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 #include <linux/fs.h>
  22
  23 kmem_cache_t *ll_file_data_slab;
  24 extern struct address_space_operations ll_aops;
  25 extern struct address_space_operations ll_dir_aops;
  26 struct super_operations ll_super_operations;
  27
  28 extern int ll_recover(struct recovd_data *, int);
  29 extern int ll_commitcbd_setup(struct ll_sb_info *);
  30 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  31
  32 static char *ll_read_opt(const char *opt, char *data)
  33 {
  34         char *value;
  35         char *retval;
  36         ENTRY;
  37
  38         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  39         if ( strncmp(opt, data, strlen(opt)) )
  40                 RETURN(NULL);
  41         if ( (value = strchr(data, '=')) == NULL )
  42                 RETURN(NULL);
  43
  44         value++;
  45         OBD_ALLOC(retval, strlen(value) + 1);
  46         if ( !retval ) {
  47                 CERROR("out of memory!\n");
  48                 RETURN(NULL);
  49         }
  50
  51         memcpy(retval, value, strlen(value)+1);
  52         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  53         RETURN(retval);
  54 }
  55
  56 static int ll_set_opt(const char *opt, char *data, int fl)
  57 {
  58         ENTRY;
  59
  60         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  61         if ( strncmp(opt, data, strlen(opt)) )
  62                 RETURN(0);
  63         else
  64                 RETURN(fl);
  65 }
  66
  67 static void ll_options(char *options, char **ost, char **mds, int *flags)
  68 {
  69         char *this_char;
  70         ENTRY;
  71
  72         if (!options) {
  73                 EXIT;
  74                 return;
  75         }
  76
  77         for (this_char = strtok (options, ",");
  78              this_char != NULL;
  79              this_char = strtok (NULL, ",")) {
  80                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
  81                 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
  82                      (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
  83                      (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
  84                       ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
  85                         continue;
  86         }
  87         EXIT;
  88 }
  89
  90 #ifndef log2
  91 #define log2(n) ffz(~(n))
  92 #endif
  93
  94 static struct super_block * ll_read_super(struct super_block *sb,
  95                                           void *data, int silent)
  96 {
  97         struct inode *root = 0;
  98         struct obd_device *obd;
  99         struct ll_sb_info *sbi;
 100         char *osc = NULL;
 101         char *mdc = NULL;
 102         int err;
 103         struct ll_fid rootfid;
 104         struct obd_statfs osfs;
 105         __u64 last_committed;
 106         __u64 last_xid;
 107         struct ptlrpc_request *request = NULL;
 108         struct ptlrpc_connection *mdc_conn;
 109         struct ll_read_inode2_cookie lic;
 110         class_uuid_t uuid;
 111
 112         ENTRY;
 113         MOD_INC_USE_COUNT;
 114
 115         OBD_ALLOC(sbi, sizeof(*sbi));
 116         if (!sbi) {
 117                 MOD_DEC_USE_COUNT;
 118                 RETURN(NULL);
 119         }
 120
 121         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 122         sbi->ll_mount_epoch = 0;
 123         generate_random_uuid(uuid);
 124         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 125
 126         sb->u.generic_sbp = sbi;
 127
 128         ll_options(data, &osc, &mdc, &sbi->ll_flags);
 129
 130         if (!osc) {
 131                 CERROR("no osc\n");
 132                 GOTO(out_free, sb = NULL);
 133         }
 134
 135         if (!mdc) {
 136                 CERROR("no mdc\n");
 137                 GOTO(out_free, sb = NULL);
 138         }
 139
 140         obd = class_uuid2obd(mdc);
 141         if (!obd) {
 142                 CERROR("MDC %s: not setup or attached\n", mdc);
 143                 GOTO(out_free, sb = NULL);
 144         }
 145
 146         err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid);
 147         if (err) {
 148                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 149                 GOTO(out_free, sb = NULL);
 150         }
 151
 152 #warning Peter: is this the right place to raise the connection level?
 153         mdc_conn = sbi2mdc(sbi)->cl_import.imp_connection;
 154         mdc_conn->c_level = LUSTRE_CONN_FULL;
 155         list_add(&mdc_conn->c_sb_chain, &sbi->ll_conn_chain);
 156
 157         obd = class_uuid2obd(osc);
 158         if (!obd) {
 159                 CERROR("OSC %s: not setup or attached\n", osc);
 160                 GOTO(out_mdc, sb = NULL);
 161         }
 162
 163         err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid);
 164         if (err) {
 165                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 166                 GOTO(out_mdc, sb = NULL);
 167         }
 168
 169         /* XXX: need to store the last_* values somewhere */
 170         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
 171                             &last_xid, &request);
 172         ptlrpc_req_finished(request);
 173         if (err) {
 174                 CERROR("cannot mds_connect: rc = %d\n", err);
 175                 GOTO(out_mdc, sb = NULL);
 176         }
 177         CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
 178         sbi->ll_rootino = rootfid.id;
 179
 180         memset(&osfs, 0, sizeof(osfs));
 181         request = NULL;
 182         err = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 183         ptlrpc_req_finished(request);
 184         sb->s_blocksize = osfs.os_bsize;
 185         sb->s_blocksize_bits = log2(osfs.os_bsize);
 186         sb->s_magic = LL_SUPER_MAGIC;
 187         sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
 188
 189         sb->s_op = &ll_super_operations;
 190
 191         /* make root inode */
 192         request = NULL;
 193         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
 194                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 195         if (err) {
 196                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 197                 GOTO(out_request, sb = NULL);
 198         }
 199
 200         /* initialize committed transaction callback daemon */
 201         spin_lock_init(&sbi->ll_commitcbd_lock);
 202         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 203         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 204         sbi->ll_commitcbd_flags = 0;
 205         err = ll_commitcbd_setup(sbi);
 206         if (err) {
 207                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 208                 GOTO(out_request, sb = NULL);
 209         }
 210
 211         lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
 212         lic.lic_lmm = NULL;
 213         root = iget4(sb, sbi->ll_rootino, NULL, &lic);
 214
 215         if (root) {
 216                 sb->s_root = d_alloc_root(root);
 217         } else {
 218                 CERROR("lustre_lite: bad iget4 for root\n");
 219                 GOTO(out_cdb, sb = NULL);
 220         }
 221
 222         ptlrpc_req_finished(request);
 223
 224 out_dev:
 225         if (mdc)
 226                 OBD_FREE(mdc, strlen(mdc) + 1);
 227         if (osc)
 228                 OBD_FREE(osc, strlen(osc) + 1);
 229
 230         RETURN(sb);
 231
 232 out_cdb:
 233         ll_commitcbd_cleanup(sbi);
 234 out_request:
 235         ptlrpc_req_finished(request);
 236         obd_disconnect(&sbi->ll_osc_conn);
 237 out_mdc:
 238         obd_disconnect(&sbi->ll_mdc_conn);
 239 out_free:
 240         OBD_FREE(sbi, sizeof(*sbi));
 241
 242         MOD_DEC_USE_COUNT;
 243         goto out_dev;
 244 } /* ll_read_super */
 245
 246 static void ll_put_super(struct super_block *sb)
 247 {
 248         struct ll_sb_info *sbi = ll_s2sbi(sb);
 249
 250         ENTRY;
 251         list_del(&sbi->ll_conn_chain);
 252         ll_commitcbd_cleanup(sbi);
 253         obd_disconnect(&sbi->ll_osc_conn);
 254         obd_disconnect(&sbi->ll_mdc_conn);
 255         OBD_FREE(sbi, sizeof(*sbi));
 256
 257         MOD_DEC_USE_COUNT;
 258         EXIT;
 259 } /* ll_put_super */
 260
 261 static void ll_clear_inode(struct inode *inode)
 262 {
 263         ENTRY;
 264
 265         if (atomic_read(&inode->i_count) == 0) {
 266                 struct ll_inode_info *lli = ll_i2info(inode);
 267                 struct lov_stripe_md *lsm = lli->lli_smd;
 268                 char *symlink_name = lli->lli_symlink_name;
 269
 270                 if (lsm) {
 271                         OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
 272                         lli->lli_smd = NULL;
 273                 }
 274                 if (symlink_name) {
 275                         OBD_FREE(symlink_name, strlen(symlink_name) + 1);
 276                         lli->lli_symlink_name = NULL;
 277                 }
 278         }
 279         EXIT;
 280 }
 281
 282 static void ll_delete_inode(struct inode *inode)
 283 {
 284         ENTRY;
 285         if (S_ISREG(inode->i_mode)) {
 286                 int err;
 287                 struct obdo *oa;
 288                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 289
 290                 if (!lsm)
 291                         GOTO(out, -EINVAL);
 292
 293                 if (lsm->lsm_object_id == 0) {
 294                         CERROR("This really happens\n");
 295                         /* No obdo was ever created */
 296                         GOTO(out, 0);
 297                 }
 298
 299                 oa = obdo_alloc();
 300                 if (oa == NULL)
 301                         GOTO(out, -ENOMEM);
 302
 303                 oa->o_id = lsm->lsm_object_id;
 304                 oa->o_easize = ll_mds_easize(inode->i_sb);
 305                 oa->o_mode = inode->i_mode;
 306                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLTYPE;
 307
 308                 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
 309                 obdo_free(oa);
 310                 CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
 311                        lsm->lsm_object_id, err);
 312         }
 313 out:
 314         clear_inode(inode);
 315         EXIT;
 316 }
 317
 318 /* like inode_setattr, but doesn't mark the inode dirty */
 319 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 320 {
 321         unsigned int ia_valid = attr->ia_valid;
 322         int error = 0;
 323
 324         if ((ia_valid & ATTR_SIZE) && trunc) {
 325                 error = vmtruncate(inode, attr->ia_size);
 326                 if (error)
 327                         goto out;
 328         } else if (ia_valid & ATTR_SIZE)
 329                 inode->i_size = attr->ia_size;
 330
 331         if (ia_valid & ATTR_UID)
 332                 inode->i_uid = attr->ia_uid;
 333         if (ia_valid & ATTR_GID)
 334                 inode->i_gid = attr->ia_gid;
 335         if (ia_valid & ATTR_ATIME)
 336                 inode->i_atime = attr->ia_atime;
 337         if (ia_valid & ATTR_MTIME)
 338                 inode->i_mtime = attr->ia_mtime;
 339         if (ia_valid & ATTR_CTIME)
 340                 inode->i_ctime = attr->ia_ctime;
 341         if (ia_valid & ATTR_MODE) {
 342                 inode->i_mode = attr->ia_mode;
 343                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 344                         inode->i_mode &= ~S_ISGID;
 345         }
 346 out:
 347         return error;
 348 }
 349
 350 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 351 {
 352         struct ptlrpc_request *request = NULL;
 353         struct ll_sb_info *sbi = ll_i2sbi(inode);
 354         int err;
 355
 356         ENTRY;
 357
 358         /* change incore inode */
 359         ll_attr2inode(inode, attr, do_trunc);
 360
 361         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 362         if (err)
 363                 CERROR("mdc_setattr fails (%d)\n", err);
 364
 365         ptlrpc_req_finished(request);
 366
 367         RETURN(err);
 368 }
 369
 370 int ll_setattr(struct dentry *de, struct iattr *attr)
 371 {
 372         int rc = inode_change_ok(de->d_inode, attr);
 373
 374         if (rc)
 375                 return rc;
 376
 377         return ll_inode_setattr(de->d_inode, attr, 1);
 378 }
 379
 380 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 381 {
 382         struct ptlrpc_request *request = NULL;
 383         struct ll_sb_info *sbi = ll_s2sbi(sb);
 384         struct obd_statfs osfs;
 385         int rc;
 386         ENTRY;
 387
 388         memset(sfs, 0, sizeof(*sfs));
 389         rc = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 390         statfs_unpack(sfs, &osfs);
 391         ptlrpc_req_finished(request);
 392         if (rc)
 393                 CERROR("mdc_statfs fails: rc = %d\n", rc);
 394         else
 395                 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
 396                        " objects "LPU64"/"LPU64"\n",
 397                        osfs.os_bavail, osfs.os_blocks,
 398                        osfs.os_ffree, osfs.os_files);
 399
 400         /* temporary until mds_statfs returns statfs info for all OSTs */
 401         if (!rc) {
 402                 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
 403                 if (rc) {
 404                         CERROR("obd_statfs fails: rc = %d\n", rc);
 405                         GOTO(out, rc);
 406                 }
 407                 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
 408                        " objects "LPU64"/"LPU64"\n",
 409                        osfs.os_bavail, osfs.os_blocks,
 410                        osfs.os_ffree, osfs.os_files);
 411
 412                 while (osfs.os_blocks > ~0UL) {
 413                         sfs->f_bsize <<= 1;
 414
 415                         osfs.os_blocks >>= 1;
 416                         osfs.os_bfree >>= 1;
 417                         osfs.os_bavail >>= 1;
 418                 }
 419                 sfs->f_blocks = osfs.os_blocks;
 420                 sfs->f_bfree = osfs.os_bfree;
 421                 sfs->f_bavail = osfs.os_bavail;
 422                 if (osfs.os_ffree < (__u64)sfs->f_ffree)
 423                         sfs->f_ffree = osfs.os_ffree;
 424         }
 425
 426 out:
 427         RETURN(rc);
 428 }
 429
 430 void ll_update_inode(struct inode *inode, struct mds_body *body)
 431 {
 432         if (body->valid & OBD_MD_FLID)
 433                 inode->i_ino = body->ino;
 434         if (body->valid & OBD_MD_FLATIME)
 435                 inode->i_atime = body->atime;
 436         if (body->valid & OBD_MD_FLMTIME)
 437                 inode->i_mtime = body->mtime;
 438         if (body->valid & OBD_MD_FLCTIME)
 439                 inode->i_ctime = body->ctime;
 440         if (body->valid & OBD_MD_FLMODE)
 441                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
 442         if (body->valid & OBD_MD_FLTYPE)
 443                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
 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
 460 static void ll_read_inode2(struct inode *inode, void *opaque)
 461 {
 462         struct ll_read_inode2_cookie *lic = opaque;
 463         struct mds_body *body = lic->lic_body;
 464         struct ll_inode_info *lli = ll_i2info(inode);
 465         ENTRY;
 466
 467         sema_init(&lli->lli_open_sem, 1);
 468         lli->lli_mount_epoch = ll_i2sbi(inode)->ll_mount_epoch;
 469
 470         /* core attributes first */
 471         ll_update_inode(inode, body);
 472
 473         //if (body->valid & OBD_MD_FLEASIZE)
 474         if (lic && lic->lic_lmm) {
 475                 struct lov_mds_md *lmm = lic->lic_lmm;
 476                 int size;
 477
 478                 if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
 479                         CERROR("Striping metadata size error %ld\n",
 480                                inode->i_ino);
 481                         LBUG();
 482                 }
 483                 size = ll_ost_easize(inode->i_sb);
 484                 OBD_ALLOC(lli->lli_smd, size);
 485                 if (!lli->lli_smd) {
 486                         CERROR("No memory for %d\n", size);
 487                         LBUG();
 488                 }
 489                 lov_unpackmd(lli->lli_smd, lmm);
 490         } else {
 491                 lli->lli_smd = NULL;
 492         }
 493
 494         /* Get the authoritative file size */
 495         if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
 496                 int rc;
 497
 498                 rc = ll_file_size(inode, lli->lli_smd);
 499                 if (rc) {
 500                         CERROR("ll_file_size: %d\n", rc);
 501                         /* FIXME: need to somehow prevent inode creation */
 502                         LBUG();
 503                 }
 504         }
 505
 506         /* OIDEBUG(inode); */
 507
 508         if (S_ISREG(inode->i_mode)) {
 509                 inode->i_op = &ll_file_inode_operations;
 510                 inode->i_fop = &ll_file_operations;
 511                 inode->i_mapping->a_ops = &ll_aops;
 512                 EXIT;
 513         } else if (S_ISDIR(inode->i_mode)) {
 514                 inode->i_op = &ll_dir_inode_operations;
 515                 inode->i_fop = &ll_dir_operations;
 516                 inode->i_mapping->a_ops = &ll_dir_aops;
 517                 EXIT;
 518         } else if (S_ISLNK(inode->i_mode)) {
 519                 inode->i_op = &ll_fast_symlink_inode_operations;
 520                 EXIT;
 521         } else {
 522                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 523                 EXIT;
 524         }
 525 }
 526
 527 static inline void invalidate_request_list(struct list_head *req_list)
 528 {
 529         struct list_head *tmp, *n;
 530         list_for_each_safe(tmp, n, req_list) {
 531                 struct ptlrpc_request *req =
 532                         list_entry(tmp, struct ptlrpc_request, rq_list);
 533                 CERROR("invalidating req xid %d op %d to %s:%d\n",
 534                        (unsigned long long)req->rq_xid, req->rq_reqmsg->opc,
 535                        req->rq_connection->c_remote_uuid,
 536                        req->rq_import->imp_client->cli_request_portal);
 537                 req->rq_flags |= PTL_RPC_FL_ERR;
 538                 wake_up(&req->rq_wait_for_rep);
 539         }
 540 }
 541
 542 void ll_umount_begin(struct super_block *sb)
 543 {
 544         struct ll_sb_info *sbi = ll_s2sbi(sb);
 545         struct list_head *ctmp;
 546
 547         ENTRY;
 548
 549         list_for_each(ctmp, &sbi->ll_conn_chain) {
 550                 struct ptlrpc_connection *conn;
 551                 conn = list_entry(ctmp, struct ptlrpc_connection, c_sb_chain);
 552
 553                 spin_lock(&conn->c_lock);
 554                 conn->c_flags |= CONN_INVALID;
 555                 invalidate_request_list(&conn->c_sending_head);
 556                 invalidate_request_list(&conn->c_delayed_head);
 557                 spin_unlock(&conn->c_unlock);
 558         }
 559
 560         EXIT;
 561 }
 562
 563 /* exported operations */
 564 struct super_operations ll_super_operations =
 565 {
 566         read_inode2: ll_read_inode2,
 567         clear_inode: ll_clear_inode,
 568         delete_inode: ll_delete_inode,
 569         put_super: ll_put_super,
 570         statfs: ll_statfs,
 571         umount_begin: ll_umount_begin
 572 };
 573
 574 struct file_system_type lustre_lite_fs_type = {
 575         "lustre_lite", 0, ll_read_super, NULL
 576 };
 577
 578 static int __init init_lustre_lite(void)
 579 {
 580         printk(KERN_INFO "Lustre Lite 0.5.14, info@clusterfs.com\n");
 581         ll_file_data_slab = kmem_cache_create("ll_file_data",
 582                                               sizeof(struct ll_file_data), 0,
 583                                               SLAB_HWCACHE_ALIGN, NULL, NULL);
 584         if (ll_file_data_slab == NULL)
 585                 return -ENOMEM;
 586         return register_filesystem(&lustre_lite_fs_type);
 587 }
 588
 589 static void __exit exit_lustre_lite(void)
 590 {
 591         unregister_filesystem(&lustre_lite_fs_type);
 592         kmem_cache_destroy(ll_file_data_slab);
 593 }
 594
 595 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 596 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 597 MODULE_LICENSE("GPL");
 598
 599 module_init(init_lustre_lite);
 600 module_exit(exit_lustre_lite);