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 recovd_data *, int);
  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 int ll_set_opt(const char *opt, char *data, int fl)
  56 {
  57         ENTRY;
  58
  59         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  60         if ( strncmp(opt, data, strlen(opt)) )
  61                 RETURN(0);
  62         else
  63                 RETURN(fl);
  64 }
  65
  66 static void ll_options(char *options, char **ost, char **mds, int *flags)
  67 {
  68         char *this_char;
  69         ENTRY;
  70
  71         if (!options) {
  72                 EXIT;
  73                 return;
  74         }
  75
  76         for (this_char = strtok (options, ",");
  77              this_char != NULL;
  78              this_char = strtok (NULL, ",")) {
  79                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
  80                 if ( (!*ost && (*ost = ll_read_opt("osc", this_char)))||
  81                      (!*mds && (*mds = ll_read_opt("mdc", this_char)))||
  82                      (!(*flags & LL_SBI_NOLCK) && ((*flags) = (*flags) |
  83                       ll_set_opt("nolock", this_char, LL_SBI_NOLCK))) )
  84                         continue;
  85         }
  86         EXIT;
  87 }
  88
  89 #ifndef log2
  90 #define log2(n) ffz(~(n))
  91 #endif
  92
  93 static struct super_block * ll_read_super(struct super_block *sb,
  94                                           void *data, int silent)
  95 {
  96         struct inode *root = 0;
  97         struct obd_device *obd;
  98         struct ll_sb_info *sbi;
  99         char *osc = NULL;
 100         char *mdc = NULL;
 101         int err;
 102         struct ll_fid rootfid;
 103         struct obd_statfs osfs;
 104         __u64 last_committed;
 105         __u64 last_xid;
 106         struct ptlrpc_request *request = NULL;
 107         struct ll_read_inode2_cookie lic;
 108         class_uuid_t uuid;
 109
 110         ENTRY;
 111         MOD_INC_USE_COUNT;
 112
 113         OBD_ALLOC(sbi, sizeof(*sbi));
 114         if (!sbi) {
 115                 MOD_DEC_USE_COUNT;
 116                 RETURN(NULL);
 117         }
 118
 119         generate_random_uuid(uuid);
 120         class_uuid_unparse(uuid, sbi->ll_sb_uuid);
 121
 122         sb->u.generic_sbp = sbi;
 123
 124         ll_options(data, &osc, &mdc, &sbi->ll_flags);
 125
 126         if (!osc) {
 127                 CERROR("no osc\n");
 128                 GOTO(out_free, sb = NULL);
 129         }
 130
 131         if (!mdc) {
 132                 CERROR("no mdc\n");
 133                 GOTO(out_free, sb = NULL);
 134         }
 135
 136         obd = class_uuid2obd(mdc);
 137         if (!obd) {
 138                 CERROR("MDC %s: not setup or attached\n", mdc);
 139                 GOTO(out_free, sb = NULL);
 140         }
 141
 142         err = obd_connect(&sbi->ll_mdc_conn, obd, sbi->ll_sb_uuid);
 143         if (err) {
 144                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 145                 GOTO(out_free, sb = NULL);
 146         }
 147
 148 #warning Peter: is this the right place to raise the connection level?
 149         sbi2mdc(sbi)->cl_import.imp_connection->c_level = LUSTRE_CONN_FULL;
 150
 151         obd = class_uuid2obd(osc);
 152         if (!obd) {
 153                 CERROR("OSC %s: not setup or attached\n", osc);
 154                 GOTO(out_mdc, sb = NULL);
 155         }
 156         err = obd_connect(&sbi->ll_osc_conn, obd, sbi->ll_sb_uuid);
 157         if (err) {
 158                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 159                 GOTO(out_mdc, sb = NULL);
 160         }
 161
 162         /* XXX: need to store the last_* values somewhere */
 163         err = mdc_getstatus(&sbi->ll_mdc_conn, &rootfid, &last_committed,
 164                             &last_xid, &request);
 165         ptlrpc_req_finished(request);
 166         if (err) {
 167                 CERROR("cannot mds_connect: rc = %d\n", err);
 168                 GOTO(out_mdc, sb = NULL);
 169         }
 170         CDEBUG(D_SUPER, "rootfid %Ld\n", (unsigned long long)rootfid.id);
 171         sbi->ll_rootino = rootfid.id;
 172
 173         memset(&osfs, 0, sizeof(osfs));
 174         request = NULL;
 175         err = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 176         ptlrpc_req_finished(request);
 177         sb->s_blocksize = osfs.os_bsize;
 178         sb->s_blocksize_bits = log2(osfs.os_bsize);
 179         sb->s_magic = LL_SUPER_MAGIC;
 180         sb->s_maxbytes = (1ULL << (32 + 9)) - osfs.os_bsize;
 181
 182         sb->s_op = &ll_super_operations;
 183
 184         /* make root inode */
 185         request = NULL;
 186         err = mdc_getattr(&sbi->ll_mdc_conn, sbi->ll_rootino, S_IFDIR,
 187                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 188         if (err) {
 189                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 190                 GOTO(out_request, sb = NULL);
 191         }
 192
 193         /* initialize committed transaction callback daemon */
 194         spin_lock_init(&sbi->ll_commitcbd_lock);
 195         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 196         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 197         sbi->ll_commitcbd_flags = 0;
 198         err = ll_commitcbd_setup(sbi);
 199         if (err) {
 200                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 201                 GOTO(out_request, sb = NULL);
 202         }
 203
 204         lic.lic_body = lustre_msg_buf(request->rq_repmsg, 0);
 205         lic.lic_lmm = NULL;
 206         root = iget4(sb, sbi->ll_rootino, NULL, &lic);
 207
 208         if (root) {
 209                 sb->s_root = d_alloc_root(root);
 210         } else {
 211                 CERROR("lustre_lite: bad iget4 for root\n");
 212                 GOTO(out_cdb, sb = NULL);
 213         }
 214
 215         ptlrpc_req_finished(request);
 216
 217 out_dev:
 218         if (mdc)
 219                 OBD_FREE(mdc, strlen(mdc) + 1);
 220         if (osc)
 221                 OBD_FREE(osc, strlen(osc) + 1);
 222
 223         RETURN(sb);
 224
 225 out_cdb:
 226         ll_commitcbd_cleanup(sbi);
 227 out_request:
 228         ptlrpc_req_finished(request);
 229         obd_disconnect(&sbi->ll_osc_conn);
 230 out_mdc:
 231         obd_disconnect(&sbi->ll_mdc_conn);
 232 out_free:
 233         OBD_FREE(sbi, sizeof(*sbi));
 234
 235         MOD_DEC_USE_COUNT;
 236         goto out_dev;
 237 } /* ll_read_super */
 238
 239 static void ll_put_super(struct super_block *sb)
 240 {
 241         struct ll_sb_info *sbi = ll_s2sbi(sb);
 242         ENTRY;
 243         ll_commitcbd_cleanup(sbi);
 244         obd_disconnect(&sbi->ll_osc_conn);
 245         obd_disconnect(&sbi->ll_mdc_conn);
 246         OBD_FREE(sbi, sizeof(*sbi));
 247
 248         MOD_DEC_USE_COUNT;
 249         EXIT;
 250 } /* ll_put_super */
 251
 252 static void ll_clear_inode(struct inode *inode)
 253 {
 254         if (atomic_read(&inode->i_count) == 0) {
 255                 struct ll_inode_info *lli = ll_i2info(inode);
 256                 struct lov_stripe_md *lsm = lli->lli_smd;
 257                 char *symlink_name = lli->lli_symlink_name;
 258
 259                 if (lsm) {
 260                         OBD_FREE(lsm, ll_ost_easize(inode->i_sb));
 261                         lli->lli_smd = NULL;
 262                 }
 263                 if (symlink_name) {
 264                         OBD_FREE(symlink_name, strlen(symlink_name) + 1);
 265                         lli->lli_symlink_name = NULL;
 266                 }
 267         }
 268 }
 269
 270 static void ll_delete_inode(struct inode *inode)
 271 {
 272         if (S_ISREG(inode->i_mode)) {
 273                 int err;
 274                 struct obdo *oa;
 275                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 276
 277                 if (!lsm)
 278                         GOTO(out, -EINVAL);
 279
 280                 if (lsm->lsm_object_id == 0) {
 281                         CERROR("This really happens\n");
 282                         /* No obdo was ever created */
 283                         GOTO(out, 0);
 284                 }
 285
 286                 oa = obdo_alloc();
 287                 if (oa == NULL)
 288                         GOTO(out, -ENOMEM);
 289
 290                 oa->o_id = lsm->lsm_object_id;
 291                 oa->o_easize = ll_mds_easize(inode->i_sb);
 292                 oa->o_mode = inode->i_mode;
 293                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLEASIZE | OBD_MD_FLMODE;
 294
 295                 err = obd_destroy(ll_i2obdconn(inode), oa, lsm);
 296                 obdo_free(oa);
 297                 CDEBUG(D_SUPER, "obd destroy of objid "LPX64" error %d\n",
 298                        lsm->lsm_object_id, err);
 299         }
 300 out:
 301         clear_inode(inode);
 302 }
 303
 304 /* like inode_setattr, but doesn't mark the inode dirty */
 305 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 306 {
 307         unsigned int ia_valid = attr->ia_valid;
 308         int error = 0;
 309
 310         if ((ia_valid & ATTR_SIZE) && trunc) {
 311                 error = vmtruncate(inode, attr->ia_size);
 312                 if (error)
 313                         goto out;
 314         } else if (ia_valid & ATTR_SIZE)
 315                 inode->i_size = attr->ia_size;
 316
 317         if (ia_valid & ATTR_UID)
 318                 inode->i_uid = attr->ia_uid;
 319         if (ia_valid & ATTR_GID)
 320                 inode->i_gid = attr->ia_gid;
 321         if (ia_valid & ATTR_ATIME)
 322                 inode->i_atime = attr->ia_atime;
 323         if (ia_valid & ATTR_MTIME)
 324                 inode->i_mtime = attr->ia_mtime;
 325         if (ia_valid & ATTR_CTIME)
 326                 inode->i_ctime = attr->ia_ctime;
 327         if (ia_valid & ATTR_MODE) {
 328                 inode->i_mode = attr->ia_mode;
 329                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 330                         inode->i_mode &= ~S_ISGID;
 331         }
 332 out:
 333         return error;
 334 }
 335
 336 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 337 {
 338         struct ptlrpc_request *request = NULL;
 339         struct ll_sb_info *sbi = ll_i2sbi(inode);
 340         int err;
 341
 342         ENTRY;
 343
 344         /* change incore inode */
 345         ll_attr2inode(inode, attr, do_trunc);
 346
 347         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 348         if (err)
 349                 CERROR("mdc_setattr fails (%d)\n", err);
 350
 351         ptlrpc_req_finished(request);
 352
 353         RETURN(err);
 354 }
 355
 356 int ll_setattr(struct dentry *de, struct iattr *attr)
 357 {
 358         int rc = inode_change_ok(de->d_inode, attr);
 359
 360         if (rc)
 361                 return rc;
 362
 363         return ll_inode_setattr(de->d_inode, attr, 1);
 364 }
 365
 366 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 367 {
 368         struct ptlrpc_request *request = NULL;
 369         struct ll_sb_info *sbi = ll_s2sbi(sb);
 370         struct obd_statfs osfs;
 371         int rc;
 372         ENTRY;
 373
 374         memset(sfs, 0, sizeof(*sfs));
 375         rc = mdc_statfs(&sbi->ll_mdc_conn, &osfs, &request);
 376         statfs_unpack(sfs, &osfs);
 377         ptlrpc_req_finished(request);
 378         if (rc)
 379                 CERROR("mdc_statfs fails: rc = %d\n", rc);
 380         else
 381                 CDEBUG(D_SUPER, "mdc_statfs shows blocks "LPU64"/"LPU64
 382                        " objects "LPU64"/"LPU64"\n",
 383                        osfs.os_bavail, osfs.os_blocks,
 384                        osfs.os_ffree, osfs.os_files);
 385
 386         /* temporary until mds_statfs returns statfs info for all OSTs */
 387         if (!rc) {
 388                 rc = obd_statfs(&sbi->ll_osc_conn, &osfs);
 389                 if (rc) {
 390                         CERROR("obd_statfs fails: rc = %d\n", rc);
 391                         GOTO(out, rc);
 392                 }
 393                 CDEBUG(D_SUPER, "obd_statfs shows blocks "LPU64"/"LPU64
 394                        " objects "LPU64"/"LPU64"\n",
 395                        osfs.os_bavail, osfs.os_blocks,
 396                        osfs.os_ffree, osfs.os_files);
 397
 398                 while (osfs.os_blocks > ~0UL) {
 399                         sfs->f_bsize <<= 1;
 400
 401                         osfs.os_blocks >>= 1;
 402                         osfs.os_bfree >>= 1;
 403                         osfs.os_bavail >>= 1;
 404                 }
 405                 sfs->f_blocks = osfs.os_blocks;
 406                 sfs->f_bfree = osfs.os_bfree;
 407                 sfs->f_bavail = osfs.os_bavail;
 408                 if (osfs.os_ffree < (__u64)sfs->f_ffree)
 409                         sfs->f_ffree = osfs.os_ffree;
 410         }
 411
 412 out:
 413         RETURN(rc);
 414 }
 415
 416 void ll_update_inode(struct inode *inode, struct mds_body *body)
 417 {
 418         if (body->valid & OBD_MD_FLID)
 419                 inode->i_ino = body->ino;
 420         if (body->valid & OBD_MD_FLATIME)
 421                 inode->i_atime = body->atime;
 422         if (body->valid & OBD_MD_FLMTIME)
 423                 inode->i_mtime = body->mtime;
 424         if (body->valid & OBD_MD_FLCTIME)
 425                 inode->i_ctime = body->ctime;
 426         if (body->valid & OBD_MD_FLMODE)
 427                 inode->i_mode = body->mode;
 428         if (body->valid & OBD_MD_FLUID)
 429                 inode->i_uid = body->uid;
 430         if (body->valid & OBD_MD_FLGID)
 431                 inode->i_gid = body->gid;
 432         if (body->valid & OBD_MD_FLFLAGS)
 433                 inode->i_flags = body->flags;
 434         if (body->valid & OBD_MD_FLNLINK)
 435                 inode->i_nlink = body->nlink;
 436         if (body->valid & OBD_MD_FLGENER)
 437                 inode->i_generation = body->generation;
 438         if (body->valid & OBD_MD_FLRDEV)
 439                 inode->i_rdev = body->extra;
 440         if (body->valid & OBD_MD_FLSIZE)
 441                 inode->i_size = body->size;
 442 }
 443
 444 static void ll_read_inode2(struct inode *inode, void *opaque)
 445 {
 446         struct ll_read_inode2_cookie *lic = opaque;
 447         struct mds_body *body = lic->lic_body;
 448         struct ll_inode_info *lli = ll_i2info(inode);
 449         ENTRY;
 450
 451         sema_init(&lli->lli_open_sem, 1);
 452
 453         /* core attributes first */
 454         ll_update_inode(inode, body);
 455
 456         //if (body->valid & OBD_MD_FLEASIZE)
 457         if (lic && lic->lic_lmm) {
 458                 struct lov_mds_md *lmm = lic->lic_lmm;
 459                 int size;
 460
 461                 if (lmm->lmm_easize != ll_mds_easize(inode->i_sb)) {
 462                         CERROR("Striping metadata size error %ld\n",
 463                                inode->i_ino);
 464                         LBUG();
 465                 }
 466                 size = ll_ost_easize(inode->i_sb);
 467                 OBD_ALLOC(lli->lli_smd, size);
 468                 if (!lli->lli_smd) {
 469                         CERROR("No memory for %d\n", size);
 470                         LBUG();
 471                 }
 472                 lov_unpackmd(lli->lli_smd, lmm);
 473         } else {
 474                 lli->lli_smd = NULL;
 475         }
 476
 477         /* Get the authoritative file size */
 478         if (lli->lli_smd && (inode->i_mode & S_IFREG)) {
 479                 int rc;
 480
 481                 rc = ll_file_size(inode, lli->lli_smd);
 482                 if (rc) {
 483                         CERROR("ll_file_size: %d\n", rc);
 484                         /* FIXME: need to somehow prevent inode creation */
 485                         LBUG();
 486                 }
 487         }
 488
 489         /* OIDEBUG(inode); */
 490
 491         if (S_ISREG(inode->i_mode)) {
 492                 inode->i_op = &ll_file_inode_operations;
 493                 inode->i_fop = &ll_file_operations;
 494                 inode->i_mapping->a_ops = &ll_aops;
 495                 EXIT;
 496         } else if (S_ISDIR(inode->i_mode)) {
 497                 inode->i_op = &ll_dir_inode_operations;
 498                 inode->i_fop = &ll_dir_operations;
 499                 inode->i_mapping->a_ops = &ll_dir_aops;
 500                 EXIT;
 501         } else if (S_ISLNK(inode->i_mode)) {
 502                 inode->i_op = &ll_fast_symlink_inode_operations;
 503                 EXIT;
 504         } else {
 505                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 506                 EXIT;
 507         }
 508 }
 509
 510 /* exported operations */
 511 struct super_operations ll_super_operations =
 512 {
 513         read_inode2: ll_read_inode2,
 514         clear_inode: ll_clear_inode,
 515         delete_inode: ll_delete_inode,
 516         put_super: ll_put_super,
 517         statfs: ll_statfs
 518 };
 519
 520 struct file_system_type lustre_lite_fs_type = {
 521         "lustre_lite", 0, ll_read_super, NULL
 522 };
 523
 524 static int __init init_lustre_lite(void)
 525 {
 526         printk(KERN_INFO "Lustre Lite 0.0.1, info@clusterfs.com\n");
 527         ll_file_data_slab = kmem_cache_create("ll_file_data",
 528                                               sizeof(struct ll_file_data), 0,
 529                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 530         if (ll_file_data_slab == NULL)
 531                 return -ENOMEM;
 532         return register_filesystem(&lustre_lite_fs_type);
 533 }
 534
 535 static void __exit exit_lustre_lite(void)
 536 {
 537         unregister_filesystem(&lustre_lite_fs_type);
 538         kmem_cache_destroy(ll_file_data_slab);
 539 }
 540
 541 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 542 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 543 MODULE_LICENSE("GPL");
 544
 545 module_init(init_lustre_lite);
 546 module_exit(exit_lustre_lite);