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