struct inode *inode)
 246 {
 247         if (atomic_read(&inode->i_count) == 0) {
 248                 struct ll_inode_info *lli = ll_i2info(inode);
 249                 struct lov_stripe_md *md = lli->lli_smd;
 250
 251                 if (md) {
 252                         OBD_FREE(md, md->lmd_easize);
 253                         lli->lli_smd = NULL;
 254                 }
 255                 if (lli->lli_symlink_name) {
 256                         OBD_FREE(lli->lli_symlink_name,
 257                                  strlen(lli->lli_symlink_name) + 1);
 258                         lli->lli_symlink_name = NULL;
 259                 }
 260         }
 261 }
 262
 263 static void ll_delete_inode(struct inode *inode)
 264 {
 265         if (S_ISREG(inode->i_mode)) {
 266                 int err;
 267                 struct obdo *oa;
 268                 struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
 269
 270                 if (!md)
 271                         GOTO(out, -EINVAL);
 272
 273                 if (md->lmd_object_id == 0) {
 274                         CERROR("This really happens\n");
 275                         /* No obdo was ever created */
 276                         GOTO(out, 0);
 277                 }
 278
 279                 oa = obdo_alloc();
 280                 if (oa == NULL)
 281                         GOTO(out, -ENOMEM);
 282
 283                 oa->o_id = md->lmd_object_id;
 284                 oa->o_easize = md->lmd_easize;
 285                 oa->o_mode = inode->i_mode;
 286                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLMODE;
 287
 288                 err = obd_destroy(ll_i2obdconn(inode), oa, md);
 289                 obdo_free(oa);
 290                 CDEBUG(D_SUPER, "obd destroy of %Ld error %d\n",
 291                        md->lmd_object_id, err);
 292         }
 293 out:
 294         clear_inode(inode);
 295 }
 296
 297 /* like inode_setattr, but doesn't mark the inode dirty */
 298 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 299 {
 300         unsigned int ia_valid = attr->ia_valid;
 301         int error = 0;
 302
 303         if ((ia_valid & ATTR_SIZE) && trunc) {
 304                 error = vmtruncate(inode, attr->ia_size);
 305                 if (error)
 306                         goto out;
 307         } else if (ia_valid & ATTR_SIZE)
 308                 inode->i_size = attr->ia_size;
 309
 310         if (ia_valid & ATTR_UID)
 311                 inode->i_uid = attr->ia_uid;
 312         if (ia_valid & ATTR_GID)
 313                 inode->i_gid = attr->ia_gid;
 314         if (ia_valid & ATTR_ATIME)
 315                 inode->i_atime = attr->ia_atime;
 316         if (ia_valid & ATTR_MTIME)
 317                 inode->i_mtime = attr->ia_mtime;
 318         if (ia_valid & ATTR_CTIME)
 319                 inode->i_ctime = attr->ia_ctime;
 320         if (ia_valid & ATTR_MODE) {
 321                 inode->i_mode = attr->ia_mode;
 322                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 323                         inode->i_mode &= ~S_ISGID;
 324         }
 325 out:
 326         return error;
 327 }
 328
 329 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 330 {
 331         struct ptlrpc_request *request = NULL;
 332         struct ll_sb_info *sbi = ll_i2sbi(inode);
 333         int err;
 334
 335         ENTRY;
 336
 337         /* change incore inode */
 338         ll_attr2inode(inode, attr, do_trunc);
 339
 340         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 341         if (err)
 342                 CERROR("mdc_setattr fails (%d)\n", err);
 343
 344         ptlrpc_req_finished(request);
 345
 346         RETURN(err);
 347 }
 348
 349 int ll_setattr(struct dentry *de, struct iattr *attr)
 350 {
 351         int rc = inode_change_ok(de->d_inode, attr);
 352
 353         if (rc)
 354                 return rc;
 355
 356         return ll_inode_setattr(de->d_inode, attr, 1);
 357 }
 358
 359 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 360 {
 361         struct ptlrpc_request *request = NULL;
 362         struct ll_sb_info *sbi = ll_s2sbi(sb);
 363         int rc;
 364         ENTRY;
 365
 366         memset(sfs, 0, sizeof(*sfs));
 367         rc = mdc_statfs(&sbi->ll_mdc_conn, sfs, &request);
 368         ptlrpc_req_finished(request);
 369         if (rc)
 370                 CERROR("obd_statfs fails: rc = %d\n", rc);
 371         else
 372                 CDEBUG(D_SUPER, "statfs shows blocks %ld/%ld objects %ld/%ld\n",
 373                        sfs->f_bavail, sfs->f_blocks, sfs->f_files,sfs->f_ffree);
 374
 375         /* temporary until mds_statfs returns statfs info for all OSTs */
 376         if (!rc) {
 377                 struct statfs obd_sfs;
 378
 379                 rc = obd_statfs(&sbi->ll_osc_conn, &obd_sfs);
 380                 if (rc) {
 381                         CERROR("obd_statfs fails: rc = %d\n", rc);
 382                         GOTO(out, rc);
 383                 }
 384                 CDEBUG(D_SUPER, "obd_statfs returns blocks %ld/%ld, "
 385                        "objects %ld/%ld\n",
 386                        obd_sfs.f_bavail, obd_sfs.f_blocks,
 387                        obd_sfs.f_ffree, obd_sfs.f_files);
 388
 389                 sfs->f_bfree = obd_sfs.f_bfree;
 390                 sfs->f_bavail = obd_sfs.f_bavail;
 391                 sfs->f_blocks = obd_sfs.f_blocks;
 392                 if (obd_sfs.f_ffree < sfs->f_ffree)
 393                         sfs->f_ffree = obd_sfs.f_ffree;
 394         }
 395
 396 out:
 397         RETURN(rc);
 398 }
 399
 400 inline int ll_stripe_mds_md_size(struct super_block *sb)
 401 {
 402         struct client_obd *mdc = sbi2mdc(ll_s2sbi(sb));
 403         return mdc->cl_max_mdsize;
 404 }
 405
 406 static int ll_file_size(struct inode *inode, struct lov_stripe_md *md,
 407                         __u64 *size)
 408 {
 409         struct ll_sb_info *sbi = ll_i2sbi(inode);
 410         struct lustre_handle *lockhs;
 411         struct obdo oa;
 412         int err, rc;
 413
 414         rc = ll_size_lock(inode, md, 0, LCK_PR, &lockhs);
 415         if (rc != ELDLM_OK) {
 416                 CERROR("lock enqueue: %d\n", rc);
 417                 RETURN(rc);
 418         }
 419
 420         /* FIXME: I don't like this; why doesn't osc_getattr get o_id from md
 421          * like lov_getattr? --phil */
 422         oa.o_id = md->lmd_object_id;
 423         oa.o_mode = S_IFREG;
 424         oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLSIZE;
 425         rc = obd_getattr(&sbi->ll_osc_conn, &oa, md);
 426         if (!rc)
 427                 *size = oa.o_size;
 428
 429         err = ll_size_unlock(inode, md, LCK_PR, lockhs);
 430         if (err != ELDLM_OK) {
 431                 CERROR("lock cancel: %d\n", err);
 432                 LBUG();
 433         }
 434         RETURN(rc);
 435 }
 436
 437 static void ll_read_inode2(struct inode *inode, void *opaque)
 438 {
 439         struct ll_inode_md *md = opaque;
 440         struct mds_body *body = md->body;
 441         struct ll_inode_info *ii = ll_i2info(inode);
 442         ENTRY;
 443
 444         sema_init(&ii->lli_open_sem, 1);
 445         INIT_LIST_HEAD(&ii->lli_osc_locks);
 446
 447         /* core attributes first */
 448         if (body->valid & OBD_MD_FLID)
 449                 inode->i_ino = body->ino;
 450         if (body->valid & OBD_MD_FLATIME)
 451                 inode->i_atime = body->atime;
 452         if (body->valid & OBD_MD_FLMTIME)
 453                 inode->i_mtime = body->mtime;
 454         if (body->valid & OBD_MD_FLCTIME)
 455                 inode->i_ctime = body->ctime;
 456         if (body->valid & OBD_MD_FLMODE)
 457                 inode->i_mode = body->mode;
 458         if (body->valid & OBD_MD_FLUID)
 459                 inode->i_uid = body->uid;
 460         if (body->valid & OBD_MD_FLGID)
 461                 inode->i_gid = body->gid;
 462         if (body->valid & OBD_MD_FLFLAGS)
 463                 inode->i_flags = body->flags;
 464         if (body->valid & OBD_MD_FLNLINK)
 465                 inode->i_nlink = body->nlink;
 466         if (body->valid & OBD_MD_FLGENER)
 467                 inode->i_generation = body->generation;
 468         if (body->valid & OBD_MD_FLRDEV)
 469                 inode->i_rdev = body->extra;
 470         //if (body->valid & OBD_MD_FLSIZE)
 471         //        inode->i_size = body->size;
 472
 473         /* Get the authoritative file size */
 474         if (md && md->md && inode->i_mode & S_IFREG) {
 475                 int rc;
 476                 rc = ll_file_size(inode, md->md, &inode->i_size);
 477                 if (rc) {
 478                         CERROR("ll_file_size: %d\n", rc);
 479                         /* FIXME: need to somehow prevent inode creation */
 480                         LBUG();
 481                 }
 482         }
 483
 484         //if (body->valid & OBD_MD_FLEASIZE)
 485         if (md && md->md && md->md->lmd_stripe_count) {
 486                 struct lov_mds_md *smd = md->md;
 487                 int size;
 488                 if (md->md->lmd_easize != ll_stripe_mds_md_size(inode->i_sb)) {
 489                         CERROR("Striping metadata size error %ld\n",
 490                                inode->i_ino);
 491                         LBUG();
 492                 }
 493                 size = sizeof(*ii->lli_smd) +
 494                         md->md->lmd_stripe_count * sizeof(struct lov_oinfo);
 495                 OBD_ALLOC(ii->lli_smd, size);
 496                 if (!ii->lli_smd){
 497                         CERROR("No memory for %d\n", size);
 498                         LBUG();
 499                 }
 500                 lov_unpackmd(ii->lli_smd, smd);
 501         }
 502
 503         /* OIDEBUG(inode); */
 504
 505         if (S_ISREG(inode->i_mode)) {
 506                 inode->i_op = &ll_file_inode_operations;
 507                 inode->i_fop = &ll_file_operations;
 508                 inode->i_mapping->a_ops = &ll_aops;
 509                 EXIT;
 510         } else if (S_ISDIR(inode->i_mode)) {
 511                 inode->i_op = &ll_dir_inode_operations;
 512                 inode->i_fop = &ll_dir_operations;
 513                 inode->i_mapping->a_ops = &ll_dir_aops;
 514                 EXIT;
 515         } else if (S_ISLNK(inode->i_mode)) {
 516                 inode->i_op = &ll_fast_symlink_inode_operations;
 517                 EXIT;
 518         } else {
 519                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 520                 EXIT;
 521         }
 522
 523         return;
 524 }
 525
 526 /* exported operations */
 527 struct super_operations ll_super_operations =
 528 {
 529         read_inode2: ll_read_inode2,
 530         clear_inode: ll_clear_inode,
 531         delete_inode: ll_delete_inode,
 532         put_super: ll_put_super,
 533         statfs: ll_statfs
 534 };
 535
 536 struct file_system_type lustre_lite_fs_type = {
 537         "lustre_lite", 0, ll_read_super, NULL
 538 };
 539
 540 static int __init init_lustre_lite(void)
 541 {
 542         printk(KERN_INFO "Lustre Lite 0.0.1, info@clusterfs.com\n");
 543         ll_file_data_slab = kmem_cache_create("ll_file_data",
 544                                               sizeof(struct ll_file_data), 0,
 545                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 546         if (ll_file_data_slab == NULL)
 547                 return -ENOMEM;
 548         return register_filesystem(&lustre_lite_fs_type);
 549 }
 550
 551 static void __exit exit_lustre_lite(void)
 552 {
 553         unregister_filesystem(&lustre_lite_fs_type);
 554         kmem_cache_destroy(ll_file_data_slab);
 555 }
 556
 557 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 558 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 559 MODULE_LICENSE("GPL");
 560
 561 module_init(init_lustre_lite);
 562 module_exit(exit_lustre_lite);