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