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/version.h>
  17 #include <linux/lustre_lite.h>
  18 #include <linux/lustre_ha.h>
  19 #include <linux/obd_lov.h>
  20 #include <linux/lustre_dlm.h>
  21 #include <linux/init.h>
  22 #include <linux/fs.h>
  23
  24 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  25 kmem_cache_t *ll_file_data_slab;
  26 extern struct address_space_operations ll_aops;
  27 extern struct address_space_operations ll_dir_aops;
  28 struct super_operations ll_super_operations;
  29
  30 extern int ll_recover(struct recovd_data *, int);
  31 extern int ll_commitcbd_setup(struct ll_sb_info *);
  32 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  33
  34 static char *ll_read_opt(const char *opt, char *data)
  35 {
  36         char *value;
  37         char *retval;
  38         ENTRY;
  39
  40         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  41         if ( strncmp(opt, data, strlen(opt)) )
  42                 RETURN(NULL);
  43         if ( (value = strchr(data, '=')) == NULL )
  44                 RETURN(NULL);
  45
  46         value++;
  47         OBD_ALLOC(retval, strlen(value) + 1);
  48         if ( !retval ) {
  49                 CERROR("out of memory!\n");
  50                 RETURN(NULL);
  51         }
  52
  53         memcpy(retval, value, strlen(value)+1);
  54         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  55         RETURN(retval);
  56 }
  57
  58 static int ll_set_opt(const char *opt, char *data, int fl)
  59 {
  60         ENTRY;
  61
  62         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  63         if ( strncmp(opt, data, strlen(opt)) )
  64                 RETURN(0);
  65         else
  66                 RETURN(fl);
  67 }
  68
  69 static void ll_options(char *options, char **ost, char **mds, int *flags)
  70 {
  71         char *this_char;
  72         ENTRY;
  73
  74         if (!options) {
  75                 EXIT;
  76                 return;
  77         }
  78
  79         for (this_char = strtok (options, ",");
  80              this_char != NULL;
  81              this_char = strtok (NULL, ",")) {
  82                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
  83                 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
  84                      (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
  85                      (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
  86                       ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
  87                         continue;
  88         }
  89         EXIT;
  90 }
  91
  92 #ifndef log2
  93 #define log2(n) ffz(~(n))
  94 #endif
  95
  96 static struct super_block * ll_read_super(struct super_block *sb,
  97                                           void *data, int silent)
  98 {
  99         struct inode *root = 0;
 100         struct obd_device *obd;
 101         struct ll_sb_info *sbi;
 102         char *osc = NULL;
 103         char *mdc = NULL;
 104         int err;
 105         struct ll_fid rootfid;
 106         struct obd_statfs osfs;
 107         __u64 last_committed;
 108         __u64 last_xid;
 109         struct ptlrpc_request *request = NULL;
 110         struct ptlrpc_connection *mdc_conn;
 111         struct ll_read_inode2_cookie lic;
 112         class_uuid_t uuid;
 113
 114         ENTRY;
 115         MOD_INC_USE_COUNT;
 116
 117         OBD_ALLOC(sbi, sizeof(*sbi));
 118         if (!sbi) {
 119                 MOD_DEC_USE_COUNT;
 120                 RETURN(NULL);
 121         }
 122
 123         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 124         generate_random_uuid(uuid);
 125         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 126
 127         sb->u.generic_sbp = sbi;
 128
 129         ll_options(data, &osc, &mdc, &sbi->ll_flags);
 130
 131         if (!osc) {
 132                 CERROR("no osc\n");
 133                 GOTO(out_free, sb = NULL);
 134         }
 135
 136         if (!mdc) {
 137                 CERROR("no mdc\n");
 138                 GOTO(out_free, sb = NULL);
 139         }
 140
 141         obd = class_uuid2obd(mdc);
 142         if (!obd) {
 143                 CERROR("MDC %s: not setup or attached\n", mdc);
 144                 GOTO(out_free, sb = NULL);
 145         }
 146
 147         err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
 148                           ptlrpc_recovd, ll_recover);
 149         if (err) {
 150                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 151                 GOTO(out_free, sb = NULL);
 152         }
 153
 154 #warning Peter: is this the right place to raise the connection level?
 155         mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
 156         mdc_conn->c_level = LUSTRE_CONN_FULL;
 157         list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
 158
 159         obd = class_uuid2obd(osc);
 160         if (!obd) {
 161                 CERROR("OSC %s: not setup or attached\n", osc);
 162                 GOTO(out_mdc, sb = NULL);
 163         }
 164
 165         err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
 166                           ptlrpc_recovd, ll_recover);
 167         if (err) {
 168                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 169                 GOTO(out_mdc, sb = NULL);
 170         }
 171
 172         /* XXX: need to store the last_* values somewhere */
 173         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
 174                             &last_xid, &request);
 175         ptlrpc_req_finished(request);
 176         if (err) {
 177                 CERROR("cannot mds_connect: rc = %d\n", err);
 178                 GOTO(out_mdc, sb = NULL);
 179         }
 180         CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
 181         sbi->ll_rootino = rootfid.id;
 182
 183         memset(&osfs, 0, sizeof(osfs));
 184         request = NULL;
 185         err = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 186         ptlrpc_req_finished(request);
 187         sb->s_blocksize = osfs.os_bsize;
 188         sb->s_blocksize_bits = log2(osfs.os_bsize);
 189         sb->s_magic = LL_SUPER_MAGIC;
 190         sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
 191
 192         sb->s_op = &ll_super_operations;
 193
 194         /* make root inode */
 195         request = NULL;
 196         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
 197                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 198         if (err) {
 199                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 200                 GOTO(out_request, sb = NULL);
 201         }
 202
 203         /* initialize committed transaction callback daemon */
 204         spin_lock_init(&sbi->ll_commitcbd_lock);
 205         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 206         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 207         sbi->ll_commitcbd_flags = 0;
 208         err = ll_commitcbd_setup(sbi);
 209         if (err) {
 210                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 211                 GOTO(out_request, sb = NULL);
 212         }
 213
 214         lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
 215         lic.lic_lmm = NULL;
 216         LASSERT(sbi->ll_rootino != 0);
 217         root = iget4(sb, sbi->ll_rootino, NULL, &lic);
 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
 254         ENTRY;
 255         list_del(&sbi->ll_conn_chain);
 256         ll_commitcbd_cleanup(sbi);
 257         obd_disconnect(&sbi->ll_osc_conn);
 258         obd_disconnect(&sbi->ll_mdc_conn);
 259         OBD_FREE(sbi, sizeof(*sbi));
 260
 261         MOD_DEC_USE_COUNT;
 262         EXIT;
 263 } /* ll_put_super */
 264
 265 static void ll_clear_inode(struct inode *inode)
 266 {
 267         ENTRY;
 268
 269         if (atomic_read(&inode->i_count) == 0) {
 270                 struct ll_inode_info *lli = ll_i2info(inode);
 271                 struct lov_stripe_md *lsm = lli->lli_smd;
 272                 char *symlink_name = lli->lli_symlink_name;
 273
 274                 if (lsm) {
 275                         OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
 276                         lli->lli_smd = NULL;
 277                 }
 278                 if (symlink_name) {
 279                         OBD_FREE(symlink_name, strlen(symlink_name) + 1);
 280                         lli->lli_symlink_name = NULL;
 281                 }
 282         }
 283         EXIT;
 284 }
 285
 286 static void ll_delete_inode(struct inode *inode)
 287 {
 288         ENTRY;
 289         if (S_ISREG(inode->i_mode)) {
 290                 int err;
 291                 struct obdo *oa;
 292                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 293
 294                 if (!lsm)
 295                         GOTO(out, -EINVAL);
 296
 297                 if (lsm->lsm_object_id == 0) {
 298                         CERROR("This really happens\n");
 299                         /* No obdo was ever created */
 300                         GOTO(out, 0);
 301                 }
 302
 303                 oa = obdo_alloc();
 304                 if (oa == NULL)
 305                         GOTO(out, -ENOMEM);
 306
 307                 oa->o_id = lsm->lsm_object_id;
 308                 oa->o_easize = ll_mds_easize(inode->i_sb);
 309                 oa->o_mode = inode->i_mode;
 310                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
 311
 312                 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
 313                 obdo_free(oa);
 314                 CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
 315                        lsm->lsm_object_id, err);
 316         }
 317 out:
 318         clear_inode(inode);
 319         EXIT;
 320 }
 321
 322 /* like inode_setattr, but doesn't mark the inode dirty */
 323 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 324 {
 325         unsigned int ia_valid = attr->ia_valid;
 326         int error = 0;
 327
 328         if ((ia_valid & ATTR_SIZE) && trunc) {
 329                 error = vmtruncate(inode, attr->ia_size);
 330                 if (error)
 331                         goto out;
 332         } else if (ia_valid & ATTR_SIZE)
 333                 inode->i_size = attr->ia_size;
 334
 335         if (ia_valid & ATTR_UID)
 336                 inode->i_uid = attr->ia_uid;
 337         if (ia_valid & ATTR_GID)
 338                 inode->i_gid = attr->ia_gid;
 339         if (ia_valid & ATTR_ATIME)
 340                 inode->i_atime = attr->ia_atime;
 341         if (ia_valid & ATTR_MTIME)
 342                 inode->i_mtime = attr->ia_mtime;
 343         if (ia_valid & ATTR_CTIME)
 344                 inode->i_ctime = attr->ia_ctime;
 345         if (ia_valid & ATTR_MODE) {
 346                 inode->i_mode = attr->ia_mode;
 347                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 348                         inode->i_mode &= ~S_ISGID;
 349         }
 350 out:
 351         return error;
 352 }
 353
 354 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 355 {
 356         struct ptlrpc_request *request = NULL;
 357         struct ll_sb_info *sbi = ll_i2sbi(inode);
 358         int err;
 359
 360         ENTRY;
 361
 362         /* change incore inode */
 363         ll_attr2inode(inode, attr, do_trunc);
 364
 365         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 366         if (err)
 367                 CERROR("mdc_setattr fails (%d)\n", err);
 368
 369         ptlrpc_req_finished(request);
 370
 371         RETURN(err);
 372 }
 373
 374 int ll_setattr(struct dentry *de, struct iattr *attr)
 375 {
 376         int rc = inode_change_ok(de->d_inode, attr);
 377
 378         if (rc)
 379                 return rc;
 380
 381         return ll_inode_setattr(de->d_inode, attr, 1);
 382 }
 383
 384 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 385 {
 386         struct ptlrpc_request *request = NULL;
 387         struct ll_sb_info *sbi = ll_s2sbi(sb);
 388         struct obd_statfs osfs;
 389         int rc;
 390         ENTRY;
 391
 392         memset(sfs, 0, sizeof(*sfs));
 393         rc = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 394         statfs_unpack(sfs, &osfs);
 395         ptlrpc_req_finished(request);
 396         if (rc)
 397                 CERROR("mdc_statfs fails: rc = %d\n", rc);
 398         else
 399                 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
 400                        " objects "LPU64"/"LPU64"\n",
 401                        osfs.os_bavail, osfs.os_blocks,
 402                        osfs.os_ffree, osfs.os_files);
 403
 404         /* temporary until mds_statfs returns statfs info for all OSTs */
 405         if (!rc) {
 406                 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
 407                 if (rc) {
 408                         CERROR("obd_statfs fails: rc = %d\n", rc);
 409                         GOTO(out, rc);
 410                 }
 411                 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
 412                        " objects "LPU64"/"LPU64"\n",
 413                        osfs.os_bavail, osfs.os_blocks,
 414                        osfs.os_ffree, osfs.os_files);
 415
 416                 while (osfs.os_blocks > ~0UL) {
 417                         sfs->f_bsize <<= 1;
 418
 419                         osfs.os_blocks >>= 1;
 420                         osfs.os_bfree >>= 1;
 421                         osfs.os_bavail >>= 1;
 422                 }
 423                 sfs->f_blocks = osfs.os_blocks;
 424                 sfs->f_bfree = osfs.os_bfree;
 425                 sfs->f_bavail = osfs.os_bavail;
 426                 if (osfs.os_ffree < (__u64)sfs->f_ffree)
 427                         sfs->f_ffree = osfs.os_ffree;
 428         }
 429
 430 out:
 431         RETURN(rc);
 432 }
 433
 434 void ll_update_inode(struct inode *inode, struct mds_body *body)
 435 {
 436         if (body->valid & OBD_MD_FLID)
 437                 inode->i_ino = body->ino;
 438         if (body->valid & OBD_MD_FLATIME)
 439                 inode->i_atime = body->atime;
 440         if (body->valid & OBD_MD_FLMTIME)
 441                 inode->i_mtime = body->mtime;
 442         if (body->valid & OBD_MD_FLCTIME)
 443                 inode->i_ctime = body->ctime;
 444         if (body->valid & OBD_MD_FLMODE)
 445                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
 446         if (body->valid & OBD_MD_FLTYPE)
 447                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
 448         if (body->valid & OBD_MD_FLUID)
 449                 inode->i_uid = body->uid;
 450         if (body->valid & OBD_MD_FLGID)
 451                 inode->i_gid = body->gid;
 452         if (body->valid & OBD_MD_FLFLAGS)
 453                 inode->i_flags = body->flags;
 454         if (body->valid & OBD_MD_FLNLINK)
 455                 inode->i_nlink = body->nlink;
 456         if (body->valid & OBD_MD_FLGENER)
 457                 inode->i_generation = body->generation;
 458         if (body->valid & OBD_MD_FLRDEV)
 459                 inode->i_rdev = body->rdev;
 460         if (body->valid & OBD_MD_FLSIZE)
 461                 inode->i_size = body->size;
 462 }
 463
 464 static void ll_read_inode2(struct inode *inode, void *opaque)
 465 {
 466         struct ll_read_inode2_cookie *lic = opaque;
 467         struct mds_body *body = lic->lic_body;
 468         struct ll_inode_info *lli = ll_i2info(inode);
 469         ENTRY;
 470
 471         sema_init(&lli->lli_open_sem, 1);
 472
 473         /* core attributes first */
 474         ll_update_inode(inode, body);
 475
 476         //if (body->valid & OBD_MD_FLEASIZE)
 477         if (lic && lic->lic_lmm) {
 478                 struct lov_mds_md *lmm = lic->lic_lmm;
 479                 int size;
 480
 481                 if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
 482                         CERROR("Striping metadata size error %ld\n",
 483                                inode->i_ino);
 484                         LBUG();
 485                 }
 486                 size = ll_ost_easize(inode->i_sb);
 487                 OBD_ALLOC(lli->lli_smd, size);
 488                 if (!lli->lli_smd) {
 489                         CERROR("No memory for %d\n", size);
 490                         LBUG();
 491                 }
 492                 lov_unpackmd(lli->lli_smd, lmm);
 493         } else {
 494                 lli->lli_smd = NULL;
 495         }
 496
 497         /* Get the authoritative file size */
 498         if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
 499                 int rc;
 500
 501                 rc = ll_file_size(inode, lli->lli_smd);
 502                 if (rc) {
 503                         CERROR("ll_file_size: %d\n", rc);
 504                         /* FIXME: need to somehow prevent inode creation */
 505                         LBUG();
 506                 }
 507         }
 508
 509         /* OIDEBUG(inode); */
 510
 511         if (S_ISREG(inode->i_mode)) {
 512                 inode->i_op = &ll_file_inode_operations;
 513                 inode->i_fop = &ll_file_operations;
 514                 inode->i_mapping->a_ops = &ll_aops;
 515                 EXIT;
 516         } else if (S_ISDIR(inode->i_mode)) {
 517                 inode->i_op = &ll_dir_inode_operations;
 518                 inode->i_fop = &ll_dir_operations;
 519                 inode->i_mapping->a_ops = &ll_dir_aops;
 520                 EXIT;
 521         } else if (S_ISLNK(inode->i_mode)) {
 522                 inode->i_op = &ll_fast_symlink_inode_operations;
 523                 EXIT;
 524         } else {
 525                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 526                 EXIT;
 527         }
 528 }
 529
 530 static inline void invalidate_request_list(struct list_head *req_list)
 531 {
 532         struct list_head *tmp, *n;
 533         list_for_each_safe(tmp, n, req_list) {
 534                 struct ptlrpc_request *req =
 535                         list_entry(tmp, struct ptlrpc_request, rq_list);
 536                 CERROR("invalidating req xid "LPD64" op %d to %s:%d\n",
 537                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 538                        req->rq_connection->c_remote_uuid,
 539                        req->rq_import->imp_client->cli_request_portal);
 540                 req->rq_flags |= PTL_RPC_FL_ERR;
 541                 wake_up(&req->rq_wait_for_rep);
 542         }
 543 }
 544
 545 void ll_umount_begin(struct super_block *sb)
 546 {
 547         struct ll_sb_info *sbi = ll_s2sbi(sb);
 548         struct list_head *ctmp;
 549
 550         ENTRY;
 551
 552         list_for_each(ctmp, &sbi->ll_conn_chain) {
 553                 struct ptlrpc_connection *conn;
 554                 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
 555
 556                 spin_lock(&conn->c_lock);
 557                 /* XXX should just be dealing with imports, probably through
 558                  * XXX iocontrol, need next-gen recovery! */
 559                 conn->c_flags |= CONN_INVALID;
 560                 invalidate_request_list(&conn->c_sending_head);
 561                 invalidate_request_list(&conn->c_delayed_head);
 562                 spin_unlock(&conn->c_lock);
 563         }
 564
 565         EXIT;
 566 }
 567
 568 /* exported operations */
 569 struct super_operations ll_super_operations =
 570 {
 571         read_inode2: ll_read_inode2,
 572         clear_inode: ll_clear_inode,
 573         delete_inode: ll_delete_inode,
 574         put_super: ll_put_super,
 575         statfs: ll_statfs,
 576         umount_begin: ll_umount_begin
 577 };
 578
 579 struct file_system_type lustre_lite_fs_type = {
 580         "lustre_lite", 0, ll_read_super, NULL
 581 };
 582
 583 static int __init init_lustre_lite(void)
 584 {
 585         printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
 586         ll_file_data_slab = kmem_cache_create("ll_file_data",
 587                                               sizeof(struct ll_file_data), 0,
 588                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
 589         if (ll_file_data_slab == NULL)
 590                 return -ENOMEM;
 591         return register_filesystem(&lustre_lite_fs_type);
 592 }
 593
 594 static void __exit exit_lustre_lite(void)
 595 {
 596         unregister_filesystem(&lustre_lite_fs_type);
 597         kmem_cache_destroy(ll_file_data_slab);
 598 }
 599
 600 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 601 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 602 MODULE_LICENSE("GPL");
 603
 604 module_init(init_lustre_lite);
 605 module_exit(exit_lustre_lite);
 606 #endif