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/lustre_lite.h>
  16 #include <linux/lustre_ha.h>
  17 #include <linux/lustre_dlm.h>
  18
  19 kmem_cache_t *ll_file_data_slab;
  20 extern struct address_space_operations ll_aops;
  21 extern struct address_space_operations ll_dir_aops;
  22 struct super_operations ll_super_operations;
  23
  24 extern int ll_recover(struct ptlrpc_client *);
  25 extern int ll_commitcbd_setup(struct ll_sb_info *);
  26 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  27
  28 static char *ll_read_opt(const char *opt, char *data)
  29 {
  30         char *value;
  31         char *retval;
  32         ENTRY;
  33
  34         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
  35         if ( strncmp(opt, data, strlen(opt)) )
  36                 RETURN(NULL);
  37         if ( (value = strchr(data, '=')) == NULL )
  38                 RETURN(NULL);
  39
  40         value++;
  41         OBD_ALLOC(retval, strlen(value) + 1);
  42         if ( !retval ) {
  43                 CERROR("out of memory!\n");
  44                 RETURN(NULL);
  45         }
  46
  47         memcpy(retval, value, strlen(value)+1);
  48         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  49         RETURN(retval);
  50 }
  51
  52 static void ll_options(char *options, char **ost, char **mds)
  53 {
  54         char *this_char;
  55         ENTRY;
  56
  57         if (!options) {
  58                 EXIT;
  59                 return;
  60         }
  61
  62         for (this_char = strtok (options, ",");
  63              this_char != NULL;
  64              this_char = strtok (NULL, ",")) {
  65                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
  66                 if ( (!*ost && (*ost = ll_read_opt("ost", this_char)))||
  67                      (!*mds && (*mds = ll_read_opt("mds", this_char))) )
  68                         continue;
  69         }
  70         EXIT;
  71 }
  72
  73 #ifndef log2
  74 #define log2(n) ffz(~(n))
  75 #endif
  76
  77 static struct super_block * ll_read_super(struct super_block *sb,
  78                                           void *data, int silent)
  79 {
  80         struct inode *root = 0;
  81         struct obd_device *obd;
  82         struct ll_sb_info *sbi;
  83         char *ost = NULL;
  84         char *mds = NULL;
  85         int err;
  86         struct ll_fid rootfid;
  87         struct statfs sfs;
  88         __u64 last_committed, last_rcvd;
  89         __u32 last_xid;
  90         struct ptlrpc_request *request = NULL;
  91         struct ll_inode_md md;
  92
  93         ENTRY;
  94         MOD_INC_USE_COUNT;
  95
  96         OBD_ALLOC(sbi, sizeof(*sbi));
  97         if (!sbi) {
  98                 MOD_DEC_USE_COUNT;
  99                 RETURN(NULL);
 100         }
 101
 102         sb->u.generic_sbp = sbi;
 103
 104         ll_options(data, &ost, &mds);
 105
 106         if (!ost) {
 107                 CERROR("no ost\n");
 108                 GOTO(out_free, sb = NULL);
 109         }
 110
 111         if (!mds) {
 112                 CERROR("no mds\n");
 113                 GOTO(out_free, sb = NULL);
 114         }
 115
 116         obd = class_uuid2obd(mds);
 117         if (!obd) {
 118                 CERROR("MDS %s: not setup or attached\n", mds);
 119                 GOTO(out_free, sb = NULL);
 120         }
 121
 122         /* First the MDS since an LOV requires an the MDC connection
 123            to find its descriptor */
 124 #if 0
 125         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mds_conn);
 126         if (err) {
 127                 CERROR("cannot connect to MDS: rc = %d\n", err);
 128                 GOTO(out_rpc, sb = NULL);
 129         }
 130 #endif
 131
 132         err = obd_connect(&sbi->ll_mdc_conn, obd);
 133         if (err) {
 134                 CERROR("cannot connect to %s: rc = %d\n", mds, err);
 135                 GOTO(out_free, sb = NULL);
 136         }
 137         sbi2mdc(sbi)->mdc_conn->c_level = LUSTRE_CONN_FULL;
 138
 139         /* now the OST, which could be an LOV */
 140
 141         obd = class_uuid2obd(ost);
 142         if (!obd) {
 143                 CERROR("OST %s: not setup or attached\n", ost);
 144                 GOTO(out_free, sb = NULL);
 145         }
 146 #warning error: need to set obd->u.lov.mdc_connh from sbi->ll_mdc_conn
 147         err = obd_connect(&sbi->ll_osc_conn, obd);
 148         if (err) {
 149                 CERROR("cannot connect to %s: rc = %d\n", ost, err);
 150                 GOTO(out_free, sb = NULL);
 151         }
 152
 153         /* XXX: need to store the last_* values somewhere */
 154         err = mdc_getstatus(&sbi->ll_mdc_conn,
 155                             &rootfid, &last_committed,
 156                             &last_rcvd, &last_xid,
 157                             &request);
 158         if (err) {
 159                 CERROR("cannot mds_connect: rc = %d\n", err);
 160                 GOTO(out_disc, 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         err = mdc_statfs(&sbi->ll_mdc_conn, &sfs, &request);
 167         sb->s_blocksize = sfs.f_bsize;
 168         sb->s_blocksize_bits = log2(sfs.f_bsize);
 169         sb->s_magic = LL_SUPER_MAGIC;
 170         sb->s_maxbytes = (1ULL << (32 + 9)) - sfs.f_bsize;
 171         ptlrpc_req_finished(request);
 172
 173         sb->s_op = &ll_super_operations;
 174
 175         /* make root inode */
 176         err = mdc_getattr(&sbi->ll_mdc_conn,
 177                           sbi->ll_rootino, S_IFDIR, OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 178         if (err) {
 179                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 180                 GOTO(out_mdc, sb = NULL);
 181         }
 182
 183         /* initialize committed transaction callback daemon */
 184         spin_lock_init(&sbi->ll_commitcbd_lock);
 185         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 186         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 187         sbi->ll_commitcbd_flags = 0;
 188         err = ll_commitcbd_setup(sbi);
 189         if (err) {
 190                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 191                 GOTO(out_request, sb = NULL);
 192         }
 193
 194         md.body = lustre_msg_buf(request->rq_repmsg, 0);
 195         md.md = NULL;
 196         root = iget4(sb, sbi->ll_rootino, NULL, &md);
 197
 198         if (root) {
 199                 sb->s_root = d_alloc_root(root);
 200         } else {
 201                 CERROR("lustre_lite: bad iget4 for root\n");
 202                 GOTO(out_cdb, sb = NULL);
 203         }
 204
 205         ptlrpc_free_req(request);
 206
 207 out_dev:
 208         if (mds)
 209                 OBD_FREE(mds, strlen(mds) + 1);
 210         if (ost)
 211                 OBD_FREE(ost, strlen(ost) + 1);
 212
 213         RETURN(sb);
 214
 215 out_cdb:
 216         ll_commitcbd_cleanup(sbi);
 217 out_request:
 218         ptlrpc_free_req(request);
 219         obd_disconnect(&sbi->ll_osc_conn);
 220 out_mdc:
 221         obd_disconnect(&sbi->ll_mdc_conn);
 222 out_free:
 223         OBD_FREE(sbi, sizeof(*sbi));
 224
 225         MOD_DEC_USE_COUNT;
 226         goto out_dev;
 227 } /* ll_read_super */
 228
 229 static void ll_put_super(struct super_block *sb)
 230 {
 231         struct ll_sb_info *sbi = ll_s2sbi(sb);
 232         ENTRY;
 233         ll_commitcbd_cleanup(sbi);
 234         obd_disconnect(&sbi->ll_osc_conn);
 235         obd_disconnect(&sbi->ll_mdc_conn);
 236         OBD_FREE(sbi, sizeof(*sbi));
 237
 238         MOD_DEC_USE_COUNT;
 239         EXIT;
 240 } /* ll_put_super */
 241
 242 static void ll_clear_inode(struct inode *inode)
 243 {
 244         if (atomic_read(&inode->i_count) == 0) {
 245                 struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
 246                 if (md) {
 247                         OBD_FREE(md, md->lmd_size);
 248                         ll_i2info(inode)->lli_smd = NULL;
 249                 }
 250                 if (ll_i2info(inode)->lli_symlink_name) {
 251                         OBD_FREE(ll_i2info(inode)->lli_symlink_name,
 252                                  strlen(ll_i2info(inode)->lli_symlink_name)+ 1);
 253                         ll_i2info(inode)->lli_symlink_name = NULL;
 254                 }
 255         }
 256 }
 257
 258 static void ll_delete_inode(struct inode *inode)
 259 {
 260         if (S_ISREG(inode->i_mode)) {
 261                 int err;
 262                 struct obdo oa;
 263                 struct lov_stripe_md *md = ll_i2info(inode)->lli_smd;
 264
 265                if (!md)
 266                         GOTO(out, -EINVAL);
 267
 268                 oa.o_id = md->lmd_object_id;
 269                 oa.o_easize = md->lmd_size;
 270                 if (oa.o_id == 0) {
 271                         CERROR("This really happens\n");
 272                         /* No obdo was ever created */
 273                         GOTO(out, 0);
 274                 }
 275
 276                 err = obd_destroy(ll_i2obdconn(inode), &oa, md);
 277                 CDEBUG(D_SUPER, "obd destroy of %Ld error %d\n",
 278                        md->lmd_object_id, err);
 279         }
 280 out:
 281         clear_inode(inode);
 282 }
 283
 284 /* like inode_setattr, but doesn't mark the inode dirty */
 285 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 286 {
 287         unsigned int ia_valid = attr->ia_valid;
 288         int error = 0;
 289
 290         if ((ia_valid & ATTR_SIZE) && trunc) {
 291                 error = vmtruncate(inode, attr->ia_size);
 292                 if (error)
 293                         goto out;
 294         } else if (ia_valid & ATTR_SIZE)
 295                 inode->i_size = attr->ia_size;
 296
 297         if (ia_valid & ATTR_UID)
 298                 inode->i_uid = attr->ia_uid;
 299         if (ia_valid & ATTR_GID)
 300                 inode->i_gid = attr->ia_gid;
 301         if (ia_valid & ATTR_ATIME)
 302                 inode->i_atime = attr->ia_atime;
 303         if (ia_valid & ATTR_MTIME)
 304                 inode->i_mtime = attr->ia_mtime;
 305         if (ia_valid & ATTR_CTIME)
 306                 inode->i_ctime = attr->ia_ctime;
 307         if (ia_valid & ATTR_MODE) {
 308                 inode->i_mode = attr->ia_mode;
 309                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 310                         inode->i_mode &= ~S_ISGID;
 311         }
 312 out:
 313         return error;
 314 }
 315
 316 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 317 {
 318         struct ptlrpc_request *request = NULL;
 319         struct ll_sb_info *sbi = ll_i2sbi(inode);
 320         int err;
 321
 322         ENTRY;
 323
 324         /* change incore inode */
 325         ll_attr2inode(inode, attr, do_trunc);
 326
 327         err = mdc_setattr(&sbi->ll_mdc_conn, inode, attr, &request);
 328         if (err)
 329                 CERROR("mdc_setattr fails (%d)\n", err);
 330
 331         ptlrpc_req_finished(request);
 332
 333         RETURN(err);
 334 }
 335
 336 int ll_setattr(struct dentry *de, struct iattr *attr)
 337 {
 338         int rc = inode_change_ok(de->d_inode, attr);
 339
 340         if (rc)
 341                 return rc;
 342
 343         return ll_inode_setattr(de->d_inode, attr, 1);
 344 }
 345
 346 static int ll_statfs(struct super_block *sb, struct statfs *sfs)
 347 {
 348         struct ptlrpc_request *request = NULL;
 349         struct ll_sb_info *sbi = ll_s2sbi(sb);
 350         int rc;
 351         ENTRY;
 352
 353         memset(sfs, 0, sizeof(*sfs));
 354         rc = mdc_statfs(&sbi->ll_mdc_conn, sfs, &request);
 355         ptlrpc_req_finished(request);
 356         if (rc)
 357                 CERROR("obd_statfs fails: rc = %d\n", rc);
 358         else
 359                 CDEBUG(D_SUPER, "statfs shows blocks %ld/%ld objects %ld/%ld\n",
 360                        sfs->f_bavail, sfs->f_blocks, sfs->f_files,sfs->f_ffree);
 361
 362         /* temporary until mds_statfs returns statfs info for all OSTs */
 363         if (!rc) {
 364                 struct statfs obd_sfs;
 365
 366                 rc = obd_statfs(&sbi->ll_osc_conn, &obd_sfs);
 367                 if (rc) {
 368                         CERROR("obd_statfs fails: rc = %d\n", rc);
 369                         GOTO(out, rc);
 370                 }
 371                 CDEBUG(D_SUPER, "obd_statfs returns blocks %ld/%ld, "
 372                        "objects %ld/%ld\n",
 373                        obd_sfs.f_bavail, obd_sfs.f_blocks,
 374                        obd_sfs.f_ffree, obd_sfs.f_files);
 375
 376                 sfs->f_bfree = obd_sfs.f_bfree;
 377                 sfs->f_bavail = obd_sfs.f_bavail;
 378                 sfs->f_blocks = obd_sfs.f_blocks;
 379                 if (obd_sfs.f_ffree < sfs->f_ffree)
 380                         sfs->f_ffree = obd_sfs.f_ffree;
 381         }
 382
 383 out:
 384         RETURN(rc);
 385 }
 386
 387 inline int ll_stripe_md_size(struct super_block *sb)
 388 {
 389         struct mdc_obd *mdc = sbi2mdc(ll_s2sbi(sb));
 390         return mdc->mdc_max_mdsize;
 391 }
 392
 393 static void ll_to_inode(struct inode *dst, struct ll_inode_md *md)
 394 {
 395         struct mds_body *body = md->body;
 396         struct ll_inode_info *ii = ll_i2info(dst);
 397
 398         /* core attributes first */
 399         if (body->valid & OBD_MD_FLID)
 400                 dst->i_ino = body->ino;
 401         if (body->valid & OBD_MD_FLATIME)
 402                 dst->i_atime = body->atime;
 403         if (body->valid & OBD_MD_FLMTIME)
 404                 dst->i_mtime = body->mtime;
 405         if (body->valid & OBD_MD_FLCTIME)
 406                 dst->i_ctime = body->ctime;
 407         if (body->valid & OBD_MD_FLSIZE)
 408                 dst->i_size = body->size;
 409         if (body->valid & OBD_MD_FLMODE)
 410                 dst->i_mode = body->mode;
 411         if (body->valid & OBD_MD_FLUID)
 412                 dst->i_uid = body->uid;
 413         if (body->valid & OBD_MD_FLGID)
 414                 dst->i_gid = body->gid;
 415         if (body->valid & OBD_MD_FLFLAGS)
 416                 dst->i_flags = body->flags;
 417         if (body->valid & OBD_MD_FLNLINK)
 418                 dst->i_nlink = body->nlink;
 419         if (body->valid & OBD_MD_FLGENER)
 420                 dst->i_generation = body->generation;
 421         if (body->valid & OBD_MD_FLRDEV)
 422                 dst->i_rdev = body->extra;
 423         if (md && md->md && md->md->lmd_stripe_count) {
 424                 struct lov_stripe_md *smd = md->md;
 425                 int size = ll_stripe_md_size(dst->i_sb);
 426                 if (md->md->lmd_size != size) {
 427                         CERROR("Striping metadata size error %ld\n",
 428                                dst->i_ino);
 429                         LBUG();
 430                 }
 431                 OBD_ALLOC(ii->lli_smd, size);
 432                 if (!ii->lli_smd){
 433                         CERROR("No memory for %d\n", size);
 434                         LBUG();
 435                 }
 436                 memcpy(ii->lli_smd, smd, size);
 437         }
 438 } /* ll_to_inode */
 439
 440 static void ll_read_inode2(struct inode *inode, void *opaque)
 441 {
 442         struct ll_inode_md *md = opaque;
 443
 444         ENTRY;
 445         ll_to_inode(inode, md);
 446
 447         /* OIDEBUG(inode); */
 448
 449         if (S_ISREG(inode->i_mode)) {
 450                 inode->i_op = &ll_file_inode_operations;
 451                 inode->i_fop = &ll_file_operations;
 452                 inode->i_mapping->a_ops = &ll_aops;
 453                 EXIT;
 454         } else if (S_ISDIR(inode->i_mode)) {
 455                 inode->i_op = &ll_dir_inode_operations;
 456                 inode->i_fop = &ll_dir_operations;
 457                 inode->i_mapping->a_ops = &ll_dir_aops;
 458                 EXIT;
 459         } else if (S_ISLNK(inode->i_mode)) {
 460                 inode->i_op = &ll_fast_symlink_inode_operations;
 461                 EXIT;
 462         } else {
 463                 init_special_inode(inode, inode->i_mode,
 464                                    ((int *)ll_i2info(inode)->lli_inline)[0]);
 465                 EXIT;
 466         }
 467
 468         return;
 469 }
 470
 471 /* exported operations */
 472 struct super_operations ll_super_operations =
 473 {
 474         read_inode2: ll_read_inode2,
 475         clear_inode: ll_clear_inode,
 476         delete_inode: ll_delete_inode,
 477         put_super: ll_put_super,
 478         statfs: ll_statfs
 479 };
 480
 481 struct file_system_type lustre_lite_fs_type = {
 482         "lustre_lite", 0, ll_read_super, NULL
 483 };
 484
 485 static int __init init_lustre_lite(void)
 486 {
 487         printk(KERN_INFO "Lustre Lite 0.0.1, info@clusterfs.com\n");
 488         ll_file_data_slab = kmem_cache_create("ll_file_data",
 489                                               sizeof(struct ll_file_data), 0,
 490                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 491         if (ll_file_data_slab == NULL)
 492                 return -ENOMEM;
 493         return register_filesystem(&lustre_lite_fs_type);
 494 }
 495
 496 static void __exit exit_lustre_lite(void)
 497 {
 498         unregister_filesystem(&lustre_lite_fs_type);
 499         kmem_cache_destroy(ll_file_data_slab);
 500 }
 501
 502 MODULE_AUTHOR("Cluster File Systems, Inc. <info@clusterfs.com>");
 503 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 504 MODULE_LICENSE("GPL");
 505
 506 module_init(init_lustre_lite);
 507 module_exit(exit_lustre_lite);