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) 1996 Peter J. Braam <braam@stelias.com>
  10  * Copryright (C) 1999 Stelias Computing Inc. <braam@stelias.com>
  11  * Copryright (C) 1999 Seagate Technology Inc.
  12  * Copryright (C) 2001 Mountain View Data, Inc.
  13  * Copryright (C) 2002 Cluster File Systems, Inc.
  14  *
  15  */
  16
  17 #define DEBUG_SUBSYSTEM S_LLITE
  18
  19 #include <linux/module.h>
  20 #include <linux/lustre_lite.h>
  21 #include <linux/lustre_ha.h>
  22 #include <linux/lustre_dlm.h>
  23
  24 kmem_cache_t *ll_file_data_slab;
  25 extern struct address_space_operations ll_aops;
  26 extern struct address_space_operations ll_dir_aops;
  27 struct super_operations ll_super_operations;
  28
  29 extern int ll_recover(struct ptlrpc_client *);
  30 extern int ll_commitcbd_setup(struct ll_sb_info *);
  31 extern int ll_commitcbd_cleanup(struct ll_sb_info *);
  32
  33 static char *ll_read_opt(const char *opt, char *data)
  34 {
  35         char *value;
  36         char *retval;
  37         ENTRY;
  38
  39         CDEBUG(D_INFO, "option: %s, data %s\n", opt, data);
  40         if ( strncmp(opt, data, strlen(opt)) )
  41                 RETURN(NULL);
  42         if ( (value = strchr(data, '=')) == NULL )
  43                 RETURN(NULL);
  44
  45         value++;
  46         OBD_ALLOC(retval, strlen(value) + 1);
  47         if ( !retval ) {
  48                 CERROR("out of memory!\n");
  49                 RETURN(NULL);
  50         }
  51
  52         memcpy(retval, value, strlen(value)+1);
  53         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
  54         RETURN(retval);
  55 }
  56
  57 static void ll_options(char *options, char **dev, char **vers)
  58 {
  59         char *this_char;
  60         ENTRY;
  61
  62         if (!options) {
  63                 EXIT;
  64                 return;
  65         }
  66
  67         for (this_char = strtok (options, ",");
  68              this_char != NULL;
  69              this_char = strtok (NULL, ",")) {
  70                 CDEBUG(D_INFO, "this_char %s\n", this_char);
  71                 if ( (!*dev && (*dev = ll_read_opt("device", this_char)))||
  72                      (!*vers && (*vers = ll_read_opt("version", this_char))) )
  73                         continue;
  74         }
  75         EXIT;
  76 }
  77
  78 static struct super_block * ll_read_super(struct super_block *sb,
  79                                           void *data, int silent)
  80 {
  81         struct inode *root = 0;
  82         struct ll_sb_info *sbi;
  83         char *device = NULL;
  84         char *version = NULL;
  85         int devno;
  86         int err;
  87         struct ll_fid rootfid;
  88         __u64 last_committed, last_rcvd;
  89         __u32 last_xid;
  90         struct ptlrpc_request *request = NULL;
  91
  92         ENTRY;
  93         MOD_INC_USE_COUNT;
  94
  95         OBD_ALLOC(sbi, sizeof(*sbi));
  96         if (!sbi) {
  97                 MOD_DEC_USE_COUNT;
  98                 RETURN(NULL);
  99         }
 100
 101         sb->u.generic_sbp = sbi;
 102
 103         ll_options(data, &device, &version);
 104
 105         if (!device) {
 106                 CERROR("no device\n");
 107                 GOTO(out_free, sb = NULL);
 108         }
 109
 110         devno = simple_strtoul(device, NULL, 0);
 111         if (devno >= MAX_OBD_DEVICES) {
 112                 CERROR("device of %s too high\n", device);
 113                 GOTO(out_free, sb = NULL);
 114         }
 115
 116         sbi->ll_conn.oc_dev = &obd_dev[devno];
 117         err = obd_connect(&sbi->ll_conn);
 118         if (err) {
 119                 CERROR("cannot connect to %s: rc = %d\n", device, err);
 120                 GOTO(out_free, sb = NULL);
 121         }
 122
 123         sbi->ll_namespace = ldlm_namespace_new(NULL, 1);
 124         if (sbi->ll_namespace == NULL) {
 125                 CERROR("failed to create local lock namespace\n");
 126                 GOTO(out_free, sb = NULL);
 127         }
 128
 129         ptlrpc_init_client(ptlrpc_connmgr, ll_recover,
 130                            MDS_REQUEST_PORTAL, MDC_REPLY_PORTAL,
 131                            &sbi->ll_mds_client);
 132
 133         sbi->ll_mds_client.cli_data = sbi;
 134         sbi->ll_mds_client.cli_name = "mdc";
 135         sbi->ll_mds_conn = ptlrpc_uuid_to_connection("mds");
 136         if (!sbi->ll_mds_conn) {
 137                 CERROR("cannot find MDS\n");
 138                 GOTO(out_disc, sb = NULL);
 139         }
 140
 141         err = connmgr_connect(ptlrpc_connmgr, sbi->ll_mds_conn);
 142         if (err) {
 143                 CERROR("cannot connect to MDS: rc = %d\n", err);
 144                 ptlrpc_put_connection(sbi->ll_mds_conn);
 145                 GOTO(out_disc, sb = NULL);
 146         }
 147
 148         sbi->ll_mds_conn->c_level = LUSTRE_CONN_FULL;
 149
 150         /* XXX: need to store the last_* values somewhere */
 151         err = mdc_connect(&sbi->ll_mds_client, sbi->ll_mds_conn,
 152                           &rootfid, &last_committed, &last_rcvd, &last_xid,
 153                           &request);
 154         if (err) {
 155                 CERROR("cannot mds_connect: rc = %d\n", err);
 156                 GOTO(out_disc, sb = NULL);
 157         }
 158         CERROR("rootfid %ld\n", (unsigned long)rootfid.id);
 159         sbi->ll_rootino = rootfid.id;
 160
 161         sb->s_maxbytes = 1ULL << 36;
 162         sb->s_blocksize = PAGE_SIZE;
 163         sb->s_blocksize_bits = (unsigned char)PAGE_SHIFT;
 164         sb->s_magic = LL_SUPER_MAGIC;
 165         sb->s_op = &ll_super_operations;
 166
 167         /* make root inode */
 168         err = mdc_getattr(&sbi->ll_mds_client, sbi->ll_mds_conn,
 169                           sbi->ll_rootino, S_IFDIR,
 170                           OBD_MD_FLNOTOBD|OBD_MD_FLBLOCKS, 0, &request);
 171         if (err) {
 172                 CERROR("mdc_getattr failed for root: rc = %d\n", err);
 173                 GOTO(out_req, sb = NULL);
 174         }
 175
 176         /* initialize committed transaction callback daemon */
 177         spin_lock_init(&sbi->ll_commitcbd_lock);
 178         init_waitqueue_head(&sbi->ll_commitcbd_waitq);
 179         init_waitqueue_head(&sbi->ll_commitcbd_ctl_waitq);
 180         sbi->ll_commitcbd_flags = 0;
 181         err = ll_commitcbd_setup(sbi);
 182         if (err) {
 183                 CERROR("failed to start commit callback daemon: rc = %d\n",err);
 184                 GOTO(out_req, sb = NULL);
 185         }
 186
 187         root = iget4(sb, sbi->ll_rootino, NULL,
 188                      lustre_msg_buf(request->rq_repmsg, 0));
 189         if (root) {
 190                 sb->s_root = d_alloc_root(root);
 191         } else {
 192                 CERROR("lustre_lite: bad iget4 for root\n");
 193                 GOTO(out_req, sb = NULL);
 194         }
 195
 196 out_req:
 197         ptlrpc_free_req(request);
 198         if (!sb) {
 199 out_disc:
 200                 obd_disconnect(&sbi->ll_conn);
 201 out_free:
 202                 MOD_DEC_USE_COUNT;
 203                 if (sbi->ll_namespace)
 204                         ldlm_namespace_free(sbi->ll_namespace);
 205                 OBD_FREE(sbi, sizeof(*sbi));
 206         }
 207         if (device)
 208                 OBD_FREE(device, strlen(device) + 1);
 209         if (version)
 210                 OBD_FREE(version, strlen(version) + 1);
 211
 212         RETURN(sb);
 213 } /* ll_read_super */
 214
 215 static void ll_put_super(struct super_block *sb)
 216 {
 217         struct ll_sb_info *sbi = sb->u.generic_sbp;
 218         ENTRY;
 219         ll_commitcbd_cleanup(sbi);
 220         obd_disconnect(&sbi->ll_conn);
 221         ldlm_namespace_free(sbi->ll_namespace);
 222         ptlrpc_put_connection(sbi->ll_mds_conn);
 223         ptlrpc_cleanup_client(&sbi->ll_mds_client);
 224         OBD_FREE(sb->u.generic_sbp, sizeof(*sbi));
 225         MOD_DEC_USE_COUNT;
 226         EXIT;
 227 } /* ll_put_super */
 228
 229
 230 extern inline struct obdo * ll_oa_from_inode(struct inode *inode, int valid);
 231 static void ll_delete_inode(struct inode *inode)
 232 {
 233         if (S_ISREG(inode->i_mode)) {
 234                 int err;
 235                 struct obdo *oa;
 236                 oa = ll_oa_from_inode(inode, OBD_MD_FLNOTOBD);
 237                 if (!oa) {
 238                         CERROR("no memory\n");
 239                 }
 240
 241                 err = obd_destroy(ll_i2obdconn(inode), oa);
 242                 CDEBUG(D_INODE, "obd destroy of %Ld error %d\n",
 243                        (unsigned long long)oa->o_id, err);
 244                 obdo_free(oa);
 245         }
 246
 247         clear_inode(inode);
 248 }
 249
 250 /* like inode_setattr, but doesn't mark the inode dirty */
 251 static int ll_attr2inode(struct inode * inode, struct iattr * attr, int trunc)
 252 {
 253         unsigned int ia_valid = attr->ia_valid;
 254         int error = 0;
 255
 256         if ((ia_valid & ATTR_SIZE) && trunc ) {
 257                 error = vmtruncate(inode, attr->ia_size);
 258                 if (error)
 259                         goto out;
 260         } else if (ia_valid & ATTR_SIZE) {
 261                 inode->i_size = attr->ia_size;
 262         }
 263
 264         if (ia_valid & ATTR_UID)
 265                 inode->i_uid = attr->ia_uid;
 266         if (ia_valid & ATTR_GID)
 267                 inode->i_gid = attr->ia_gid;
 268         if (ia_valid & ATTR_ATIME)
 269                 inode->i_atime = attr->ia_atime;
 270         if (ia_valid & ATTR_MTIME)
 271                 inode->i_mtime = attr->ia_mtime;
 272         if (ia_valid & ATTR_CTIME)
 273                 inode->i_ctime = attr->ia_ctime;
 274         if (ia_valid & ATTR_MODE) {
 275                 inode->i_mode = attr->ia_mode;
 276                 if (!in_group_p(inode->i_gid) && !capable(CAP_FSETID))
 277                         inode->i_mode &= ~S_ISGID;
 278         }
 279 out:
 280         return error;
 281 }
 282
 283 int ll_inode_setattr(struct inode *inode, struct iattr *attr, int do_trunc)
 284 {
 285         struct ptlrpc_request *request = NULL;
 286         struct ll_sb_info *sbi = ll_i2sbi(inode);
 287         int err;
 288
 289         ENTRY;
 290
 291         /* change incore inode */
 292         ll_attr2inode(inode, attr, do_trunc);
 293
 294         err = mdc_setattr(&sbi->ll_mds_client, sbi->ll_mds_conn, inode, attr,
 295                           &request);
 296         if (err)
 297                 CERROR("mdc_setattr fails (%d)\n", err);
 298
 299         ptlrpc_req_finished(request);
 300
 301         RETURN(err);
 302 }
 303
 304 int ll_setattr(struct dentry *de, struct iattr *attr)
 305 {
 306         return ll_inode_setattr(de->d_inode, attr, 1);
 307 }
 308
 309 static int ll_statfs(struct super_block *sb, struct statfs *buf)
 310 {
 311         struct statfs tmp;
 312         int err;
 313         ENTRY;
 314
 315         err = obd_statfs(&ll_s2sbi(sb)->ll_conn, &tmp);
 316         if (err) {
 317                 CERROR("obd_statfs fails (%d)\n", err);
 318                 RETURN(err);
 319         }
 320         memcpy(buf, &tmp, sizeof(*buf));
 321         CDEBUG(D_SUPER, "statfs returns avail %ld\n", tmp.f_bavail);
 322
 323         RETURN(err);
 324 }
 325
 326 static void inline ll_to_inode(struct inode *dst, struct mds_body *body)
 327 {
 328         struct ll_inode_info *ii =
 329                 (struct ll_inode_info *) &dst->u.generic_ip;
 330
 331         /* core attributes first */
 332         if ( body->valid & OBD_MD_FLID )
 333                 dst->i_ino = body->ino;
 334         if ( body->valid & OBD_MD_FLATIME )
 335                 dst->i_atime = body->atime;
 336         if ( body->valid & OBD_MD_FLMTIME )
 337                 dst->i_mtime = body->mtime;
 338         if ( body->valid & OBD_MD_FLCTIME )
 339                 dst->i_ctime = body->ctime;
 340         if ( body->valid & OBD_MD_FLSIZE )
 341                 dst->i_size = body->size;
 342         if ( body->valid & OBD_MD_FLMODE )
 343                 dst->i_mode = body->mode;
 344         if ( body->valid & OBD_MD_FLUID )
 345                 dst->i_uid = body->uid;
 346         if ( body->valid & OBD_MD_FLGID )
 347                 dst->i_gid = body->gid;
 348         if ( body->valid & OBD_MD_FLFLAGS )
 349                 dst->i_flags = body->flags;
 350         if ( body->valid & OBD_MD_FLNLINK )
 351                 dst->i_nlink = body->nlink;
 352         if ( body->valid & OBD_MD_FLGENER )
 353                 dst->i_generation = body->generation;
 354
 355         /* this will become more elaborate for striping etc */
 356         if (body->valid & OBD_MD_FLOBJID)
 357                 ii->lli_objid = body->objid;
 358 #if 0
 359
 360         if (obdo_has_inline(oa)) {
 361                 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 362                     S_ISFIFO(inode->i_mode)) {
 363                         obd_rdev rdev = *((obd_rdev *)oa->o_inline);
 364                         CDEBUG(D_INODE,
 365                                "copying device %x from obdo to inode\n", rdev);
 366                         init_special_inode(inode, inode->i_mode, rdev);
 367                 } else {
 368                         CDEBUG(D_INFO, "copying inline from obdo to inode\n");
 369                         memcpy(oinfo->lli_inline, oa->o_inline, OBD_INLINESZ);
 370                 }
 371                 oinfo->lli_flags |= OBD_FL_INLINEDATA;
 372         }
 373 #endif
 374 } /* ll_to_inode */
 375
 376 static inline void ll_read_inode2(struct inode *inode, void *opaque)
 377 {
 378         struct mds_body *body = opaque;
 379
 380         ENTRY;
 381         ll_to_inode(inode, body);
 382
 383         /* OIDEBUG(inode); */
 384
 385         if (S_ISREG(inode->i_mode)) {
 386                 inode->i_op = &ll_file_inode_operations;
 387                 inode->i_fop = &ll_file_operations;
 388                 inode->i_mapping->a_ops = &ll_aops;
 389                 EXIT;
 390         } else if (S_ISDIR(inode->i_mode)) {
 391                 inode->i_op = &ll_dir_inode_operations;
 392                 inode->i_fop = &ll_dir_operations;
 393                 inode->i_mapping->a_ops = &ll_dir_aops;
 394                 EXIT;
 395         } else if (S_ISLNK(inode->i_mode)) {
 396                 inode->i_op = &ll_fast_symlink_inode_operations;
 397                 EXIT;
 398         } else {
 399                 init_special_inode(inode, inode->i_mode,
 400                                    ((int *)ll_i2info(inode)->lli_inline)[0]);
 401                 EXIT;
 402         }
 403
 404         return;
 405 }
 406
 407 /* exported operations */
 408 struct super_operations ll_super_operations =
 409 {
 410         read_inode2: ll_read_inode2,
 411         delete_inode: ll_delete_inode,
 412         put_super: ll_put_super,
 413         statfs: ll_statfs
 414 };
 415
 416 struct file_system_type lustre_lite_fs_type = {
 417         "lustre_lite", 0, ll_read_super, NULL
 418 };
 419
 420 static int __init init_lustre_lite(void)
 421 {
 422         printk(KERN_INFO "Lustre Lite 0.0.1, braam@clusterfs.com\n");
 423         ll_file_data_slab = kmem_cache_create("ll_file_data",
 424                                               sizeof(struct ll_file_data), 0,
 425                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
 426         if (ll_file_data_slab == NULL)
 427                 return -ENOMEM;
 428         return register_filesystem(&lustre_lite_fs_type);
 429 }
 430
 431 static void __exit exit_lustre_lite(void)
 432 {
 433         unregister_filesystem(&lustre_lite_fs_type);
 434         kmem_cache_destroy(ll_file_data_slab);
 435 }
 436
 437 MODULE_AUTHOR("Peter J. Braam <braam@clusterfs.com>");
 438 MODULE_DESCRIPTION("Lustre Lite Client File System v1.0");
 439 MODULE_LICENSE("GPL");
 440
 441 module_init(init_lustre_lite);
 442 module_exit(exit_lustre_lite);