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