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         sbi->ll_mount_epoch = 0;
 125         generate_random_uuid(uuid);
 126         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 127
 128         sb->u.generic_sbp = sbi;
 129
 130         ll_options(data, &osc, &mdc, &sbi->ll_flags);
 131
 132         if (!osc) {
 133                 CERROR("no osc\n");
 134                 GOTO(out_free, sb = NULL);
 135         }
 136
 137         if (!mdc) {
 138                 CERROR("no mdc\n");
 139                 GOTO(out_free, sb = NULL);
 140         }
 141
 142         obd = class_uuid2obd(mdc);
 143         if (!obd) {
 144                 CERROR("MDC %s: not setup or attached\n", mdc);
 145                 GOTO(out_free, sb = NULL);
 146         }
 147
 148         err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid,
 149                           ptlrpc_recovd, ll_recover);
 150         if (err) {
 151                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 152                 GOTO(out_free, sb = NULL);
 153         }
 154
 155 #warning Peter: is this the right place to raise the connection level?
 156         mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
 157         mdc_conn->c_level = LUSTRE_CONN_FULL;
 158         list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
 159
 160         obd = class_uuid2obd(osc);
 161         if (!obd) {
 162                 CERROR("OSC %s: not setup or attached\n", osc);
 163                 GOTO(out_mdc, sb = NULL);
 164         }
 165
 166         err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid,
 167                           ptlrpc_recovd, ll_recover);
 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         lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
 216         lic.lic_lmm = NULL;
 217         LASSERT(sbi->ll_rootino != 0);
 218         root = iget4(sb, sbi->ll_rootino, NULL, &lic);
 219
 220         if (root) {
 221                 sb->s_root = d_alloc_root(root);
 222         } else {
 223                 CERROR("lustre_lite: bad iget4 for root\n");
 224                 GOTO(out_cdb, sb = NULL);
 225         }
 226
 227         ptlrpc_req_finished(request);
 228
 229 out_dev:
 230         if (mdc)
 231                 OBD_FREE(mdc, strlen(mdc) + 1);
 232         if (osc)
 233                 OBD_FREE(osc, strlen(osc) + 1);
 234
 235         RETURN(sb);
 236
 237 out_cdb:
 238         ll_commitcbd_cleanup(sbi);
 239 out_request:
 240         ptlrpc_req_finished(request);
 241         obd_disconnect(&sbi->ll_osc_conn);
 242 out_mdc:
 243         obd_disconnect(&sbi->ll_mdc_conn);
 244 out_free:
 245         OBD_FREE(sbi, sizeof(*sbi));
 246
 247         MOD_DEC_USE_COUNT;
 248         goto out_dev;
 249 } /* ll_read_super */
 250
 251 static void ll_put_super(struct super_block *sb)
 252 {
 253         struct ll_sb_info *sbi = ll_s2sbi(sb);
 254
 255         ENTRY;
 256         list_del(&sbi->ll_conn_chain);
 257         ll_commitcbd_cleanup(sbi);
 258         obd_disconnect(&sbi->ll_osc_conn);
 259         obd_disconnect(&sbi->ll_mdc_conn);
 260         OBD_FREE(sbi, sizeof(*sbi));
 261
 262         MOD_DEC_USE_COUNT;
 263         EXIT;
 264 } /* ll_put_super */
 265
 266 static void ll_clear_inode(struct inode *inode)
 267 {
 268         ENTRY;
 269
 270         if (atomic_read(&inode->i_count) == 0) {
 271                 struct ll_inode_info *lli = ll_i2info(inode);
 272                 struct lov_stripe_md *lsm = lli->lli_smd;
 273                 char *symlink_name = lli->lli_symlink_name;
 274
 275                 if (lsm) {
 276                         OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
 277                         lli->lli_smd = NULL;
 278                 }
 279                 if (symlink_name) {
 280                         OBD_FREE(symlink_name, strlen(symlink_name) + 1);
 281                         lli->lli_symlink_name = NULL;
 282                 }
 283         }
 284         EXIT;
 285 }
 286
 287 static void ll_delete_inode(struct inode *inode)
 288 {
 289         ENTRY;
 290         if (S_ISREG(inode->i_mode)) {
 291                 int err;
 292                 struct obdo *oa;
 293                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 294
 295                 if (!lsm)
 296                         GOTO(out, -EINVAL);
 297
 298                 if (lsm->lsm_object_id == 0) {
 299                         CERROR("This really happens\n");
 300                         /* No obdo was ever created */
 301                         GOTO(out, 0);
 302                 }
 303
 304                 oa = obdo_alloc();
 305                 if (oa == NULL)
 306                         GOTO(out, -ENOMEM);
 307
 308                 oa->o_id = lsm->lsm_object_id;
 309                 oa->o_easize = ll_mds_easize(inode->i_sb);
 310                 oa->o_mode = inode->i_mode;
 311                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
 312
 313                 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
 314                 obdo_free(oa);
 315                 CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
 316                        lsm->lsm_object_id, err);
 317         }
 318 out:
 319         clear_inode(inode);
 320         EXIT;
 321 }
 322
 323 /* like inode_setattr, but doesn't mark the inode dirty */
 324 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 325 {
 326         unsigned int ia_valid = attr->ia_valid;
 327         int error = 0;
 328
 329         if ((ia_valid & ATTR_SIZE) && trunc) {
 330                 error = vmtruncate(inode, attr->ia_size);
 331                 if (error)
 332                         goto out;
 333         } else if (ia_valid & ATTR_SIZE)
 334                 inode->i_size = attr->ia_size;
 335
 336         if (ia_valid & ATTR_UID)
 337                 inode->i_uid = attr->ia_uid;
 338         if (ia_valid & ATTR_GID)
 339                 inode->i_gid = attr->ia_gid;
 340         if (ia_valid & ATTR_ATIME)
 341                 inode->i_atime = attr->ia_atime;
 342         if (ia_valid & ATTR_MTIME)
 343                 inode->i_mtime = attr->ia_mtime;
 344         if (ia_valid & ATTR_CTIME)
 345                 inode->i_ctime = attr->ia_ctime;
 346         if (ia_valid & ATTR_MODE) {
 347                 inode->i_mode = attr->ia_mode;
 348                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 349                         inode->i_mode &= ~S_ISGID;
 350         }
 351 out:
 352         return error;
 353 }
 354
 355 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 356 {
 357         struct ptlrpc_request *request = NULL;
 358         struct ll_sb_info *sbi = ll_i2sbi(inode);
 359         int err;
 360
 361         ENTRY;
 362
 363         /* change incore inode */
 364         ll_attr2inode(inode, attr, do_trunc);
 365
 366         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 367         if (err)
 368                 CERROR("mdc_setattr fails (%d)\n", err);
 369
 370         ptlrpc_req_finished(request);
 371
 372         RETURN(err);
 373 }
 374
 375 int ll_setattr(struct dentry *de, struct iattr *attr)
 376 {
 377         int rc = inode_change_ok(de->d_inode, attr);
 378
 379         if (rc)
 380                 return rc;
 381
 382         return ll_inode_setattr(de->d_inode, attr, 1);
 383 }
 384
 385 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 386 {
 387         struct ptlrpc_request *request = NULL;
 388         struct ll_sb_info *sbi = ll_s2sbi(sb);
 389         struct obd_statfs osfs;
 390         int rc;
 391         ENTRY;
 392
 393         memset(sfs, 0, sizeof(*sfs));
 394         rc = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 395         statfs_unpack(sfs, &osfs);
 396         ptlrpc_req_finished(request);
 397         if (rc)
 398                 CERROR("mdc_statfs fails: rc = %d\n", rc);
 399         else
 400                 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
 401                        " objects "LPU64"/"LPU64"\n",
 402                        osfs.os_bavail, osfs.os_blocks,
 403                        osfs.os_ffree, osfs.os_files);
 404
 405         /* temporary until mds_statfs returns statfs info for all OSTs */
 406         if (!rc) {
 407                 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
 408                 if (rc) {
 409                         CERROR("obd_statfs fails: rc = %d\n", rc);
 410                         GOTO(out, rc);
 411                 }
 412                 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
 413                        " objects "LPU64"/"LPU64"\n",
 414                        osfs.os_bavail, osfs.os_blocks,
 415                        osfs.os_ffree, osfs.os_files);
 416
 417                 while (osfs.os_blocks > ~0UL) {
 418                         sfs->f_bsize <<= 1;
 419
 420                         osfs.os_blocks >>= 1;
 421                         osfs.os_bfree >>= 1;
 422                         osfs.os_bavail >>= 1;
 423                 }
 424                 sfs->f_blocks = osfs.os_blocks;
 425                 sfs->f_bfree = osfs.os_bfree;
 426                 sfs->f_bavail = osfs.os_bavail;
 427                 if (osfs.os_ffree < (__u64)sfs->f_ffree)
 428                         sfs->f_ffree = osfs.os_ffree;
 429         }
 430
 431 out:
 432         RETURN(rc);
 433 }
 434
 435 void ll_update_inode(struct inode *inode, struct mds_body *body)
 436 {
 437         if (body->valid & OBD_MD_FLID)
 438                 inode->i_ino = body->ino;
 439         if (body->valid & OBD_MD_FLATIME)
 440                 inode->i_atime = body->atime;
 441         if (body->valid & OBD_MD_FLMTIME)
 442                 inode->i_mtime = body->mtime;
 443         if (body->valid & OBD_MD_FLCTIME)
 444                 inode->i_ctime = body->ctime;
 445         if (body->valid & OBD_MD_FLMODE)
 446                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
 447         if (body->valid & OBD_MD_FLTYPE)
 448                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
 449         if (body->valid & OBD_MD_FLUID)
 450                 inode->i_uid = body->uid;
 451         if (body->valid & OBD_MD_FLGID)
 452                 inode->i_gid = body->gid;
 453         if (body->valid & OBD_MD_FLFLAGS)
 454                 inode->i_flags = body->flags;
 455         if (body->valid & OBD_MD_FLNLINK)
 456                 inode->i_nlink = body->nlink;
 457         if (body->valid & OBD_MD_FLGENER)
 458                 inode->i_generation = body->generation;
 459         if (body->valid & OBD_MD_FLRDEV)
 460                 inode->i_rdev = body->rdev;
 461         if (body->valid & OBD_MD_FLSIZE)
 462                 inode->i_size = body->size;
 463 }
 464
 465 static void ll_read_inode2(struct inode *inode, void *opaque)
 466 {
 467         struct ll_read_inode2_cookie *lic = opaque;
 468         struct mds_body *body = lic->lic_body;
 469         struct ll_inode_info *lli = ll_i2info(inode);
 470         ENTRY;
 471
 472         sema_init(&lli->lli_open_sem, 1);
 473         lli->lli_mount_epoch = ll_i2sbi(inode)->ll_mount_epoch;
 474
 475         /* core attributes first */
 476         ll_update_inode(inode, body);
 477
 478         //if (body->valid & OBD_MD_FLEASIZE)
 479         if (lic && lic->lic_lmm) {
 480                 struct lov_mds_md *lmm = lic->lic_lmm;
 481                 int size;
 482
 483                 if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
 484                         CERROR("Striping metadata size error %ld\n",
 485                                inode->i_ino);
 486                         LBUG();
 487                 }
 488                 size = ll_ost_easize(inode->i_sb);
 489                 OBD_ALLOC(lli->lli_smd, size);
 490                 if (!lli->lli_smd) {
 491                         CERROR("No memory for %d\n", size);
 492                         LBUG();
 493                 }
 494                 lov_unpackmd(lli->lli_smd, lmm);
 495         } else {
 496                 lli->lli_smd = NULL;
 497         }
 498
 499         /* Get the authoritative file size */
 500         if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
 501                 int rc;
 502
 503                 rc = ll_file_size(inode, lli->lli_smd);
 504                 if (rc) {
 505                         CERROR("ll_file_size: %d\n", rc);
 506                         /* FIXME: need to somehow prevent inode creation */
 507                         LBUG();
 508                 }
 509         }
 510
 511         /* OIDEBUG(inode); */
 512
 513         if (S_ISREG(inode->i_mode)) {
 514                 inode->i_op = &ll_file_inode_operations;
 515                 inode->i_fop = &ll_file_operations;
 516                 inode->i_mapping->a_ops = &ll_aops;
 517                 EXIT;
 518         } else if (S_ISDIR(inode->i_mode)) {
 519                 inode->i_op = &ll_dir_inode_operations;
 520                 inode->i_fop = &ll_dir_operations;
 521                 inode->i_mapping->a_ops = &ll_dir_aops;
 522                 EXIT;
 523         } else if (S_ISLNK(inode->i_mode)) {
 524                 inode->i_op = &ll_fast_symlink_inode_operations;
 525                 EXIT;
 526         } else {
 527                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 528                 EXIT;
 529         }
 530 }
 531
 532 static inline void invalidate_request_list(struct list_head *req_list)
 533 {
 534         struct list_head *tmp, *n;
 535         list_for_each_safe(tmp, n, req_list) {
 536                 struct ptlrpc_request *req =
 537                         list_entry(tmp, struct ptlrpc_request, rq_list);
 538                 CERROR("invalidating req xid "LPD64" op %d to %s:%d\n",
 539                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 540                        req->rq_connection->c_remote_uuid,
 541                        req->rq_import->imp_client->cli_request_portal);
 542                 req->rq_flags |= PTL_RPC_FL_ERR;
 543                 wake_up(&req->rq_wait_for_rep);
 544         }
 545 }
 546
 547 void ll_umount_begin(struct super_block *sb)
 548 {
 549         struct ll_sb_info *sbi = ll_s2sbi(sb);
 550         struct list_head *ctmp;
 551
 552         ENTRY;
 553
 554         list_for_each(ctmp, &sbi->ll_conn_chain) {
 555                 struct ptlrpc_connection *conn;
 556                 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
 557
 558                 spin_lock(&conn->c_lock);
 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