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