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/lustre_dlm.h>
  19 #include <linux/init.h>
  20
  21 kmem_cache_t *ll_file_data_slab;
  22 extern struct address_space_operations ll_aops;
  23 extern struct address_space_operations ll_dir_aops;
  24 struct super_operations ll_super_operations;
  25
  26 extern int ll_recover(struct ptlrpc_client *);
  27 extern int ll_commitcbd_setup(struct ll_sb_info *);
  28 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  29
  30 static char *ll_read_opt(const char *opt, char *data)
  31 {
  32         char *value;
  33         char *retval;
  34         ENTRY;
  35
  36         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  37         if ( strncmp(opt, data, strlen(opt)) )
  38                 RETURN(NULL);
  39         if ( (value = strchr(data, '=')) == NULL )
  40                 RETURN(NULL);
  41
  42         value++;
  43         OBD_ALLOC(retval, strlen(value) + 1);
  44         if ( !retval ) {
  45                 CERROR("out of memory!\n");
  46                 RETURN(NULL);
  47         }
  48
  49         memcpy(retval, value, strlen(value)+1);
  50         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  51         RETURN(retval);
  52 }
  53
  54 static void ll_options(char *options, char **ost, char **mds)
  55 {
  56         char *this_char;
  57         ENTRY;
  58
  59         if (!options) {
  60                 EXIT;
  61                 return;
  62         }
  63
  64         for (this_char = strtok (options, ",");
  65              this_char != NULL;
  66              this_char = strtok (NULL, ",")) {
  67                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
  68                 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
  69                      (!*mds && (*mds = ll_read_opt("mdc", this_char))) )
  70                         continue;
  71         }
  72         EXIT;
  73 }
  74
  75 #ifndef log2
  76 #define log2(n) ffz(~(n))
  77 #endif
  78
  79 static struct super_block * ll_read_super(struct super_block *sb,
  80                                           void *data, int silent)
  81 {
  82         struct inode *root = 0;
  83         struct obd_device *obd;
  84         struct ll_sb_info *sbi;
  85         char *osc = NULL;
  86         char *mdc = NULL;
  87         int err;
  88         struct ll_fid rootfid;
  89         struct statfs sfs;
  90         __u64 last_committed;
  91         __u32 last_xid;
  92         struct ptlrpc_request *request = NULL;
  93         struct ll_inode_md md;
  94         class_uuid_t uuid;
  95
  96         ENTRY;
  97         MOD_INC_USE_COUNT;
  98
  99         OBD_ALLOC(sbi, sizeof(*sbi));
 100         if (!sbi) {
 101                 MOD_DEC_USE_COUNT;
 102                 RETURN(NULL);
 103         }
 104
 105         generate_random_uuid(uuid);
 106         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 107
 108         sb->u.generic_sbp = sbi;
 109
 110         ll_options(data, &osc, &mdc);
 111
 112         if (!osc) {
 113                 CERROR("no osc\n");
 114                 GOTO(out_free, sb = NULL);
 115         }
 116
 117         if (!mdc) {
 118                 CERROR("no mdc\n");
 119                 GOTO(out_free, sb = NULL);
 120         }
 121
 122         obd = class_uuid2obd(mdc);
 123         if (!obd) {
 124                 CERROR("MDC %s: not setup or attached\n", mdc);
 125                 GOTO(out_free, sb = NULL);
 126         }
 127
 128 #if 0
 129         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mdc_conn);
 130         if (err) {
 131                 CERROR("cannot connect to MDC: rc = %d\n", err);
 132                 GOTO(out_rpc, sb = NULL);
 133         }
 134 #endif
 135
 136         err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid);
 137         if (err) {
 138                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 139                 GOTO(out_free, sb = NULL);
 140         }
 141         sbi2mdc(sbi)->cl_conn->c_level = LUSTRE_CONN_FULL;
 142
 143         obd = class_uuid2obd(osc);
 144         if (!obd) {
 145                 CERROR("OSC %s: not setup or attached\n", osc);
 146                 GOTO(out_mdc, sb = NULL);
 147         }
 148         err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid);
 149         if (err) {
 150                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 151                 GOTO(out_mdc, sb = NULL);
 152         }
 153
 154         /* XXX: need to store the last_* values somewhere */
 155         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
 156                             &last_xid, &request);
 157         ptlrpc_req_finished(request);
 158         if (err) {
 159                 CERROR("cannot mds_connect: rc = %d\n", err);
 160                 GOTO(out_mdc, sb = NULL);
 161         }
 162         CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
 163         sbi->ll_rootino = rootfid.id;
 164
 165         memset(&sfs, 0, sizeof(sfs));
 166         request = NULL;
 167         err = mdc_statfs(&sbi->ll_mdc_conn, &sfs, &request);
 168         ptlrpc_req_finished(request);
 169         sb->s_blocksize = sfs.f_bsize;
 170         sb->s_blocksize_bits = log2(sfs.f_bsize);
 171         sb->s_magic = LL_SUPER_MAGIC;
 172         sb->s_maxbytes = (1ULL << (32 + 9)) - sfs.f_bsize;
 173
 174         sb->s_op = &ll_super_operations;
 175
 176         /* make root inode */
 177         request = NULL;
 178         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
 179                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 180         if (err) {
 181                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 182                 GOTO(out_request, sb = NULL);
 183         }
 184
 185         /* initialize committed transaction callback daemon */
 186         spin_lock_init(&sbi->ll_commitcbd_lock);
 187         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 188         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 189         sbi->ll_commitcbd_flags = 0;
 190         err = ll_commitcbd_setup(sbi);
 191         if (err) {
 192                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 193                 GOTO(out_request, sb = NULL);
 194         }
 195
 196         md.body = lustre_msg_buf(request->rq_repmsg, 0);
 197         md.md = NULL;
 198         root = iget4(sb, sbi->ll_rootino, NULL, &md);
 199
 200         if (root) {
 201                 sb->s_root = d_alloc_root(root);
 202         } else {
 203                 CERROR("lustre_lite: bad iget4 for root\n");
 204                 GOTO(out_cdb, sb = NULL);
 205         }
 206
 207         ptlrpc_req_finished(request);
 208
 209 out_dev:
 210         if (mdc)
 211                 OBD_FREE(mdc, strlen(mdc) + 1);
 212         if (osc)
 213                 OBD_FREE(osc, strlen(osc) + 1);
 214
 215         RETURN(sb);
 216
 217 out_cdb:
 218         ll_commitcbd_cleanup(sbi);
 219 out_request:
 220         ptlrpc_req_finished(request);
 221         obd_disconnect(&sbi->ll_osc_conn);
 222 out_mdc:
 223         obd_disconnect(&sbi->ll_mdc_conn);
 224 out_free:
 225         OBD_FREE(sbi, sizeof(*sbi));
 226
 227         MOD_DEC_USE_COUNT;
 228         goto out_dev;
 229 } /* ll_read_super */
 230
 231 static void ll_put_super(struct super_block *sb)
 232 {
 233         struct ll_sb_info *sbi = ll_s2sbi(sb);
 234         ENTRY;
 235         ll_commitcbd_cleanup(sbi);
 236         obd_disconnect(&sbi->ll_osc_conn);
 237         obd_disconnect(&sbi->ll_mdc_conn);
 238         OBD_FREE(sbi, sizeof(*sbi));
 239
 240         MOD_DEC_USE_COUNT;
 241         EXIT;
 242 } /* ll_put_super */
 243
 244 static void ll_clear_inode(struct inode *inode)
 245 {
 246         if (atomic_read(&inode->i_count) == 0) {
 247                 struct ll_inode_info *lli = ll_i2info(inode);
 248                 struct lov_stripe_md *md = lli->lli_smd;
 249
 250                 if (md) {
 251                         OBD_FREE(md, md->lmd_easize);
 252                         lli->lli_smd = NULL;
 253                 }
 254                 if (lli->lli_symlink_name) {
 255                         OBD_FREE(lli->lli_symlink_name,
 256                                  strlen(lli->lli_symlink_name) + 1);
 257                         lli->lli_symlink_name = NULL;
 258                 }
 259         }
 260 }
 261
 262 static void ll_delete_inode(struct inode *inode)
 263 {
 264         if (S_ISREG(inode->i_mode)) {
 265                 int err;
 266                 struct obdo *oa;
 267                 struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
 268
 269                 if (!md)
 270                         GOTO(out, -EINVAL);
 271
 272                 if (md->lmd_object_id == 0) {
 273                         CERROR("This really happens\n");
 274                         /* No obdo was ever created */
 275                         GOTO(out, 0);
 276                 }
 277
 278                 oa = obdo_alloc();
 279                 if (oa == NULL)
 280                         GOTO(out, -ENOMEM);
 281
 282                 oa->o_id = md->lmd_object_id;
 283                 oa->o_easize = md->lmd_easize;
 284                 oa->o_mode = inode->i_mode;
 285                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLMODE;
 286
 287                 err = obd_destroy(ll_i2obdconn(inode), oa, md);
 288                 obdo_free(oa);
 289                 CDEBUG(D_SUPER, "obd destroy of %Ld error %d\n",
 290                        md->lmd_object_id, err);
 291         }
 292 out:
 293         clear_inode(inode);
 294 }
 295
 296 /* like inode_setattr, but doesn't mark the inode dirty */
 297 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 298 {
 299         unsigned int ia_valid = attr->ia_valid;
 300         int error = 0;
 301
 302         if ((ia_valid & ATTR_SIZE) && trunc) {
 303                 error = vmtruncate(inode, attr->ia_size);
 304                 if (error)
 305                         goto out;
 306         } else if (ia_valid & ATTR_SIZE)
 307                 inode->i_size = attr->ia_size;
 308
 309         if (ia_valid & ATTR_UID)
 310                 inode->i_uid = attr->ia_uid;
 311         if (ia_valid & ATTR_GID)
 312                 inode->i_gid = attr->ia_gid;
 313         if (ia_valid & ATTR_ATIME)
 314                 inode->i_atime = attr->ia_atime;
 315         if (ia_valid & ATTR_MTIME)
 316                 inode->i_mtime = attr->ia_mtime;
 317         if (ia_valid & ATTR_CTIME)
 318                 inode->i_ctime = attr->ia_ctime;
 319         if (ia_valid & ATTR_MODE) {
 320                 inode->i_mode = attr->ia_mode;
 321                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 322                         inode->i_mode &= ~S_ISGID;
 323         }
 324 out:
 325         return error;
 326 }
 327
 328 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 329 {
 330         struct ptlrpc_request *request = NULL;
 331         struct ll_sb_info *sbi = ll_i2sbi(inode);
 332         int err;
 333
 334         ENTRY;
 335
 336         /* change incore inode */
 337         ll_attr2inode(inode, attr, do_trunc);
 338
 339         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 340         if (err)
 341                 CERROR("mdc_setattr fails (%d)\n", err);
 342
 343         ptlrpc_req_finished(request);
 344
 345         RETURN(err);
 346 }
 347
 348 int ll_setattr(struct dentry *de, struct iattr *attr)
 349 {
 350         int rc = inode_change_ok(de->d_inode, attr);
 351
 352         if (rc)
 353                 return rc;
 354
 355         return ll_inode_setattr(de->d_inode, attr, 1);
 356 }
 357
 358 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 359 {
 360         struct ptlrpc_request *request = NULL;
 361         struct ll_sb_info *sbi = ll_s2sbi(sb);
 362         int rc;
 363         ENTRY;
 364
 365         memset(sfs, 0, sizeof(*sfs));
 366         rc = mdc_statfs(&sbi->ll_mdc_conn, sfs, &request);
 367         ptlrpc_req_finished(request);
 368         if (rc)
 369                 CERROR("obd_statfs fails: rc = %d\n", rc);
 370         else
 371                 CDEBUG(D_SUPER, "statfs shows blocks %ld/%ld objects %ld/%ld\n",
 372                        sfs->f_bavail, sfs->f_blocks, sfs->f_files,sfs->f_ffree);
 373
 374         /* temporary until mds_statfs returns statfs info for all OSTs */
 375         if (!rc) {
 376                 struct statfs obd_sfs;
 377
 378                 rc = obd_statfs(&sbi->ll_osc_conn, &obd_sfs);
 379                 if (rc) {
 380                         CERROR("obd_statfs fails: rc = %d\n", rc);
 381                         GOTO(out, rc);
 382                 }
 383                 CDEBUG(D_SUPER, "obd_statfs returns blocks %ld/%ld, "
 384                        "objects %ld/%ld\n",
 385                        obd_sfs.f_bavail, obd_sfs.f_blocks,
 386                        obd_sfs.f_ffree, obd_sfs.f_files);
 387
 388                 sfs->f_bfree = obd_sfs.f_bfree;
 389                 sfs->f_bavail = obd_sfs.f_bavail;
 390                 sfs->f_blocks = obd_sfs.f_blocks;
 391                 if (obd_sfs.f_ffree < sfs->f_ffree)
 392                         sfs->f_ffree = obd_sfs.f_ffree;
 393         }
 394
 395 out:
 396         RETURN(rc);
 397 }
 398
 399 inline int ll_stripe_md_size(struct super_block *sb)
 400 {
 401         struct client_obd *mdc = sbi2mdc(ll_s2sbi(sb));
 402         return mdc->cl_max_mdsize;
 403 }
 404
 405 static int ll_file_size(struct inode *inode, struct lov_stripe_md *md,
 406                         __u64 *size)
 407 {
 408         struct ll_sb_info *sbi = ll_i2sbi(inode);
 409         struct lustre_handle *lockhs;
 410         struct obdo oa;
 411         int err, rc;
 412
 413         rc = ll_size_lock(inode, md, 0, LCK_PR, &lockhs);
 414         if (rc != ELDLM_OK) {
 415                 CERROR("lock enqueue: %d\n", rc);
 416                 RETURN(rc);
 417         }
 418
 419         /* FIXME: I don't like this; why doesn't osc_getattr get o_id from md
 420          * like lov_getattr? --phil */
 421         oa.o_id = md->lmd_object_id;
 422         oa.o_mode = S_IFREG;
 423         oa.o_valid = OBD_MD_FLID | OBD_MD_FLMODE | OBD_MD_FLSIZE;
 424         rc = obd_getattr(&sbi->ll_osc_conn, &oa, md);
 425         if (!rc)
 426                 *size = oa.o_size;
 427
 428         err = ll_size_unlock(inode, md, LCK_PR, lockhs);
 429         if (err != ELDLM_OK) {
 430                 CERROR("lock cancel: %d\n", err);
 431                 LBUG();
 432         }
 433         RETURN(rc);
 434 }
 435
 436 static void ll_read_inode2(struct inode *inode, void *opaque)
 437 {
 438         struct ll_inode_md *md = opaque;
 439         struct mds_body *body = md->body;
 440         struct ll_inode_info *ii = ll_i2info(inode);
 441         ENTRY;
 442
 443         sema_init(&ii->lli_open_sem, 1);
 444         INIT_LIST_HEAD(&ii->lli_osc_locks);
 445
 446         /* core attributes first */
 447         if (body->valid & OBD_MD_FLID)
 448                 inode->i_ino = body->ino;
 449         if (body->valid & OBD_MD_FLATIME)
 450                 inode->i_atime = body->atime;
 451         if (body->valid & OBD_MD_FLMTIME)
 452                 inode->i_mtime = body->mtime;
 453         if (body->valid & OBD_MD_FLCTIME)
 454                 inode->i_ctime = body->ctime;
 455         if (body->valid & OBD_MD_FLMODE)
 456                 inode->i_mode = body->mode;
 457         if (body->valid & OBD_MD_FLUID)
 458                 inode->i_uid = body->uid;
 459         if (body->valid & OBD_MD_FLGID)
 460                 inode->i_gid = body->gid;
 461         if (body->valid & OBD_MD_FLFLAGS)
 462                 inode->i_flags = body->flags;
 463         if (body->valid & OBD_MD_FLNLINK)
 464                 inode->i_nlink = body->nlink;
 465         if (body->valid & OBD_MD_FLGENER)
 466                 inode->i_generation = body->generation;
 467         if (body->valid & OBD_MD_FLRDEV)
 468                 inode->i_rdev = body->extra;
 469         //if (body->valid & OBD_MD_FLSIZE)
 470         //        inode->i_size = body->size;
 471
 472         /* Get the authoritative file size */
 473         if (md && md->md && inode->i_mode & S_IFREG) {
 474                 int rc;
 475                 rc = ll_file_size(inode, md->md, &inode->i_size);
 476                 if (rc) {
 477                         CERROR("ll_file_size: %d\n", rc);
 478                         /* FIXME: need to somehow prevent inode creation */
 479                         LBUG();
 480                 }
 481         }
 482
 483         //if (body->valid & OBD_MD_FLEASIZE)
 484         if (md && md->md && md->md->lmd_stripe_count) {
 485                 struct lov_stripe_md *smd = md->md;
 486                 int size = ll_stripe_md_size(inode->i_sb);
 487                 if (md->md->lmd_easize != size) {
 488                         CERROR("Striping metadata size error %ld\n",
 489                                inode->i_ino);
 490                         LBUG();
 491                 }
 492                 OBD_ALLOC(ii->lli_smd, size);
 493                 if (!ii->lli_smd){
 494                         CERROR("No memory for %d\n", size);
 495                         LBUG();
 496                 }
 497                 memcpy(ii->lli_smd, smd, size);
 498         }
 499
 500         /* OIDEBUG(inode); */
 501
 502         if (S_ISREG(inode->i_mode)) {
 503                 inode->i_op = &ll_file_inode_operations;
 504                 inode->i_fop = &ll_file_operations;
 505                 inode->i_mapping->a_ops = &ll_aops;
 506                 EXIT;
 507         } else if (S_ISDIR(inode->i_mode)) {
 508                 inode->i_op = &ll_dir_inode_operations;
 509                 inode->i_fop = &ll_dir_operations;
 510                 inode->i_mapping->a_ops = &ll_dir_aops;
 511                 EXIT;
 512         } else if (S_ISLNK(inode->i_mode)) {
 513                 inode->i_op = &ll_fast_symlink_inode_operations;
 514                 EXIT;
 515         } else {
 516                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 517                 EXIT;
 518         }
 519
 520         return;
 521 }
 522
 523 /* exported operations */
 524 struct super_operations ll_super_operations =
 525 {
 526         read_inode2: ll_read_inode2,
 527         clear_inode: ll_clear_inode,
 528         delete_inode: ll_delete_inode,
 529         put_super: ll_put_super,
 530         statfs: ll_statfs
 531 };
 532
 533 struct file_system_type lustre_lite_fs_type = {
 534         "lustre_lite", 0, ll_read_super, NULL
 535 };
 536
 537 static int __init init_lustre_lite(void)
 538 {
 539         printk(KERN_INFO "Lustre Lite 0.0.1, info@clusterfs.com\n");
 540         ll_file_data_slab = kmem_cache_create("ll_file_data",
 541                                               sizeof(struct ll_file_data), 0,
 542                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 543         if (ll_file_data_slab == NULL)
 544                 return -ENOMEM;
 545         return register_filesystem(&lustre_lite_fs_type);
 546 }
 547
 548 static void __exit exit_lustre_lite(void)
 549 {
 550         unregister_filesystem(&lustre_lite_fs_type);
 551         kmem_cache_destroy(ll_file_data_slab);
 552 }
 553
 554 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 555 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 556 MODULE_LICENSE("GPL");
 557
 558 module_init(init_lustre_lite);
 559 module_exit(exit_lustre_lite);