lustre/llite/llite_lib.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  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/random.h>
  29 #include <linux/version.h>
  30
  31 #include <linux/lustre_lite.h>
  32 #include <linux/lustre_ha.h>
  33 #include <linux/lustre_dlm.h>
  34 #include <linux/lprocfs_status.h>
  35 #include <linux/lustre_acl.h>
  36 #include "llite_internal.h"
  37
  38 kmem_cache_t *ll_file_data_slab;
  39 kmem_cache_t *ll_intent_slab;
  40
  41 extern struct address_space_operations ll_aops;
  42 extern struct address_space_operations ll_dir_aops;
  43
  44 #ifndef log2
  45 #define log2(n) ffz(~(n))
  46 #endif
  47
  48 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
  49 {
  50         struct ll_sb_info *sbi = NULL;
  51         class_uuid_t uuid;
  52         ENTRY;
  53
  54         OBD_ALLOC(sbi, sizeof(*sbi));
  55         if (!sbi)
  56                 RETURN(NULL);
  57
  58         spin_lock_init(&sbi->ll_lock);
  59         INIT_LIST_HEAD(&sbi->ll_pglist);
  60         sbi->ll_pglist_gen = 0;
  61         if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
  62                 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
  63         else
  64                 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
  65         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  66         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  67         INIT_LIST_HEAD(&sbi->ll_mnt_list);
  68
  69         sema_init(&sbi->ll_gns_sem, 1);
  70         spin_lock_init(&sbi->ll_gns_lock);
  71         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
  72         init_waitqueue_head(&sbi->ll_gns_waitq);
  73
  74         /* this later may be reset via /proc/fs/... */
  75         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
  76         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
  77
  78         /* this later may be reset via /proc/fs/... */
  79         memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
  80                strlen("/usr/sbin/gns_upcall"));
  81         sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
  82
  83         /* default values, may be changed via /proc/fs/... */
  84         sbi->ll_gns_state = LL_GNS_IDLE;
  85         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  86         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  87
  88         sbi->ll_gns_timer.data = (unsigned long)sbi;
  89         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  90         init_timer(&sbi->ll_gns_timer);
  91
  92         ll_set_sbi(sb, sbi);
  93
  94         generate_random_uuid(uuid);
  95         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  96         RETURN(sbi);
  97 }
  98
  99 void lustre_free_sbi(struct super_block *sb)
 100 {
 101         struct ll_sb_info *sbi = ll_s2sbi(sb);
 102         ENTRY;
 103
 104         if (sbi != NULL) {
 105                 list_del(&sbi->ll_gns_sbi_head);
 106                 del_timer(&sbi->ll_gns_timer);
 107                 OBD_FREE(sbi, sizeof(*sbi));
 108         }
 109         ll_set_sbi(sb, NULL);
 110         EXIT;
 111 }
 112
 113 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 114 {
 115         int valsize, rc;
 116         ENTRY;
 117
 118         valsize = sizeof(sbi->ll_dt_desc);
 119         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 120         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 121                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 122         RETURN(rc);
 123 }
 124
 125 extern struct dentry_operations ll_d_ops;
 126
 127 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 128                              int async, char *security, __u32 *nllu,
 129                              __u64 *remote)
 130 {
 131         struct ll_sb_info *sbi = ll_s2sbi(sb);
 132         struct ptlrpc_request *request = NULL;
 133         struct lustre_handle dt_conn = {0, };
 134         struct lustre_handle md_conn = {0, };
 135         struct obd_connect_data *data;
 136         struct inode *root = NULL;
 137         struct obd_device *obd;
 138         struct obd_statfs osfs;
 139         struct lustre_md md;
 140         kdev_t devno;
 141         int err;
 142         __u32 valsize;
 143         ENTRY;
 144
 145         obd = class_name2obd(lmv);
 146         if (!obd) {
 147                 CERROR("MDC %s: not setup or attached\n", lmv);
 148                 RETURN(-EINVAL);
 149         }
 150         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 151                      sizeof(async), &async);
 152
 153         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 154             (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 155                 CERROR("wrong remote flag "LPX64"\n", *remote);
 156                 RETURN(-EINVAL);
 157         }
 158
 159         OBD_ALLOC(data, sizeof(*data));
 160         if (!data)
 161                 RETURN(-ENOMEM);
 162
 163         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 164                                               OBD_CONNECT_REMOTE);
 165         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 166
 167         if (security == NULL)
 168                 security = "null";
 169
 170         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 171                            strlen(security), security);
 172         if (err) {
 173                 CERROR("LMV %s: failed to set security %s, err %d\n",
 174                         lmv, security, err);
 175                 OBD_FREE(data, sizeof(*data));
 176                 RETURN(err);
 177         }
 178
 179         if (proc_lustre_fs_root) {
 180                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 181                                                   lov, lmv);
 182                 if (err < 0)
 183                         CERROR("could not register mount in /proc/lustre");
 184         }
 185
 186         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
 187                           OBD_OPT_REAL_CLIENT);
 188         if (err == -EBUSY) {
 189                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 190                        " client is not a part.  Please wait for recovery to "
 191                        "complete, abort, or time out.\n", lmv);
 192                 GOTO(out, err);
 193         } else if (err) {
 194                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 195                 GOTO(out, err);
 196         }
 197         sbi->ll_md_exp = class_conn2export(&md_conn);
 198         err = obd_statfs(obd, &osfs, jiffies - HZ);
 199         if (err)
 200                 GOTO(out_lmv, err);
 201
 202         if (!osfs.os_bsize) {
 203                 CERROR("Invalid block size is detected.");
 204                 GOTO(out_lmv, err);
 205         }
 206
 207         sb->s_magic = LL_SUPER_MAGIC;
 208         sb->s_blocksize = osfs.os_bsize;
 209         sb->s_blocksize_bits = log2(osfs.os_bsize);
 210         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 211
 212         devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 213                              strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 214
 215         sb->s_dev = devno;
 216
 217         /* after statfs, we are supposed to have connected to MDSs,
 218          * so it's ok to check remote flag returned.
 219          */
 220         valsize = sizeof(&sbi->ll_remote);
 221         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 222                            &valsize, &sbi->ll_remote);
 223         if (err) {
 224                 CERROR("fail to obtain remote flag\n");
 225                 GOTO(out, err);
 226         }
 227
 228         obd = class_name2obd(lov);
 229         if (!obd) {
 230                 CERROR("OSC %s: not setup or attached\n", lov);
 231                 GOTO(out_lmv, err);
 232         }
 233         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 234                      sizeof(async), &async);
 235
 236         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
 237                           OBD_OPT_REAL_CLIENT);
 238         if (err == -EBUSY) {
 239                 CERROR("An OST (lov %s) is performing recovery, of which this"
 240                        " client is not a part.  Please wait for recovery to "
 241                        "complete, abort, or time out.\n", lov);
 242                 GOTO(out, err);
 243         } else if (err) {
 244                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 245                 GOTO(out_lmv, err);
 246         }
 247         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 248
 249         err = lustre_init_dt_desc(sbi);
 250         if (err == 0) {
 251                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 252                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
 253                                  sbi->ll_dt_desc.ld_tgt_count *
 254                                  sizeof(struct llog_cookie));
 255         }
 256
 257         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 258         if (err) {
 259                 CERROR("cannot mds_connect: rc = %d\n", err);
 260                 GOTO(out_lov, err);
 261         }
 262         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 263
 264         sb->s_op = &lustre_super_operations;
 265
 266         /* make root inode
 267          * XXX: move this to after cbd setup? */
 268         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 269                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
 270                          0, &request);
 271         if (err) {
 272                 CERROR("md_getattr failed for root: rc = %d\n", err);
 273                 GOTO(out_lov, err);
 274         }
 275
 276         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 277                                 sbi->ll_dt_exp, &md);
 278         if (err) {
 279                 CERROR("failed to understand root inode md: rc = %d\n", err);
 280                 ptlrpc_req_finished(request);
 281                 GOTO(out_lov, err);
 282         }
 283
 284         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 285         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 286
 287         ptlrpc_req_finished(request);
 288
 289         if (root == NULL || is_bad_inode(root)) {
 290                 CERROR("lustre_lite: bad iget4 for root\n");
 291                 GOTO(out_root, err = -EBADF);
 292         }
 293
 294         err = ll_close_thread_start(&sbi->ll_lcq);
 295         if (err) {
 296                 CERROR("cannot start close thread: rc %d\n", err);
 297                 GOTO(out_root, err);
 298         }
 299
 300         ll_gns_add_timer(sbi);
 301
 302         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 303            backing dev info assigned to inode mapping is used for
 304            determining maximal readahead. */
 305 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 306     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 307         /* bug 2805 - set VM readahead to zero */
 308         vm_max_readahead = vm_min_readahead = 0;
 309 #endif
 310
 311         sb->s_root = d_alloc_root(root);
 312         sb->s_root->d_op = &ll_d_ops;
 313
 314         sb->s_flags |= MS_POSIXACL;
 315 #ifdef S_PDIROPS
 316         CWARN("Enabling PDIROPS\n");
 317         sb->s_flags |= S_PDIROPS;
 318 #endif
 319
 320         if (data != NULL)
 321                 OBD_FREE(data, sizeof(*data));
 322         RETURN(err);
 323 out_root:
 324         if (root)
 325                 iput(root);
 326 out_lov:
 327         obd_disconnect(sbi->ll_dt_exp, 0);
 328 out_lmv:
 329         obd_disconnect(sbi->ll_md_exp, 0);
 330 out:
 331         if (data != NULL)
 332                 OBD_FREE(data, sizeof(*data));
 333         lprocfs_unregister_mountpoint(sbi);
 334         return err;
 335 }
 336
 337 void lustre_common_put_super(struct super_block *sb)
 338 {
 339         struct ll_sb_info *sbi = ll_s2sbi(sb);
 340         struct hlist_node *tmp, *next;
 341         ENTRY;
 342
 343         ll_gns_del_timer(sbi);
 344         ll_close_thread_shutdown(sbi->ll_lcq);
 345
 346         list_del(&sbi->ll_conn_chain);
 347         obd_disconnect(sbi->ll_dt_exp, 0);
 348
 349         lprocfs_unregister_mountpoint(sbi);
 350         if (sbi->ll_proc_root) {
 351                 lprocfs_remove(sbi->ll_proc_root);
 352                 sbi->ll_proc_root = NULL;
 353         }
 354
 355         obd_disconnect(sbi->ll_md_exp, 0);
 356
 357         // We do this to get rid of orphaned dentries. That is not really trw.
 358         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 359                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 360                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 361                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 362                 shrink_dcache_parent(dentry);
 363         }
 364         EXIT;
 365 }
 366
 367 char *ll_read_opt(const char *opt, char *data)
 368 {
 369         char *value;
 370         char *retval;
 371         ENTRY;
 372
 373         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 374         if (strncmp(opt, data, strlen(opt)))
 375                 RETURN(NULL);
 376         if ((value = strchr(data, '=')) == NULL)
 377                 RETURN(NULL);
 378
 379         value++;
 380         OBD_ALLOC(retval, strlen(value) + 1);
 381         if (!retval) {
 382                 CERROR("out of memory!\n");
 383                 RETURN(NULL);
 384         }
 385
 386         memcpy(retval, value, strlen(value)+1);
 387         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 388         RETURN(retval);
 389 }
 390
 391 int ll_set_opt(const char *opt, char *data, int fl)
 392 {
 393         ENTRY;
 394
 395         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 396         if (strncmp(opt, data, strlen(opt)))
 397                 RETURN(0);
 398         else
 399                 RETURN(fl);
 400 }
 401
 402 void ll_options(char *options, char **lov, char **lmv, char **sec,
 403                 int *async, int *flags)
 404 {
 405         char *this_char;
 406 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 407         char *opt_ptr = options;
 408 #endif
 409         ENTRY;
 410
 411         if (!options) {
 412                 EXIT;
 413                 return;
 414         }
 415
 416         *async = 0;
 417 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 418         for (this_char = strtok (options, ",");
 419              this_char != NULL;
 420              this_char = strtok (NULL, ",")) {
 421 #else
 422         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 423 #endif
 424                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 425                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 426                         continue;
 427                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 428                         continue;
 429                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 430                         *async = 1;
 431                         continue;
 432                 }
 433                 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
 434                         continue;
 435                 if (!(*flags & LL_SBI_NOLCK) &&
 436                     ((*flags) = (*flags) |
 437                                 ll_set_opt("nolock", this_char,
 438                                            LL_SBI_NOLCK)))
 439                         continue;
 440         }
 441
 442         EXIT;
 443 }
 444
 445 void ll_lli_init(struct ll_inode_info *lli)
 446 {
 447         sema_init(&lli->lli_open_sem, 1);
 448         sema_init(&lli->lli_size_sem, 1);
 449         lli->lli_flags = 0;
 450         lli->lli_size_pid = 0;
 451         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 452         spin_lock_init(&lli->lli_lock);
 453         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 454         lli->lli_inode_magic = LLI_INODE_MAGIC;
 455         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 456         sema_init(&lli->lli_och_sem, 1);
 457         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 458         lli->lli_mds_exec_och = NULL;
 459         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 460         lli->lli_open_fd_exec_count = 0;
 461 }
 462
 463 int ll_fill_super(struct super_block *sb, void *data, int silent)
 464 {
 465         struct ll_sb_info *sbi;
 466         char *lov = NULL;
 467         char *lmv = NULL;
 468         int async, err;
 469         char *sec = NULL;
 470         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 471         __u64 remote_flag = 0;
 472         ENTRY;
 473
 474         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 475
 476         sbi = lustre_init_sbi(sb);
 477         if (!sbi)
 478                 RETURN(-ENOMEM);
 479
 480         sbi->ll_flags |= LL_SBI_READAHEAD;
 481         ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
 482
 483         if (!lov) {
 484                 CERROR("no osc\n");
 485                 GOTO(out, err = -EINVAL);
 486         }
 487
 488         if (!lmv) {
 489                 CERROR("no mdc\n");
 490                 GOTO(out, err = -EINVAL);
 491         }
 492
 493         err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
 494                                        &remote_flag);
 495         EXIT;
 496 out:
 497         if (err)
 498                 lustre_free_sbi(sb);
 499
 500         if (sec)
 501                 OBD_FREE(sec, strlen(sec) + 1);
 502         if (lmv)
 503                 OBD_FREE(lmv, strlen(lmv) + 1);
 504         if (lov)
 505                 OBD_FREE(lov, strlen(lov) + 1);
 506         return err;
 507 } /* ll_read_super */
 508
 509 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 510                               struct config_llog_instance *cfg, int allow_recov)
 511 {
 512         struct lustre_cfg lcfg;
 513         struct portals_cfg pcfg;
 514         char *peer = "MDS_PEER_UUID";
 515         struct obd_device *obd;
 516         struct lustre_handle md_conn = {0, };
 517         struct obd_export *exp;
 518         char *name = "mdc_dev";
 519         class_uuid_t uuid;
 520         struct obd_uuid lmv_uuid;
 521         struct llog_ctxt *ctxt;
 522         int rc = 0, err = 0;
 523         ENTRY;
 524
 525         if (lmd_bad_magic(lmd))
 526                 RETURN(-EINVAL);
 527
 528         generate_random_uuid(uuid);
 529         class_uuid_unparse(uuid, &lmv_uuid);
 530
 531         if (lmd->lmd_local_nid) {
 532                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 533                 pcfg.pcfg_nal = lmd->lmd_nal;
 534                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 535                 rc = libcfs_nal_cmd(&pcfg);
 536                 if (rc < 0)
 537                         GOTO(out, rc);
 538         }
 539
 540         if (lmd->lmd_nal == SOCKNAL ||
 541             lmd->lmd_nal == OPENIBNAL ||
 542             lmd->lmd_nal == IIBNAL ||
 543             lmd->lmd_nal == VIBNAL ||
 544             lmd->lmd_nal == RANAL) {
 545                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 546                 pcfg.pcfg_nal     = lmd->lmd_nal;
 547                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 548                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 549                 pcfg.pcfg_misc    = lmd->lmd_port;
 550                 rc = libcfs_nal_cmd(&pcfg);
 551                 if (rc < 0)
 552                         GOTO(out, rc);
 553         }
 554
 555         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
 556         lcfg.lcfg_nid = lmd->lmd_server_nid;
 557         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 558         lcfg.lcfg_inlbuf1 = peer;
 559         lcfg.lcfg_nal = lmd->lmd_nal;
 560         rc = class_process_config(&lcfg);
 561         if (rc < 0)
 562                 GOTO(out_del_conn, rc);
 563
 564         LCFG_INIT(lcfg, LCFG_ATTACH, name);
 565         lcfg.lcfg_inlbuf1 = "mdc";
 566         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 567         lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
 568         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 569         err = class_process_config(&lcfg);
 570         if (rc < 0)
 571                 GOTO(out_del_uuid, rc);
 572
 573         LCFG_INIT(lcfg, LCFG_SETUP, name);
 574         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
 575         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 576         lcfg.lcfg_inlbuf2 = peer;
 577         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 578         rc = class_process_config(&lcfg);
 579         if (rc < 0)
 580                 GOTO(out_detach, rc);
 581
 582         obd = class_name2obd(name);
 583         if (obd == NULL)
 584                 GOTO(out_cleanup, rc = -EINVAL);
 585
 586         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 587                           strlen(lmd->lmd_security), lmd->lmd_security);
 588         if (rc)
 589                 GOTO(out_cleanup, rc);
 590
 591         /* Disable initial recovery on this import */
 592         rc = obd_set_info(obd->obd_self_export,
 593                           strlen("initial_recov"), "initial_recov",
 594                           sizeof(allow_recov), &allow_recov);
 595         if (rc)
 596                 GOTO(out_cleanup, rc);
 597
 598         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 599         if (rc) {
 600                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 601                 GOTO(out_cleanup, rc);
 602         }
 603
 604         exp = class_conn2export(&md_conn);
 605
 606         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 607         rc = class_config_process_llog(ctxt, profile, cfg);
 608         if (rc)
 609                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 610
 611         err = obd_disconnect(exp, 0);
 612
 613         EXIT;
 614 out_cleanup:
 615         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
 616         err = class_process_config(&lcfg);
 617         if (err < 0)
 618                 GOTO(out, err);
 619
 620 out_detach:
 621         LCFG_INIT(lcfg, LCFG_DETACH, name);
 622         err = class_process_config(&lcfg);
 623         if (err < 0)
 624                 GOTO(out, err);
 625
 626 out_del_uuid:
 627         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
 628         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 629         lcfg.lcfg_inlbuf1 = peer;
 630         err = class_process_config(&lcfg);
 631
 632 out_del_conn:
 633         if (lmd->lmd_nal == SOCKNAL ||
 634             lmd->lmd_nal == OPENIBNAL ||
 635             lmd->lmd_nal == IIBNAL ||
 636             lmd->lmd_nal == VIBNAL ||
 637             lmd->lmd_nal == RANAL) {
 638                 int err2;
 639
 640                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 641                 pcfg.pcfg_nal     = lmd->lmd_nal;
 642                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 643                 pcfg.pcfg_flags   = 1;          /* single_share */
 644                 err2 = libcfs_nal_cmd(&pcfg);
 645                 if (err2 && !err)
 646                         err = err2;
 647                 if (err < 0)
 648                         GOTO(out, err);
 649         }
 650 out:
 651         if (rc == 0)
 652                 rc = err;
 653
 654         return rc;
 655 }
 656
 657 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 658 {
 659         struct lustre_cfg lcfg;
 660         struct obd_device *obd;
 661         int next = 0;
 662
 663         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 664         {
 665                 int err;
 666
 667                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
 668                 err = class_process_config(&lcfg);
 669                 if (err) {
 670                         CERROR("cleanup failed: %s\n", obd->obd_name);
 671                         //continue;
 672                 }
 673
 674                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
 675                 err = class_process_config(&lcfg);
 676                 if (err) {
 677                         CERROR("detach failed: %s\n", obd->obd_name);
 678                         //continue;
 679                 }
 680         }
 681
 682         if (sbi->ll_lmd != NULL)
 683                 class_del_profile(sbi->ll_lmd->lmd_profile);
 684 }
 685
 686 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 687 {
 688         struct lustre_mount_data * lmd = data;
 689         char *lov = NULL, *lmv = NULL;
 690         struct ll_sb_info *sbi;
 691         int err;
 692         ENTRY;
 693
 694         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 695         if (lmd_bad_magic(lmd))
 696                 RETURN(-EINVAL);
 697
 698         sbi = lustre_init_sbi(sb);
 699         if (!sbi)
 700                 RETURN(-ENOMEM);
 701
 702         sbi->ll_flags |= LL_SBI_READAHEAD;
 703
 704         if (lmd->lmd_profile) {
 705                 struct lustre_profile *lprof;
 706                 struct config_llog_instance cfg;
 707                 int len;
 708
 709                 if (lmd->lmd_mds[0] == '\0') {
 710                         CERROR("no mds name\n");
 711                         GOTO(out_free, err = -EINVAL);
 712                 }
 713                 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
 714
 715                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 716                 if (sbi->ll_lmd == NULL)
 717                         GOTO(out_free, err = -ENOMEM);
 718                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 719
 720                 /* generate a string unique to this super, let's try
 721                  the address of the super itself.*/
 722                 len = (sizeof(sb) * 2) + 1;
 723                 OBD_ALLOC(sbi->ll_instance, len);
 724                 if (sbi->ll_instance == NULL)
 725                         GOTO(out_free, err = -ENOMEM);
 726                 sprintf(sbi->ll_instance, "%p", sb);
 727
 728                 cfg.cfg_instance = sbi->ll_instance;
 729                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 730                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 731                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 732                 if (err < 0) {
 733                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 734                         GOTO(out_free, err);
 735                 }
 736
 737                 lprof = class_get_profile(lmd->lmd_profile);
 738                 if (lprof == NULL) {
 739                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 740                         GOTO(out_free, err = -EINVAL);
 741                 }
 742                 if (lov)
 743                         OBD_FREE(lov, strlen(lov) + 1);
 744                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
 745                           strlen(sbi->ll_instance) + 2);
 746                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 747
 748                 if (lmv)
 749                         OBD_FREE(lmv, strlen(lmv) + 1);
 750                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
 751                           strlen(sbi->ll_instance) + 2);
 752                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 753         }
 754
 755         if (!lov) {
 756                 CERROR("no osc\n");
 757                 GOTO(out_free, err = -EINVAL);
 758         }
 759
 760         if (!lmv) {
 761                 CERROR("no mdc\n");
 762                 GOTO(out_free, err = -EINVAL);
 763         }
 764
 765         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
 766                                        lmd->lmd_security, &lmd->lmd_nllu,
 767                                        &lmd->lmd_remote_flag);
 768
 769         if (err)
 770                 GOTO(out_free, err);
 771
 772 out_dev:
 773         if (lmv)
 774                 OBD_FREE(lmv, strlen(lmv) + 1);
 775         if (lov)
 776                 OBD_FREE(lov, strlen(lov) + 1);
 777
 778         RETURN(err);
 779
 780 out_free:
 781         if (sbi->ll_lmd) {
 782                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 783                 int err;
 784
 785                 if (sbi->ll_instance != NULL) {
 786                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 787                         struct config_llog_instance cfg;
 788                         char *cl_prof;
 789
 790                         cfg.cfg_instance = sbi->ll_instance;
 791                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 792
 793                         OBD_ALLOC(cl_prof, len);
 794                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 795                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 796                         if (err < 0) {
 797                                 CERROR("Unable to process log: %s\n", cl_prof);
 798                                 lustre_manual_cleanup(sbi);
 799                         }
 800                         OBD_FREE(cl_prof, len);
 801                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 802                 }
 803                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 804         }
 805         lustre_free_sbi(sb);
 806         goto out_dev;
 807 } /* lustre_fill_super */
 808
 809 void lustre_put_super(struct super_block *sb)
 810 {
 811         struct obd_device *obd;
 812         struct ll_sb_info *sbi = ll_s2sbi(sb);
 813         int force_umount = 0;
 814         ENTRY;
 815
 816         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 817         obd = class_exp2obd(sbi->ll_md_exp);
 818         if (obd)
 819                 force_umount = obd->obd_no_recov;
 820         obd = NULL;
 821
 822         lustre_common_put_super(sb);
 823         if (sbi->ll_lmd != NULL) {
 824                 char *cl_prof;
 825                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 826                 int err;
 827                 struct config_llog_instance cfg;
 828
 829                 if (force_umount) {
 830                         CERROR("force umount, doing manual cleanup\n");
 831                         lustre_manual_cleanup(sbi);
 832                         GOTO(free_lmd, 0);
 833                 }
 834
 835                 cfg.cfg_instance = sbi->ll_instance;
 836                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 837
 838                 OBD_ALLOC(cl_prof, len);
 839                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 840                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
 841                 if (err < 0) {
 842                         CERROR("Unable to process log: %s, doing manual cleanup"
 843                                "\n", cl_prof);
 844                         lustre_manual_cleanup(sbi);
 845                 }
 846
 847                 OBD_FREE(cl_prof, len);
 848         free_lmd:
 849                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 850                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 851         }
 852
 853         lustre_free_sbi(sb);
 854
 855         EXIT;
 856 } /* lustre_put_super */
 857
 858 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 859 {
 860         struct obd_export *md_exp = sbi->ll_md_exp;
 861         struct lustre_mount_data *lmd = sbi->ll_lmd;
 862         struct llog_ctxt *ctxt;
 863         struct config_llog_instance cfg;
 864         char *profile = lmd->lmd_profile, *name = NULL;
 865         int rc, namelen =  0, version;
 866         ENTRY;
 867
 868         if (profile == NULL)
 869                 RETURN(0);
 870         if (lmd == NULL) {
 871                 CERROR("Client not mounted with zero-conf; cannot process "
 872                        "update log.\n");
 873                 RETURN(0);
 874         }
 875
 876         rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
 877                                     LDLM_FL_CONFIG_CHANGE, NULL);
 878         if (rc != 0)
 879                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
 880
 881         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
 882                                NULL);
 883         if (rc != 0)
 884                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 885
 886         cfg.cfg_instance = sbi->ll_instance;
 887         cfg.cfg_uuid = sbi->ll_sb_uuid;
 888         cfg.cfg_local_nid = lmd->lmd_local_nid;
 889
 890         namelen = strlen(profile) + 20; /* -clean-######### */
 891         OBD_ALLOC(name, namelen);
 892         if (name == NULL)
 893                 RETURN(-ENOMEM);
 894
 895         if (clean) {
 896                 version = sbi->ll_config_version - 1;
 897                 sprintf(name, "%s-clean-%d", profile, version);
 898         } else {
 899                 version = sbi->ll_config_version + 1;
 900                 sprintf(name, "%s-%d", profile, version);
 901         }
 902
 903         CWARN("Applying configuration log %s\n", name);
 904
 905         ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
 906                                 LLOG_CONFIG_REPL_CTXT);
 907         rc = class_config_process_llog(ctxt, name, &cfg);
 908         if (rc == 0)
 909                 sbi->ll_config_version = version;
 910         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 911
 912         if (rc == 0 && clean == 0) {
 913                 struct lov_desc desc;
 914                 int rc, valsize;
 915                 valsize = sizeof(desc);
 916                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 917                                   "lovdesc", &valsize, &desc);
 918
 919                 rc = obd_init_ea_size(md_exp,
 920                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
 921                                       (desc.ld_tgt_count *
 922                                        sizeof(struct llog_cookie)));
 923         }
 924         OBD_FREE(name, namelen);
 925         RETURN(rc);
 926 }
 927
 928 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 929 {
 930         struct inode *inode = NULL;
 931         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 932         if (lock->l_ast_data) {
 933                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 934                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 935                         inode = igrab(lock->l_ast_data);
 936                 } else {
 937                         inode = lock->l_ast_data;
 938                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
 939                                "l_ast_data %p is bogus: magic %0x8\n",
 940                                lock->l_ast_data, lli->lli_inode_magic);
 941                         inode = NULL;
 942                 }
 943         }
 944         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 945         return inode;
 946 }
 947
 948 int null_if_equal(struct ldlm_lock *lock, void *data)
 949 {
 950         if (data == lock->l_ast_data) {
 951                 lock->l_ast_data = NULL;
 952
 953                 if (lock->l_req_mode != lock->l_granted_mode)
 954                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 955         }
 956
 957         return LDLM_ITER_CONTINUE;
 958 }
 959
 960 void ll_clear_inode(struct inode *inode)
 961 {
 962         struct lustre_id id;
 963         struct ll_inode_info *lli = ll_i2info(inode);
 964         struct ll_sb_info *sbi = ll_i2sbi(inode);
 965         ENTRY;
 966
 967         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 968                inode->i_generation, inode);
 969
 970         lli->lli_inode_magic = LLI_INODE_DEAD;
 971         ll_inode2id(&id, inode);
 972
 973         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 974         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
 975
 976         LASSERT(!lli->lli_open_fd_write_count);
 977         LASSERT(!lli->lli_open_fd_read_count);
 978         LASSERT(!lli->lli_open_fd_exec_count);
 979
 980         if (lli->lli_mds_write_och)
 981                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
 982         if (lli->lli_mds_exec_och)
 983                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
 984         if (lli->lli_mds_read_och)
 985                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
 986
 987         if (lli->lli_smd)
 988                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 989                                   null_if_equal, inode);
 990
 991         if (lli->lli_smd) {
 992                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
 993                 lli->lli_smd = NULL;
 994         }
 995
 996         if (lli->lli_mea) {
 997                 obd_free_memmd(sbi->ll_md_exp,
 998                                (struct lov_stripe_md **) &lli->lli_mea);
 999                 lli->lli_mea = NULL;
1000         }
1001
1002         if (lli->lli_symlink_name) {
1003                 OBD_FREE(lli->lli_symlink_name,
1004                          strlen(lli->lli_symlink_name) + 1);
1005                 lli->lli_symlink_name = NULL;
1006         }
1007         lli->lli_inode_magic = LLI_INODE_DEAD;
1008
1009         EXIT;
1010 }
1011
1012 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1013  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1014  * keep these values until such a time that objects are allocated for it.
1015  * We do the MDS operations first, as it is checking permissions for us.
1016  * We don't to the MDS RPC if there is nothing that we want to store there,
1017  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1018  * going to do an RPC anyways.
1019  *
1020  * If we are doing a truncate, we will send the mtime and ctime updates
1021  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1022  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1023  * at the same time.
1024  */
1025 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1026 {
1027         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1028         struct ll_sb_info *sbi = ll_i2sbi(inode);
1029         struct ptlrpc_request *request = NULL;
1030         struct mdc_op_data *op_data;
1031         int ia_valid = attr->ia_valid;
1032         int rc = 0;
1033         ENTRY;
1034
1035         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1036         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1037
1038         if (ia_valid & ATTR_SIZE) {
1039                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1040                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1041                                attr->ia_size, ll_file_maxbytes(inode));
1042                         RETURN(-EFBIG);
1043                 }
1044
1045                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1046         }
1047
1048         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1049         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1050                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1051                         RETURN(-EPERM);
1052         }
1053
1054         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1055         if (attr->ia_valid & ATTR_CTIME) {
1056                 attr->ia_ctime = CURRENT_TIME;
1057                 attr->ia_valid |= ATTR_CTIME_SET;
1058         }
1059         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1060                 attr->ia_atime = CURRENT_TIME;
1061                 attr->ia_valid |= ATTR_ATIME_SET;
1062         }
1063         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1064                 attr->ia_mtime = CURRENT_TIME;
1065                 attr->ia_valid |= ATTR_MTIME_SET;
1066         }
1067
1068         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1069                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1070                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1071                        LTIME_S(CURRENT_TIME));
1072
1073         if (lsm)
1074                 attr->ia_valid &= ~ATTR_SIZE;
1075
1076         /* If only OST attributes being set on objects, don't do MDS RPC.
1077          * In that case, we need to check permissions and update the local
1078          * inode ourselves so we can call obdo_from_inode() always. */
1079         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1080                 struct lustre_md md;
1081
1082                 OBD_ALLOC(op_data, sizeof(*op_data));
1083                 if (op_data == NULL)
1084                         RETURN(-ENOMEM);
1085                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1086
1087                 rc = md_setattr(sbi->ll_md_exp, op_data,
1088                                 attr, NULL, 0, NULL, 0, &request);
1089                 OBD_FREE(op_data, sizeof(*op_data));
1090                 if (rc) {
1091                         ptlrpc_req_finished(request);
1092                         if (rc != -EPERM && rc != -EACCES)
1093                                 CERROR("md_setattr fails: rc = %d\n", rc);
1094                         RETURN(rc);
1095                 }
1096
1097                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1098                                        sbi->ll_dt_exp, &md);
1099                 if (rc) {
1100                         ptlrpc_req_finished(request);
1101                         RETURN(rc);
1102                 }
1103
1104                 /* We call inode_setattr to adjust timestamps, but we first
1105                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1106                  *
1107                  * NB: ATTR_SIZE will only be set at this point if the size
1108                  * resides on the MDS, ie, this file has no objects. */
1109                 attr->ia_valid &= ~ATTR_SIZE;
1110                 inode_setattr(inode, attr);
1111
1112                 ll_update_inode(inode, &md);
1113                 ptlrpc_req_finished(request);
1114
1115                 if (!lsm || !S_ISREG(inode->i_mode)) {
1116                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1117                         RETURN(0);
1118                 }
1119         } else {
1120                 /* The OST doesn't check permissions, but the alternative is
1121                  * a gratuitous RPC to the MDS.  We already rely on the client
1122                  * to do read/write/truncate permission checks, so is mtime OK?
1123                  */
1124                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1125                         /* from sys_utime() */
1126                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1127                                 if (current->fsuid != inode->i_uid &&
1128                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1129                                         RETURN(rc);
1130                         } else {
1131                                 /* from inode_change_ok() */
1132                                 if (current->fsuid != inode->i_uid &&
1133                                     !capable(CAP_FOWNER))
1134                                         RETURN(-EPERM);
1135                         }
1136                 }
1137
1138                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1139                 inode_setattr(inode, attr);
1140         }
1141
1142         /* We really need to get our PW lock before we change inode->i_size.
1143          * If we don't we can race with other i_size updaters on our node, like
1144          * ll_file_read.  We can also race with i_size propogation to other
1145          * nodes through dirtying and writeback of final cached pages.  This
1146          * last one is especially bad for racing o_append users on other
1147          * nodes. */
1148         if (ia_valid & ATTR_SIZE) {
1149                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1150                                                            OBD_OBJECT_EOF } };
1151                 struct lustre_handle lockh = { 0 };
1152                 struct ll_inode_info *lli = ll_i2info(inode);
1153                 int err, ast_flags = 0;
1154                 /* XXX when we fix the AST intents to pass the discard-range
1155                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1156                  * XXX here. */
1157                 if (attr->ia_size == 0)
1158                         ast_flags = LDLM_AST_DISCARD_DATA;
1159
1160                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1161                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1162
1163                 if (rc != 0)
1164                         RETURN(rc);
1165
1166                 down(&lli->lli_size_sem);
1167                 lli->lli_size_pid = current->pid;
1168                 rc = vmtruncate(inode, attr->ia_size);
1169                 if (rc != 0) {
1170                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1171                         lli->lli_size_pid = 0;
1172                         up(&lli->lli_size_sem);
1173                 }
1174
1175                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1176                 if (err) {
1177                         CERROR("ll_extent_unlock failed: %d\n", err);
1178                         if (!rc)
1179                                 rc = err;
1180                 }
1181         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1182                 struct obdo *oa = NULL;
1183
1184                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1185                        inode->i_ino, LTIME_S(attr->ia_mtime));
1186
1187                 oa = obdo_alloc();
1188                 if (oa == NULL)
1189                         RETURN(-ENOMEM);
1190
1191                 oa->o_id = lsm->lsm_object_id;
1192                 oa->o_gr = lsm->lsm_object_gr;
1193                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1194                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1195                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1196                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1197                 obdo_free(oa);
1198                 if (rc)
1199                         CERROR("obd_setattr fails: rc = %d\n", rc);
1200         }
1201         RETURN(rc);
1202 }
1203
1204 int ll_setattr(struct dentry *de, struct iattr *attr)
1205 {
1206         LASSERT(de->d_inode);
1207         return ll_setattr_raw(de->d_inode, attr);
1208 }
1209
1210 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1211                        unsigned long max_age)
1212 {
1213         struct ll_sb_info *sbi = ll_s2sbi(sb);
1214         struct obd_statfs obd_osfs;
1215         int rc;
1216         ENTRY;
1217
1218         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1219         if (rc) {
1220                 CERROR("obd_statfs fails: rc = %d\n", rc);
1221                 RETURN(rc);
1222         }
1223
1224         osfs->os_type = sb->s_magic;
1225
1226         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1227                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1228
1229         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1230         if (rc) {
1231                 CERROR("obd_statfs fails: rc = %d\n", rc);
1232                 RETURN(rc);
1233         }
1234
1235         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1236                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1237                obd_osfs.os_files);
1238
1239         osfs->os_blocks = obd_osfs.os_blocks;
1240         osfs->os_bfree = obd_osfs.os_bfree;
1241         osfs->os_bavail = obd_osfs.os_bavail;
1242
1243         /* If we don't have as many objects free on the OST as inodes
1244          * on the MDS, we reduce the total number of inodes to
1245          * compensate, so that the "inodes in use" number is correct.
1246          */
1247         if (obd_osfs.os_ffree < osfs->os_ffree) {
1248                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1249                         obd_osfs.os_ffree;
1250                 osfs->os_ffree = obd_osfs.os_ffree;
1251         }
1252
1253         RETURN(rc);
1254 }
1255
1256 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1257 {
1258         struct obd_statfs osfs;
1259         int rc;
1260
1261         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1262         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1263
1264         /* For now we will always get up-to-date statfs values, but in the
1265          * future we may allow some amount of caching on the client (e.g.
1266          * from QOS or lprocfs updates). */
1267         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1268         if (rc)
1269                 return rc;
1270
1271         statfs_unpack(sfs, &osfs);
1272
1273         if (sizeof(sfs->f_blocks) == 4) {
1274                 while (osfs.os_blocks > ~0UL) {
1275                         sfs->f_bsize <<= 1;
1276
1277                         osfs.os_blocks >>= 1;
1278                         osfs.os_bfree >>= 1;
1279                         osfs.os_bavail >>= 1;
1280                 }
1281         }
1282
1283         sfs->f_blocks = osfs.os_blocks;
1284         sfs->f_bfree = osfs.os_bfree;
1285         sfs->f_bavail = osfs.os_bavail;
1286
1287         return 0;
1288 }
1289
1290 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1291 {
1292         struct ll_inode_info *lli = ll_i2info(inode);
1293         struct lov_stripe_md *lsm = md->lsm;
1294         struct mds_body *body = md->body;
1295         struct mea *mea = md->mea;
1296         struct posix_acl *ll_acl_access = md->acl_access;
1297         ENTRY;
1298
1299         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1300
1301         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1302                 /* check for default striping info for dir. */
1303                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1304         }
1305
1306         if (lsm != NULL) {
1307                 LASSERT(lsm->lsm_object_gr > 0);
1308                 if (lli->lli_smd == NULL) {
1309                         lli->lli_smd = lsm;
1310                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1311                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1312                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1313                 } else {
1314                         int i;
1315                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1316                                 CERROR("lsm mismatch for inode %ld\n",
1317                                        inode->i_ino);
1318                                 CERROR("lli_smd:\n");
1319                                 dump_lsm(D_ERROR, lli->lli_smd);
1320                                 CERROR("lsm:\n");
1321                                 dump_lsm(D_ERROR, lsm);
1322                                 LBUG();
1323                         }
1324                         /* XXX FIXME -- We should decide on a safer (atomic) and
1325                          * more elegant way to update the lsm */
1326                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1327                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1328                                         lsm->lsm_oinfo[i].loi_id;
1329                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1330                                         lsm->lsm_oinfo[i].loi_gr;
1331                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1332                                         lsm->lsm_oinfo[i].loi_ost_idx;
1333                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1334                                         lsm->lsm_oinfo[i].loi_ost_gen;
1335                         }
1336                 }
1337                 /* bug 2844 - limit i_blksize for broken user-space apps */
1338                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1339                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1340                 if (lli->lli_smd != lsm)
1341                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1342         }
1343
1344         if (mea != NULL) {
1345                 if (lli->lli_mea == NULL) {
1346                         lli->lli_mea = mea;
1347                 } else {
1348                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1349                                 CERROR("mea mismatch for inode %lu\n",
1350                                         inode->i_ino);
1351                                 LBUG();
1352                         }
1353                 }
1354                 if (lli->lli_mea != mea)
1355                         obd_free_memmd(ll_i2mdexp(inode),
1356                                        (struct lov_stripe_md **) &mea);
1357         }
1358
1359         if (body->valid & OBD_MD_FID)
1360                 id_assign_fid(&lli->lli_id, &body->id1);
1361
1362         if (body->valid & OBD_MD_FLID)
1363                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1364
1365         if (body->valid & OBD_MD_FLGENER)
1366                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1367
1368         spin_lock(&lli->lli_lock);
1369         if (ll_acl_access != NULL) {
1370                 if (lli->lli_acl_access != NULL)
1371                         posix_acl_release(lli->lli_acl_access);
1372                 lli->lli_acl_access = ll_acl_access;
1373         }
1374         spin_unlock(&lli->lli_lock);
1375
1376         if (body->valid & OBD_MD_FLID)
1377                 inode->i_ino = id_ino(&body->id1);
1378         if (body->valid & OBD_MD_FLGENER)
1379                 inode->i_generation = id_gen(&body->id1);
1380         if (body->valid & OBD_MD_FLATIME)
1381                 LTIME_S(inode->i_atime) = body->atime;
1382         if (body->valid & OBD_MD_FLMTIME &&
1383             body->mtime > LTIME_S(inode->i_mtime)) {
1384                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1385                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1386                 LTIME_S(inode->i_mtime) = body->mtime;
1387         }
1388         if (body->valid & OBD_MD_FLCTIME &&
1389             body->ctime > LTIME_S(inode->i_ctime))
1390                 LTIME_S(inode->i_ctime) = body->ctime;
1391         if (body->valid & OBD_MD_FLMODE) {
1392                 inode->i_mode = (inode->i_mode & S_IFMT) |
1393                         (body->mode & ~S_IFMT);
1394         }
1395         if (body->valid & OBD_MD_FLTYPE) {
1396                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1397                         (body->mode & S_IFMT);
1398         }
1399         if (body->valid & OBD_MD_FLUID)
1400                 inode->i_uid = body->uid;
1401         if (body->valid & OBD_MD_FLGID)
1402                 inode->i_gid = body->gid;
1403         if (body->valid & OBD_MD_FLFLAGS)
1404                 inode->i_flags = body->flags;
1405         if (body->valid & OBD_MD_FLNLINK)
1406                 inode->i_nlink = body->nlink;
1407         if (body->valid & OBD_MD_FLRDEV)
1408 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1409                 inode->i_rdev = body->rdev;
1410 #else
1411                 inode->i_rdev = old_decode_dev(body->rdev);
1412 #endif
1413         if (body->valid & OBD_MD_FLSIZE)
1414                 inode->i_size = body->size;
1415         if (body->valid & OBD_MD_FLBLOCKS)
1416                 inode->i_blocks = body->blocks;
1417
1418         if (body->valid & OBD_MD_FLSIZE)
1419                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1420
1421 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1422         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1423 #endif
1424         LASSERT(id_fid(&lli->lli_id) != 0);
1425 }
1426
1427 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1428 static struct backing_dev_info ll_backing_dev_info = {
1429         .ra_pages       = 0,    /* No readahead */
1430         .memory_backed  = 0,    /* Does contribute to dirty memory */
1431 };
1432 #endif
1433
1434 void ll_read_inode2(struct inode *inode, void *opaque)
1435 {
1436         struct lustre_md *md = opaque;
1437         struct ll_inode_info *lli = ll_i2info(inode);
1438         ENTRY;
1439
1440         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1441                inode->i_generation, inode);
1442
1443         ll_lli_init(lli);
1444
1445         LASSERT(!lli->lli_smd);
1446
1447         /* Core attributes from the MDS first.  This is a new inode, and
1448          * the VFS doesn't zero times in the core inode so we have to do
1449          * it ourselves.  They will be overwritten by either MDS or OST
1450          * attributes - we just need to make sure they aren't newer. */
1451         LTIME_S(inode->i_mtime) = 0;
1452         LTIME_S(inode->i_atime) = 0;
1453         LTIME_S(inode->i_ctime) = 0;
1454
1455         inode->i_rdev = 0;
1456         ll_update_inode(inode, md);
1457
1458         /* OIDEBUG(inode); */
1459
1460         if (S_ISREG(inode->i_mode)) {
1461                 inode->i_op = &ll_file_inode_operations;
1462                 inode->i_fop = &ll_file_operations;
1463                 inode->i_mapping->a_ops = &ll_aops;
1464                 EXIT;
1465         } else if (S_ISDIR(inode->i_mode)) {
1466                 inode->i_op = &ll_dir_inode_operations;
1467                 inode->i_fop = &ll_dir_operations;
1468                 inode->i_mapping->a_ops = &ll_dir_aops;
1469                 EXIT;
1470         } else if (S_ISLNK(inode->i_mode)) {
1471                 inode->i_op = &ll_fast_symlink_inode_operations;
1472                 EXIT;
1473         } else {
1474                 inode->i_op = &ll_special_inode_operations;
1475
1476 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1477                 init_special_inode(inode, inode->i_mode,
1478                                    kdev_t_to_nr(inode->i_rdev));
1479
1480                 /* initializing backing dev info. */
1481                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1482 #else
1483                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1484 #endif
1485                 lli->ll_save_ifop = inode->i_fop;
1486
1487                 if (S_ISCHR(inode->i_mode))
1488                         inode->i_fop = &ll_special_chr_inode_fops;
1489                 else if (S_ISBLK(inode->i_mode))
1490                         inode->i_fop = &ll_special_blk_inode_fops;
1491                 else if (S_ISFIFO(inode->i_mode))
1492                         inode->i_fop = &ll_special_fifo_inode_fops;
1493                 else if (S_ISSOCK(inode->i_mode))
1494                         inode->i_fop = &ll_special_sock_inode_fops;
1495
1496                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1497                       inode->i_fop);
1498
1499                 if (lli->ll_save_ifop->owner) {
1500                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1501                               lli->ll_save_ifop->owner);
1502                 }
1503                 EXIT;
1504         }
1505 }
1506
1507 void ll_delete_inode(struct inode *inode)
1508 {
1509         struct ll_sb_info *sbi = ll_i2sbi(inode);
1510         struct lustre_id id;
1511         int rc;
1512         ENTRY;
1513
1514         ll_inode2id(&id, inode);
1515
1516         rc = md_delete_inode(sbi->ll_md_exp, &id);
1517         if (rc) {
1518                 CERROR("md_delete_inode() failed, error %d\n",
1519                        rc);
1520         }
1521
1522         clear_inode(inode);
1523         EXIT;
1524 }
1525
1526 int ll_iocontrol(struct inode *inode, struct file *file,
1527                  unsigned int cmd, unsigned long arg)
1528 {
1529         struct ll_sb_info *sbi = ll_i2sbi(inode);
1530         struct ptlrpc_request *req = NULL;
1531         int rc, flags = 0;
1532         ENTRY;
1533
1534         switch(cmd) {
1535         case EXT3_IOC_GETFLAGS: {
1536                 struct lustre_id id;
1537                 __u64 valid = OBD_MD_FLFLAGS;
1538                 struct mds_body *body;
1539
1540                 ll_inode2id(&id, inode);
1541                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1542                 if (rc) {
1543                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1544                         RETURN(-abs(rc));
1545                 }
1546
1547                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1548
1549                 if (body->flags & S_APPEND)
1550                         flags |= EXT3_APPEND_FL;
1551                 if (body->flags & S_IMMUTABLE)
1552                         flags |= EXT3_IMMUTABLE_FL;
1553                 if (body->flags & S_NOATIME)
1554                         flags |= EXT3_NOATIME_FL;
1555
1556                 ptlrpc_req_finished (req);
1557
1558                 RETURN(put_user(flags, (int *)arg));
1559         }
1560         case EXT3_IOC_SETFLAGS: {
1561                 struct mdc_op_data *op_data;
1562                 struct iattr attr;
1563                 struct obdo *oa;
1564                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1565
1566                 if (get_user(flags, (int *)arg))
1567                         RETURN(-EFAULT);
1568
1569                 oa = obdo_alloc();
1570                 if (!oa)
1571                         RETURN(-ENOMEM);
1572
1573                 OBD_ALLOC(op_data, sizeof(*op_data));
1574                 if (op_data == NULL) {
1575                         obdo_free(oa);
1576                         RETURN(-ENOMEM);
1577                 }
1578                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1579
1580                 memset(&attr, 0x0, sizeof(attr));
1581                 attr.ia_attr_flags = flags;
1582                 attr.ia_valid |= ATTR_ATTR_FLAG;
1583
1584                 rc = md_setattr(sbi->ll_md_exp, op_data,
1585                                 &attr, NULL, 0, NULL, 0, &req);
1586                 OBD_FREE(op_data, sizeof(*op_data));
1587                 if (rc) {
1588                         ptlrpc_req_finished(req);
1589                         if (rc != -EPERM && rc != -EACCES)
1590                                 CERROR("md_setattr fails: rc = %d\n", rc);
1591                         obdo_free(oa);
1592                         RETURN(rc);
1593                 }
1594                 ptlrpc_req_finished(req);
1595
1596                 oa->o_id = lsm->lsm_object_id;
1597                 oa->o_gr = lsm->lsm_object_gr;
1598                 oa->o_flags = flags;
1599                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1600
1601                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1602                 obdo_free(oa);
1603                 if (rc) {
1604                         if (rc != -EPERM && rc != -EACCES)
1605                                 CERROR("md_setattr fails: rc = %d\n", rc);
1606                         RETURN(rc);
1607                 }
1608
1609                 if (flags & EXT3_APPEND_FL)
1610                         inode->i_flags |= S_APPEND;
1611                 else
1612                         inode->i_flags &= ~S_APPEND;
1613                 if (flags & EXT3_IMMUTABLE_FL)
1614                         inode->i_flags |= S_IMMUTABLE;
1615                 else
1616                         inode->i_flags &= ~S_IMMUTABLE;
1617                 if (flags & EXT3_NOATIME_FL)
1618                         inode->i_flags |= S_NOATIME;
1619                 else
1620                         inode->i_flags &= ~S_NOATIME;
1621
1622                 RETURN(0);
1623         }
1624         default:
1625                 RETURN(-ENOSYS);
1626         }
1627
1628         RETURN(0);
1629 }
1630
1631 /* this is only called in the case of forced umount. */
1632 void ll_umount_begin(struct super_block *sb)
1633 {
1634         struct ll_sb_info *sbi = ll_s2sbi(sb);
1635         struct obd_ioctl_data ioc_data = { 0 };
1636         struct obd_device *obd;
1637         ENTRY;
1638
1639         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1640                sb->s_count, atomic_read(&sb->s_active));
1641
1642         obd = class_exp2obd(sbi->ll_md_exp);
1643         if (obd == NULL) {
1644                 CERROR("Invalid MDC connection handle "LPX64"\n",
1645                        sbi->ll_md_exp->exp_handle.h_cookie);
1646                 EXIT;
1647                 return;
1648         }
1649         obd->obd_no_recov = 1;
1650         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1651                       sizeof(ioc_data), &ioc_data, NULL);
1652
1653         obd = class_exp2obd(sbi->ll_dt_exp);
1654         if (obd == NULL) {
1655                 CERROR("Invalid LOV connection handle "LPX64"\n",
1656                        sbi->ll_dt_exp->exp_handle.h_cookie);
1657                 EXIT;
1658                 return;
1659         }
1660
1661         obd->obd_no_recov = 1;
1662         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1663                       sizeof(ioc_data), &ioc_data, NULL);
1664
1665         /*
1666          * really, we'd like to wait until there are no requests outstanding,
1667          * and then continue.  For now, we just invalidate the requests,
1668          * schedule, and hope.
1669          */
1670         schedule();
1671
1672         EXIT;
1673 }
1674
1675 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1676                   struct inode **inode, struct ptlrpc_request *req,
1677                   int offset, struct super_block *sb)
1678 {
1679         struct lustre_md md;
1680         int rc = 0;
1681
1682         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1683         if (rc)
1684                 RETURN(rc);
1685
1686         if (*inode) {
1687                 ll_update_inode(*inode, &md);
1688         } else {
1689                 LASSERT(sb);
1690                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1691                 if (*inode == NULL || is_bad_inode(*inode)) {
1692                         /* free the lsm if we allocated one above */
1693                         if (md.lsm != NULL)
1694                                 obd_free_memmd(dt_exp, &md.lsm);
1695                         if (md.mea != NULL)
1696                                 obd_free_memmd(md_exp,
1697                                                (struct lov_stripe_md**)&md.mea);
1698                         rc = -ENOMEM;
1699                         CERROR("new_inode -fatal: rc %d\n", rc);
1700                 }
1701         }
1702
1703         RETURN(rc);
1704 }
1705
1706 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1707                char *filename, struct lustre_id *ret)
1708 {
1709         struct ptlrpc_request *request = NULL;
1710         struct mds_body *body;
1711         int rc;
1712
1713         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1714                              OBD_MD_FID, 0, &request);
1715         if (rc < 0) {
1716                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1717                        filename, rc);
1718                 return rc;
1719         }
1720
1721         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1722         LASSERT(body != NULL);
1723         LASSERT_REPSWABBED(request, 0);
1724
1725         *ret = body->id1;
1726         ptlrpc_req_finished(request);
1727
1728         return rc;
1729 }