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         init_completion(&sbi->ll_gns_mount_finished);
  74
  75         /* this later may be reset via /proc/fs/... */
  76         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
  77         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
  78
  79         /* this later may be reset via /proc/fs/... */
  80         memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
  81                strlen("/usr/sbin/gns_upcall"));
  82         sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
  83
  84         /* default values, may be changed via /proc/fs/... */
  85         sbi->ll_gns_state = LL_GNS_IDLE;
  86         sbi->ll_gns_pending_dentry = NULL;
  87         atomic_set(&sbi->ll_gns_enabled, 1);
  88         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  89         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  90
  91         sbi->ll_gns_timer.data = (unsigned long)sbi;
  92         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  93         init_timer(&sbi->ll_gns_timer);
  94
  95         ll_set_sbi(sb, sbi);
  96
  97         generate_random_uuid(uuid);
  98         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  99         RETURN(sbi);
 100 }
 101
 102 void lustre_free_sbi(struct super_block *sb)
 103 {
 104         struct ll_sb_info *sbi = ll_s2sbi(sb);
 105         ENTRY;
 106
 107         if (sbi != NULL) {
 108                 list_del(&sbi->ll_gns_sbi_head);
 109                 del_timer(&sbi->ll_gns_timer);
 110                 OBD_FREE(sbi, sizeof(*sbi));
 111         }
 112         ll_set_sbi(sb, NULL);
 113         EXIT;
 114 }
 115
 116 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 117 {
 118         __u32 valsize;
 119         int rc = 0;
 120         ENTRY;
 121
 122         valsize = sizeof(sbi->ll_dt_desc);
 123         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 124         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 125                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 126         RETURN(rc);
 127 }
 128
 129 extern struct dentry_operations ll_d_ops;
 130
 131 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 132                              int async, char *security, __u32 *nllu,
 133                              __u64 *remote)
 134 {
 135         struct ll_sb_info *sbi = ll_s2sbi(sb);
 136         struct ptlrpc_request *request = NULL;
 137         struct lustre_handle dt_conn = {0, };
 138         struct lustre_handle md_conn = {0, };
 139         struct obd_connect_data *data;
 140         struct inode *root = NULL;
 141         struct obd_device *obd;
 142         struct obd_statfs osfs;
 143         struct lustre_md md;
 144         kdev_t devno;
 145         int err;
 146         __u32 valsize;
 147         ENTRY;
 148
 149         obd = class_name2obd(lmv);
 150         if (!obd) {
 151                 CERROR("MDC %s: not setup or attached\n", lmv);
 152                 RETURN(-EINVAL);
 153         }
 154         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 155                      sizeof(async), &async);
 156
 157         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 158             (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 159                 CERROR("wrong remote flag "LPX64"\n", *remote);
 160                 RETURN(-EINVAL);
 161         }
 162
 163         OBD_ALLOC(data, sizeof(*data));
 164         if (!data)
 165                 RETURN(-ENOMEM);
 166
 167         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 168                                               OBD_CONNECT_REMOTE);
 169         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 170
 171         if (security == NULL)
 172                 security = "null";
 173
 174         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 175                            strlen(security), security);
 176         if (err) {
 177                 CERROR("LMV %s: failed to set security %s, err %d\n",
 178                         lmv, security, err);
 179                 OBD_FREE(data, sizeof(*data));
 180                 RETURN(err);
 181         }
 182
 183         if (proc_lustre_fs_root) {
 184                 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
 185                                                   sb, lov, lmv);
 186                 if (err < 0)
 187                         CERROR("could not register mount in /proc/lustre");
 188         }
 189
 190         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
 191                           OBD_OPT_REAL_CLIENT);
 192         if (err == -EBUSY) {
 193                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 194                        " client is not a part.  Please wait for recovery to "
 195                        "complete, abort, or time out.\n", lmv);
 196                 GOTO(out, err);
 197         } else if (err) {
 198                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 199                 GOTO(out, err);
 200         }
 201         sbi->ll_md_exp = class_conn2export(&md_conn);
 202         err = obd_statfs(obd, &osfs, jiffies - HZ);
 203         if (err)
 204                 GOTO(out_lmv, err);
 205
 206         if (!osfs.os_bsize) {
 207                 CERROR("Invalid block size is detected.");
 208                 GOTO(out_lmv, err);
 209         }
 210
 211         sb->s_magic = LL_SUPER_MAGIC;
 212         sb->s_blocksize = osfs.os_bsize;
 213         sb->s_blocksize_bits = log2(osfs.os_bsize);
 214         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 215
 216         devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 217                              strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 218
 219         sb->s_dev = devno;
 220
 221         /* after statfs, we are supposed to have connected to MDSs,
 222          * so it's ok to check remote flag returned.
 223          */
 224         valsize = sizeof(&sbi->ll_remote);
 225         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 226                            &valsize, &sbi->ll_remote);
 227         if (err) {
 228                 CERROR("fail to obtain remote flag\n");
 229                 GOTO(out, err);
 230         }
 231
 232         obd = class_name2obd(lov);
 233         if (!obd) {
 234                 CERROR("OSC %s: not setup or attached\n", lov);
 235                 GOTO(out_lmv, err);
 236         }
 237         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 238                      sizeof(async), &async);
 239
 240         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
 241                           OBD_OPT_REAL_CLIENT);
 242         if (err == -EBUSY) {
 243                 CERROR("An OST (lov %s) is performing recovery, of which this"
 244                        " client is not a part.  Please wait for recovery to "
 245                        "complete, abort, or time out.\n", lov);
 246                 GOTO(out, err);
 247         } else if (err) {
 248                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 249                 GOTO(out_lmv, err);
 250         }
 251         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 252
 253         err = lustre_init_dt_desc(sbi);
 254         if (err == 0) {
 255                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 256                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
 257                                  sbi->ll_dt_desc.ld_tgt_count *
 258                                  sizeof(struct llog_cookie));
 259         }
 260
 261         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 262         if (err) {
 263                 CERROR("cannot mds_connect: rc = %d\n", err);
 264                 GOTO(out_lov, err);
 265         }
 266         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 267
 268         sb->s_op = &lustre_super_operations;
 269
 270         /* make root inode */
 271         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 272                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
 273                          NULL, 0, 0, &request);
 274         if (err) {
 275                 CERROR("md_getattr failed for root: rc = %d\n", err);
 276                 GOTO(out_lov, err);
 277         }
 278
 279         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 280                                 sbi->ll_dt_exp, &md);
 281         if (err) {
 282                 CERROR("failed to understand root inode md: rc = %d\n", err);
 283                 ptlrpc_req_finished(request);
 284                 GOTO(out_lov, err);
 285         }
 286
 287         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 288         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 289
 290         ptlrpc_req_finished(request);
 291
 292         if (root == NULL || is_bad_inode(root)) {
 293                 if (md.lsm != NULL)
 294                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 295                 if (md.mea != NULL)
 296                     obd_free_memmd(sbi->ll_md_exp,
 297                                    (struct lov_stripe_md**)&md.mea);
 298                 CERROR("lustre_lite: bad iget4 for root\n");
 299                 GOTO(out_root, err = -EBADF);
 300         }
 301
 302         err = ll_close_thread_start(&sbi->ll_lcq);
 303         if (err) {
 304                 CERROR("cannot start close thread: rc %d\n", err);
 305                 GOTO(out_root, err);
 306         }
 307
 308         ll_gns_add_timer(sbi);
 309
 310         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 311            backing dev info assigned to inode mapping is used for
 312            determining maximal readahead. */
 313 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 314     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 315         /* bug 2805 - set VM readahead to zero */
 316         vm_max_readahead = vm_min_readahead = 0;
 317 #endif
 318
 319         sb->s_root = d_alloc_root(root);
 320         sb->s_root->d_op = &ll_d_ops;
 321
 322         sb->s_flags |= MS_POSIXACL;
 323 #ifdef S_PDIROPS
 324         CWARN("Enabling PDIROPS\n");
 325         sb->s_flags |= S_PDIROPS;
 326 #endif
 327
 328         if (data != NULL)
 329                 OBD_FREE(data, sizeof(*data));
 330         RETURN(err);
 331 out_root:
 332         if (root)
 333                 iput(root);
 334 out_lov:
 335         obd_disconnect(sbi->ll_dt_exp, 0);
 336 out_lmv:
 337         obd_disconnect(sbi->ll_md_exp, 0);
 338 out:
 339         if (data != NULL)
 340                 OBD_FREE(data, sizeof(*data));
 341         lprocfs_unregister_mountpoint(sbi);
 342         return err;
 343 }
 344
 345 void lustre_common_put_super(struct super_block *sb)
 346 {
 347         struct ll_sb_info *sbi = ll_s2sbi(sb);
 348         struct hlist_node *tmp, *next;
 349         ENTRY;
 350
 351         ll_gns_del_timer(sbi);
 352         ll_close_thread_shutdown(sbi->ll_lcq);
 353
 354         list_del(&sbi->ll_conn_chain);
 355         obd_disconnect(sbi->ll_dt_exp, 0);
 356
 357         lprocfs_unregister_mountpoint(sbi);
 358         if (sbi->ll_proc_root) {
 359                 lprocfs_remove(sbi->ll_proc_root);
 360                 sbi->ll_proc_root = NULL;
 361         }
 362
 363         obd_disconnect(sbi->ll_md_exp, 0);
 364
 365         // We do this to get rid of orphaned dentries. That is not really trw.
 366         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 367                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 368                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 369                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 370                 shrink_dcache_parent(dentry);
 371         }
 372         EXIT;
 373 }
 374
 375 char *ll_read_opt(const char *opt, char *data)
 376 {
 377         char *value;
 378         char *retval;
 379         ENTRY;
 380
 381         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 382         if (strncmp(opt, data, strlen(opt)))
 383                 RETURN(NULL);
 384         if ((value = strchr(data, '=')) == NULL)
 385                 RETURN(NULL);
 386
 387         value++;
 388         OBD_ALLOC(retval, strlen(value) + 1);
 389         if (!retval) {
 390                 CERROR("out of memory!\n");
 391                 RETURN(NULL);
 392         }
 393
 394         memcpy(retval, value, strlen(value)+1);
 395         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 396         RETURN(retval);
 397 }
 398
 399 int ll_set_opt(const char *opt, char *data, int fl)
 400 {
 401         ENTRY;
 402
 403         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 404         if (strncmp(opt, data, strlen(opt)))
 405                 RETURN(0);
 406         else
 407                 RETURN(fl);
 408 }
 409
 410 void ll_options(char *options, char **lov, char **lmv, char **sec,
 411                 int *async, int *flags)
 412 {
 413         char *this_char;
 414 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 415         char *opt_ptr = options;
 416 #endif
 417         ENTRY;
 418
 419         if (!options) {
 420                 EXIT;
 421                 return;
 422         }
 423
 424         *async = 0;
 425 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 426         for (this_char = strtok (options, ",");
 427              this_char != NULL;
 428              this_char = strtok (NULL, ",")) {
 429 #else
 430         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 431 #endif
 432                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 433                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 434                         continue;
 435                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 436                         continue;
 437                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 438                         *async = 1;
 439                         continue;
 440                 }
 441                 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
 442                         continue;
 443                 if (!(*flags & LL_SBI_NOLCK) &&
 444                     ((*flags) = (*flags) |
 445                                 ll_set_opt("nolock", this_char,
 446                                            LL_SBI_NOLCK)))
 447                         continue;
 448         }
 449
 450         EXIT;
 451 }
 452
 453 void ll_lli_init(struct ll_inode_info *lli)
 454 {
 455         sema_init(&lli->lli_open_sem, 1);
 456         sema_init(&lli->lli_size_sem, 1);
 457         lli->lli_flags = 0;
 458         lli->lli_size_pid = 0;
 459         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 460         spin_lock_init(&lli->lli_lock);
 461         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 462         lli->lli_inode_magic = LLI_INODE_MAGIC;
 463         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 464         sema_init(&lli->lli_och_sem, 1);
 465         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 466         lli->lli_mds_exec_och = NULL;
 467         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 468         lli->lli_open_fd_exec_count = 0;
 469 }
 470
 471 int ll_fill_super(struct super_block *sb, void *data, int silent)
 472 {
 473         struct ll_sb_info *sbi;
 474         char *lov = NULL;
 475         char *lmv = NULL;
 476         int async, err;
 477         char *sec = NULL;
 478         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 479         __u64 remote_flag = 0;
 480         ENTRY;
 481
 482         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 483
 484         sbi = lustre_init_sbi(sb);
 485         if (!sbi)
 486                 RETURN(-ENOMEM);
 487
 488         sbi->ll_flags |= LL_SBI_READAHEAD;
 489         ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
 490
 491         if (!lov) {
 492                 CERROR("no osc\n");
 493                 GOTO(out, err = -EINVAL);
 494         }
 495
 496         if (!lmv) {
 497                 CERROR("no mdc\n");
 498                 GOTO(out, err = -EINVAL);
 499         }
 500
 501         err = lustre_common_fill_super(sb, lmv, lov, async, sec,
 502                                        nllu, &remote_flag);
 503         EXIT;
 504 out:
 505         if (err)
 506                 lustre_free_sbi(sb);
 507
 508         if (sec)
 509                 OBD_FREE(sec, strlen(sec) + 1);
 510         if (lmv)
 511                 OBD_FREE(lmv, strlen(lmv) + 1);
 512         if (lov)
 513                 OBD_FREE(lov, strlen(lov) + 1);
 514         return err;
 515 } /* ll_read_super */
 516
 517 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 518                               struct config_llog_instance *cfg, int allow_recov)
 519 {
 520         struct lustre_cfg *lcfg = NULL;
 521         struct lustre_cfg_bufs bufs;
 522         struct portals_cfg pcfg;
 523         char *peer = "MDS_PEER_UUID";
 524         struct obd_device *obd;
 525         struct lustre_handle md_conn = {0, };
 526         struct obd_export *exp;
 527         char *name = "mdc_dev";
 528         class_uuid_t uuid;
 529         struct obd_uuid lmv_uuid;
 530         struct llog_ctxt *ctxt;
 531         int rc = 0, err = 0;
 532         ENTRY;
 533
 534         if (lmd_bad_magic(lmd))
 535                 RETURN(-EINVAL);
 536
 537         generate_random_uuid(uuid);
 538         class_uuid_unparse(uuid, &lmv_uuid);
 539
 540         if (lmd->lmd_local_nid) {
 541                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 542                 pcfg.pcfg_nal = lmd->lmd_nal;
 543                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 544                 rc = libcfs_nal_cmd(&pcfg);
 545                 if (rc < 0)
 546                         GOTO(out, rc);
 547         }
 548
 549         if (lmd->lmd_nal == SOCKNAL ||
 550             lmd->lmd_nal == OPENIBNAL ||
 551             lmd->lmd_nal == IIBNAL ||
 552             lmd->lmd_nal == VIBNAL ||
 553             lmd->lmd_nal == RANAL) {
 554                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 555                 pcfg.pcfg_nal     = lmd->lmd_nal;
 556                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 557                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 558                 pcfg.pcfg_misc    = lmd->lmd_port;
 559                 rc = libcfs_nal_cmd(&pcfg);
 560                 if (rc < 0)
 561                         GOTO(out, rc);
 562         }
 563         lustre_cfg_bufs_reset(&bufs, name);
 564         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 565
 566         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 567         lcfg->lcfg_nal = lmd->lmd_nal;
 568         lcfg->lcfg_nid = lmd->lmd_server_nid;
 569         LASSERT(lcfg->lcfg_nal);
 570         LASSERT(lcfg->lcfg_nid);
 571         err = class_process_config(lcfg);
 572         lustre_cfg_free(lcfg);
 573         if (err < 0)
 574                 GOTO(out_del_conn, err);
 575
 576         lustre_cfg_bufs_reset(&bufs, name);
 577         lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MDC_NAME);
 578         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 579
 580         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 581         err = class_process_config(lcfg);
 582         lustre_cfg_free(lcfg);
 583         if (err < 0)
 584                 GOTO(out_del_uuid, err);
 585
 586         lustre_cfg_bufs_reset(&bufs, name);
 587         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 588         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 589
 590         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 591         err = class_process_config(lcfg);
 592         lustre_cfg_free(lcfg);
 593         if (err < 0)
 594                 GOTO(out_detach, err);
 595
 596         obd = class_name2obd(name);
 597         if (obd == NULL)
 598                 GOTO(out_cleanup, rc = -EINVAL);
 599
 600         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 601                           strlen(lmd->lmd_security), lmd->lmd_security);
 602         if (rc)
 603                 GOTO(out_cleanup, rc);
 604
 605         /* Disable initial recovery on this import */
 606         rc = obd_set_info(obd->obd_self_export,
 607                           strlen("initial_recov"), "initial_recov",
 608                           sizeof(allow_recov), &allow_recov);
 609         if (rc)
 610                 GOTO(out_cleanup, rc);
 611
 612         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 613         if (rc) {
 614                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 615                 GOTO(out_cleanup, rc);
 616         }
 617
 618         exp = class_conn2export(&md_conn);
 619
 620         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 621         rc = class_config_process_llog(ctxt, profile, cfg);
 622         if (rc)
 623                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 624
 625         err = obd_disconnect(exp, 0);
 626
 627         EXIT;
 628 out_cleanup:
 629         lustre_cfg_bufs_reset(&bufs, name);
 630         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 631         err = class_process_config(lcfg);
 632         lustre_cfg_free(lcfg);
 633         if (err < 0)
 634                 GOTO(out, err);
 635 out_detach:
 636         lustre_cfg_bufs_reset(&bufs, name);
 637         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 638         err = class_process_config(lcfg);
 639         lustre_cfg_free(lcfg);
 640         if (err < 0)
 641                 GOTO(out, err);
 642
 643 out_del_uuid:
 644         lustre_cfg_bufs_reset(&bufs, name);
 645         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 646         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 647         err = class_process_config(lcfg);
 648         lustre_cfg_free(lcfg);
 649
 650 out_del_conn:
 651         if (lmd->lmd_nal == SOCKNAL ||
 652             lmd->lmd_nal == OPENIBNAL ||
 653             lmd->lmd_nal == IIBNAL ||
 654             lmd->lmd_nal == VIBNAL ||
 655             lmd->lmd_nal == RANAL) {
 656                 int err2;
 657
 658                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 659                 pcfg.pcfg_nal     = lmd->lmd_nal;
 660                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 661                 pcfg.pcfg_flags   = 1;          /* single_share */
 662                 err2 = libcfs_nal_cmd(&pcfg);
 663                 if (err2 && !err)
 664                         err = err2;
 665                 if (err < 0)
 666                         GOTO(out, err);
 667         }
 668 out:
 669         if (rc == 0)
 670                 rc = err;
 671
 672         return rc;
 673 }
 674
 675 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 676 {
 677         struct lustre_cfg *lcfg;
 678         struct lustre_cfg_bufs bufs;
 679         struct obd_device *obd;
 680         int next = 0;
 681
 682         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 683         {
 684                 int err;
 685
 686                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 687                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 688                 err = class_process_config(lcfg);
 689                 if (err) {
 690                         CERROR("cleanup failed: %s\n", obd->obd_name);
 691                         //continue;
 692                 }
 693
 694                 lcfg->lcfg_command = LCFG_DETACH;
 695                 err = class_process_config(lcfg);
 696                 lustre_cfg_free(lcfg);
 697                 if (err) {
 698                         CERROR("detach failed: %s\n", obd->obd_name);
 699                         //continue;
 700                 }
 701         }
 702
 703         if (sbi->ll_lmd != NULL)
 704                 class_del_profile(sbi->ll_lmd->lmd_profile);
 705 }
 706
 707 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 708 {
 709         struct lustre_mount_data * lmd = data;
 710         char *lov = NULL, *lmv = NULL;
 711         struct ll_sb_info *sbi;
 712         int err;
 713         ENTRY;
 714
 715         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 716         if (lmd_bad_magic(lmd))
 717                 RETURN(-EINVAL);
 718
 719         sbi = lustre_init_sbi(sb);
 720         if (!sbi)
 721                 RETURN(-ENOMEM);
 722
 723         sbi->ll_flags |= LL_SBI_READAHEAD;
 724
 725         if (lmd->lmd_profile) {
 726                 struct lustre_profile *lprof;
 727                 struct config_llog_instance cfg;
 728                 int len;
 729
 730                 if (lmd->lmd_mds[0] == '\0') {
 731                         CERROR("no mds name\n");
 732                         GOTO(out_free, err = -EINVAL);
 733                 }
 734                 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
 735
 736                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 737                 if (sbi->ll_lmd == NULL)
 738                         GOTO(out_free, err = -ENOMEM);
 739                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 740
 741                 /* generate a string unique to this super, let's try
 742                  the address of the super itself.*/
 743                 len = (sizeof(sb) * 2) + 1;
 744                 OBD_ALLOC(sbi->ll_instance, len);
 745                 if (sbi->ll_instance == NULL)
 746                         GOTO(out_free, err = -ENOMEM);
 747                 sprintf(sbi->ll_instance, "%p", sb);
 748
 749                 cfg.cfg_instance = sbi->ll_instance;
 750                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 751                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 752                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 753                 if (err < 0) {
 754                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 755                         GOTO(out_free, err);
 756                 }
 757
 758                 lprof = class_get_profile(lmd->lmd_profile);
 759                 if (lprof == NULL) {
 760                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 761                         GOTO(out_free, err = -EINVAL);
 762                 }
 763                 if (lov)
 764                         OBD_FREE(lov, strlen(lov) + 1);
 765                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
 766                           strlen(sbi->ll_instance) + 2);
 767                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 768
 769                 if (lmv)
 770                         OBD_FREE(lmv, strlen(lmv) + 1);
 771                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
 772                           strlen(sbi->ll_instance) + 2);
 773                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 774         }
 775
 776         if (!lov) {
 777                 CERROR("no osc\n");
 778                 GOTO(out_free, err = -EINVAL);
 779         }
 780
 781         if (!lmv) {
 782                 CERROR("no mdc\n");
 783                 GOTO(out_free, err = -EINVAL);
 784         }
 785
 786         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
 787                                        lmd->lmd_security, &lmd->lmd_nllu,
 788                                        &lmd->lmd_remote_flag);
 789
 790         if (err)
 791                 GOTO(out_free, err);
 792
 793 out_dev:
 794         if (lmv)
 795                 OBD_FREE(lmv, strlen(lmv) + 1);
 796         if (lov)
 797                 OBD_FREE(lov, strlen(lov) + 1);
 798
 799         RETURN(err);
 800
 801 out_free:
 802         if (sbi->ll_lmd) {
 803                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 804                 int err;
 805
 806                 if (sbi->ll_instance != NULL) {
 807                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 808                         struct config_llog_instance cfg;
 809                         char *cl_prof;
 810
 811                         cfg.cfg_instance = sbi->ll_instance;
 812                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 813
 814                         OBD_ALLOC(cl_prof, len);
 815                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 816                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 817                         if (err < 0) {
 818                                 CERROR("Unable to process log: %s\n", cl_prof);
 819                                 lustre_manual_cleanup(sbi);
 820                         }
 821                         OBD_FREE(cl_prof, len);
 822                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 823                 }
 824                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 825         }
 826         lustre_free_sbi(sb);
 827         goto out_dev;
 828 } /* lustre_fill_super */
 829
 830 void lustre_put_super(struct super_block *sb)
 831 {
 832         struct obd_device *obd;
 833         struct ll_sb_info *sbi = ll_s2sbi(sb);
 834         int force_umount = 0;
 835         ENTRY;
 836
 837         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 838         obd = class_exp2obd(sbi->ll_md_exp);
 839         if (obd)
 840                 force_umount = obd->obd_no_recov;
 841         obd = NULL;
 842
 843         lustre_common_put_super(sb);
 844         if (sbi->ll_lmd != NULL) {
 845                 char *cl_prof;
 846                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 847                 int err;
 848                 struct config_llog_instance cfg;
 849
 850                 if (force_umount) {
 851                         CERROR("force umount, doing manual cleanup\n");
 852                         lustre_manual_cleanup(sbi);
 853                         GOTO(free_lmd, 0);
 854                 }
 855
 856                 cfg.cfg_instance = sbi->ll_instance;
 857                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 858
 859                 OBD_ALLOC(cl_prof, len);
 860                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 861                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
 862                 if (err < 0) {
 863                         CERROR("Unable to process log: %s, doing manual cleanup"
 864                                "\n", cl_prof);
 865                         lustre_manual_cleanup(sbi);
 866                 }
 867
 868                 OBD_FREE(cl_prof, len);
 869         free_lmd:
 870                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 871                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 872         }
 873
 874         lustre_free_sbi(sb);
 875
 876         EXIT;
 877 } /* lustre_put_super */
 878
 879 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 880 {
 881         struct lustre_mount_data *lmd = sbi->ll_lmd;
 882         char *profile = lmd->lmd_profile, *name = NULL;
 883         struct config_llog_instance cfg;
 884         int rc, namelen =  0, version;
 885         struct llog_ctxt *ctxt;
 886         ENTRY;
 887
 888         if (profile == NULL)
 889                 RETURN(0);
 890         if (lmd == NULL) {
 891                 CERROR("Client not mounted with zero-conf; cannot "
 892                        "process update log.\n");
 893                 RETURN(0);
 894         }
 895
 896         rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
 897                                LDLM_FL_CONFIG_CHANGE, NULL);
 898         if (rc != 0)
 899                 CWARN("obd_cancel_unused(mdc): %d\n", rc);
 900
 901         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
 902                                LDLM_FL_CONFIG_CHANGE, NULL);
 903         if (rc != 0)
 904                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 905
 906         cfg.cfg_instance = sbi->ll_instance;
 907         cfg.cfg_uuid = sbi->ll_sb_uuid;
 908         cfg.cfg_local_nid = lmd->lmd_local_nid;
 909
 910         namelen = strlen(profile) + 20; /* -clean-######### */
 911         OBD_ALLOC(name, namelen);
 912         if (name == NULL)
 913                 RETURN(-ENOMEM);
 914
 915         if (clean) {
 916                 version = sbi->ll_config_version - 1;
 917                 sprintf(name, "%s-clean-%d", profile, version);
 918         } else {
 919                 version = sbi->ll_config_version + 1;
 920                 sprintf(name, "%s-%d", profile, version);
 921         }
 922
 923         CWARN("Applying configuration log %s\n", name);
 924
 925         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
 926                                 LLOG_CONFIG_REPL_CTXT);
 927         rc = class_config_process_llog(ctxt, name, &cfg);
 928         if (rc == 0)
 929                 sbi->ll_config_version = version;
 930         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 931
 932         if (rc == 0 && clean == 0) {
 933                 struct lov_desc desc;
 934                 __u32 valsize;
 935                 int rc = 0;
 936
 937                 valsize = sizeof(desc);
 938                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 939                                   "lovdesc", &valsize, &desc);
 940
 941                 rc = obd_init_ea_size(sbi->ll_md_exp,
 942                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
 943                                       (desc.ld_tgt_count *
 944                                        sizeof(struct llog_cookie)));
 945         }
 946         OBD_FREE(name, namelen);
 947         RETURN(rc);
 948 }
 949
 950 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 951 {
 952         struct inode *inode = NULL;
 953         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 954         if (lock->l_ast_data) {
 955                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 956                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 957                         inode = igrab(lock->l_ast_data);
 958                 } else {
 959                         inode = lock->l_ast_data;
 960                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
 961                                "l_ast_data %p is bogus: magic %0x8\n",
 962                                lock->l_ast_data, lli->lli_inode_magic);
 963                         inode = NULL;
 964                 }
 965         }
 966         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 967         return inode;
 968 }
 969
 970 int null_if_equal(struct ldlm_lock *lock, void *data)
 971 {
 972         if (data == lock->l_ast_data) {
 973                 lock->l_ast_data = NULL;
 974
 975                 if (lock->l_req_mode != lock->l_granted_mode)
 976                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 977         }
 978
 979         return LDLM_ITER_CONTINUE;
 980 }
 981
 982 void ll_clear_inode(struct inode *inode)
 983 {
 984         struct lustre_id id;
 985         struct ll_inode_info *lli = ll_i2info(inode);
 986         struct ll_sb_info *sbi = ll_i2sbi(inode);
 987         ENTRY;
 988
 989         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 990                inode->i_generation, inode);
 991
 992         lli->lli_inode_magic = LLI_INODE_DEAD;
 993         ll_inode2id(&id, inode);
 994
 995         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 996         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
 997
 998         LASSERT(!lli->lli_open_fd_write_count);
 999         LASSERT(!lli->lli_open_fd_read_count);
1000         LASSERT(!lli->lli_open_fd_exec_count);
1001         if (lli->lli_mds_write_och)
1002                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1003         if (lli->lli_mds_exec_och)
1004                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1005         if (lli->lli_mds_read_och)
1006                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1007         if (lli->lli_smd)
1008                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1009                                   null_if_equal, inode);
1010
1011         if (lli->lli_smd) {
1012                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1013                 lli->lli_smd = NULL;
1014         }
1015
1016         if (lli->lli_mea) {
1017                 obd_free_memmd(sbi->ll_md_exp,
1018                                (struct lov_stripe_md **) &lli->lli_mea);
1019                 lli->lli_mea = NULL;
1020         }
1021
1022         if (lli->lli_symlink_name) {
1023                 OBD_FREE(lli->lli_symlink_name,
1024                          strlen(lli->lli_symlink_name) + 1);
1025                 lli->lli_symlink_name = NULL;
1026         }
1027         lli->lli_inode_magic = LLI_INODE_DEAD;
1028
1029         EXIT;
1030 }
1031
1032 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1033  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1034  * keep these values until such a time that objects are allocated for it.
1035  * We do the MDS operations first, as it is checking permissions for us.
1036  * We don't to the MDS RPC if there is nothing that we want to store there,
1037  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1038  * going to do an RPC anyways.
1039  *
1040  * If we are doing a truncate, we will send the mtime and ctime updates
1041  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1042  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1043  * at the same time.
1044  */
1045 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1046 {
1047         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1048         struct ll_sb_info *sbi = ll_i2sbi(inode);
1049         struct ptlrpc_request *request = NULL;
1050         struct mdc_op_data *op_data;
1051         int ia_valid = attr->ia_valid;
1052         int err, rc = 0;
1053         ENTRY;
1054
1055         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1056         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1057
1058         if (ia_valid & ATTR_SIZE) {
1059                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1060                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1061                                attr->ia_size, ll_file_maxbytes(inode));
1062                         RETURN(-EFBIG);
1063                 }
1064
1065                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1066         }
1067
1068         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1069         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1070                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1071                         RETURN(-EPERM);
1072         }
1073
1074         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1075         if (attr->ia_valid & ATTR_CTIME) {
1076                 attr->ia_ctime = CURRENT_TIME;
1077                 attr->ia_valid |= ATTR_CTIME_SET;
1078         }
1079         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1080                 attr->ia_atime = CURRENT_TIME;
1081                 attr->ia_valid |= ATTR_ATIME_SET;
1082         }
1083         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1084                 attr->ia_mtime = CURRENT_TIME;
1085                 attr->ia_valid |= ATTR_MTIME_SET;
1086         }
1087
1088         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1089                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1090                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1091                        LTIME_S(CURRENT_TIME));
1092
1093         if (lsm)
1094                 attr->ia_valid &= ~ATTR_SIZE;
1095
1096         /* If only OST attributes being set on objects, don't do MDS RPC.
1097          * In that case, we need to check permissions and update the local
1098          * inode ourselves so we can call obdo_from_inode() always. */
1099         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1100                 struct lustre_md md;
1101
1102                 OBD_ALLOC(op_data, sizeof(*op_data));
1103                 if (op_data == NULL)
1104                         RETURN(-ENOMEM);
1105                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1106
1107                 rc = md_setattr(sbi->ll_md_exp, op_data,
1108                                 attr, NULL, 0, NULL, 0, &request);
1109                 OBD_FREE(op_data, sizeof(*op_data));
1110                 if (rc) {
1111                         ptlrpc_req_finished(request);
1112                         if (rc != -EPERM && rc != -EACCES)
1113                                 CERROR("md_setattr fails: rc = %d\n", rc);
1114                         RETURN(rc);
1115                 }
1116
1117                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1118                                        sbi->ll_dt_exp, &md);
1119                 if (rc) {
1120                         ptlrpc_req_finished(request);
1121                         RETURN(rc);
1122                 }
1123
1124                 /* We call inode_setattr to adjust timestamps, but we first
1125                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1126                  *
1127                  * NB: ATTR_SIZE will only be set at this point if the size
1128                  * resides on the MDS, ie, this file has no objects. */
1129                 attr->ia_valid &= ~ATTR_SIZE;
1130
1131                 /*
1132                  * assigning inode_setattr() to @err to disable warning that
1133                  * function's result should be checked by by caller. error is
1134                  * impossible here, as vmtruncate() control path is disabled.
1135                  */
1136                 err = inode_setattr(inode, attr);
1137                 ll_update_inode(inode, &md);
1138                 ptlrpc_req_finished(request);
1139
1140                 if (!lsm || !S_ISREG(inode->i_mode)) {
1141                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1142                         RETURN(0);
1143                 }
1144         } else {
1145                 /* The OST doesn't check permissions, but the alternative is
1146                  * a gratuitous RPC to the MDS.  We already rely on the client
1147                  * to do read/write/truncate permission checks, so is mtime OK?
1148                  */
1149                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1150                         /* from sys_utime() */
1151                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1152                                 if (current->fsuid != inode->i_uid &&
1153                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1154                                         RETURN(rc);
1155                         } else {
1156                                 /* from inode_change_ok() */
1157                                 if (current->fsuid != inode->i_uid &&
1158                                     !capable(CAP_FOWNER))
1159                                         RETURN(-EPERM);
1160                         }
1161                 }
1162
1163                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1164                 err = inode_setattr(inode, attr);
1165                 /*
1166                  * assigning inode_setattr() to @err to disable warning that
1167                  * function's result should be checked by by caller. error is
1168                  * impossible here, as vmtruncate() control path is disabled.
1169                  */
1170         }
1171
1172         /* We really need to get our PW lock before we change inode->i_size.
1173          * If we don't we can race with other i_size updaters on our node, like
1174          * ll_file_read.  We can also race with i_size propogation to other
1175          * nodes through dirtying and writeback of final cached pages.  This
1176          * last one is especially bad for racing o_append users on other
1177          * nodes. */
1178         if (ia_valid & ATTR_SIZE) {
1179                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1180                                                            OBD_OBJECT_EOF } };
1181                 struct lustre_handle lockh = { 0 };
1182                 struct ll_inode_info *lli = ll_i2info(inode);
1183                 int err, ast_flags = 0;
1184                 /* XXX when we fix the AST intents to pass the discard-range
1185                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1186                  * XXX here. */
1187                 if (attr->ia_size == 0)
1188                         ast_flags = LDLM_AST_DISCARD_DATA;
1189
1190                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1191                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1192
1193                 if (rc != 0)
1194                         RETURN(rc);
1195
1196                 down(&lli->lli_size_sem);
1197                 lli->lli_size_pid = current->pid;
1198                 rc = vmtruncate(inode, attr->ia_size);
1199                 if (rc != 0) {
1200                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1201                         lli->lli_size_pid = 0;
1202                         up(&lli->lli_size_sem);
1203                 }
1204
1205                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1206                 if (err) {
1207                         CERROR("ll_extent_unlock failed: %d\n", err);
1208                         if (!rc)
1209                                 rc = err;
1210                 }
1211         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1212                 struct obdo *oa = NULL;
1213
1214                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1215                        inode->i_ino, LTIME_S(attr->ia_mtime));
1216
1217                 oa = obdo_alloc();
1218                 if (oa == NULL)
1219                         RETURN(-ENOMEM);
1220
1221                 oa->o_id = lsm->lsm_object_id;
1222                 oa->o_gr = lsm->lsm_object_gr;
1223                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1224                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1225                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1226                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1227                 obdo_free(oa);
1228                 if (rc)
1229                         CERROR("obd_setattr fails: rc = %d\n", rc);
1230         }
1231         RETURN(rc);
1232 }
1233
1234 int ll_setattr(struct dentry *de, struct iattr *attr)
1235 {
1236         LASSERT(de->d_inode);
1237         return ll_setattr_raw(de->d_inode, attr);
1238 }
1239
1240 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1241                        unsigned long max_age)
1242 {
1243         struct ll_sb_info *sbi = ll_s2sbi(sb);
1244         struct obd_statfs obd_osfs;
1245         int rc;
1246         ENTRY;
1247
1248         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1249         if (rc) {
1250                 CERROR("obd_statfs fails: rc = %d\n", rc);
1251                 RETURN(rc);
1252         }
1253
1254         osfs->os_type = sb->s_magic;
1255
1256         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1257                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1258
1259         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1260         if (rc) {
1261                 CERROR("obd_statfs fails: rc = %d\n", rc);
1262                 RETURN(rc);
1263         }
1264
1265         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1266                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1267                obd_osfs.os_files);
1268
1269         osfs->os_blocks = obd_osfs.os_blocks;
1270         osfs->os_bfree = obd_osfs.os_bfree;
1271         osfs->os_bavail = obd_osfs.os_bavail;
1272
1273         /* If we don't have as many objects free on the OST as inodes
1274          * on the MDS, we reduce the total number of inodes to
1275          * compensate, so that the "inodes in use" number is correct.
1276          */
1277         if (obd_osfs.os_ffree < osfs->os_ffree) {
1278                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1279                         obd_osfs.os_ffree;
1280                 osfs->os_ffree = obd_osfs.os_ffree;
1281         }
1282
1283         RETURN(rc);
1284 }
1285
1286 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1287 {
1288         struct obd_statfs osfs;
1289         int rc;
1290
1291         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1292         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1293
1294         /* For now we will always get up-to-date statfs values, but in the
1295          * future we may allow some amount of caching on the client (e.g.
1296          * from QOS or lprocfs updates). */
1297         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1298         if (rc)
1299                 return rc;
1300
1301         statfs_unpack(sfs, &osfs);
1302
1303         if (sizeof(sfs->f_blocks) == 4) {
1304                 while (osfs.os_blocks > ~0UL) {
1305                         sfs->f_bsize <<= 1;
1306
1307                         osfs.os_blocks >>= 1;
1308                         osfs.os_bfree >>= 1;
1309                         osfs.os_bavail >>= 1;
1310                 }
1311         }
1312
1313         sfs->f_blocks = osfs.os_blocks;
1314         sfs->f_bfree = osfs.os_bfree;
1315         sfs->f_bavail = osfs.os_bavail;
1316
1317         return 0;
1318 }
1319
1320 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1321 {
1322         struct ll_inode_info *lli = ll_i2info(inode);
1323         struct lov_stripe_md *lsm = md->lsm;
1324         struct mds_body *body = md->body;
1325         struct mea *mea = md->mea;
1326         struct posix_acl *ll_acl_access = md->acl_access;
1327         ENTRY;
1328
1329         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1330
1331         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1332                 /* check for default striping info for dir. */
1333                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1334         }
1335
1336         if (lsm != NULL) {
1337                 LASSERT(lsm->lsm_object_gr > 0);
1338                 if (lli->lli_smd == NULL) {
1339                         lli->lli_smd = lsm;
1340                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1341                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1342                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1343                 } else {
1344                         int i;
1345                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1346                                 CERROR("lsm mismatch for inode %ld\n",
1347                                        inode->i_ino);
1348                                 CERROR("lli_smd:\n");
1349                                 dump_lsm(D_ERROR, lli->lli_smd);
1350                                 CERROR("lsm:\n");
1351                                 dump_lsm(D_ERROR, lsm);
1352                                 LBUG();
1353                         }
1354                         /* XXX FIXME -- We should decide on a safer (atomic) and
1355                          * more elegant way to update the lsm */
1356                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1357                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1358                                         lsm->lsm_oinfo[i].loi_id;
1359                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1360                                         lsm->lsm_oinfo[i].loi_gr;
1361                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1362                                         lsm->lsm_oinfo[i].loi_ost_idx;
1363                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1364                                         lsm->lsm_oinfo[i].loi_ost_gen;
1365                         }
1366                 }
1367                 /* bug 2844 - limit i_blksize for broken user-space apps */
1368                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1369                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1370                 if (lli->lli_smd != lsm)
1371                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1372         }
1373
1374         if (mea != NULL) {
1375                 if (lli->lli_mea == NULL) {
1376                         lli->lli_mea = mea;
1377                 } else {
1378                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1379                                 CERROR("mea mismatch for inode %lu\n",
1380                                         inode->i_ino);
1381                                 LBUG();
1382                         }
1383                 }
1384                 if (lli->lli_mea != mea)
1385                         obd_free_memmd(ll_i2mdexp(inode),
1386                                        (struct lov_stripe_md **) &mea);
1387         }
1388
1389         if (body->valid & OBD_MD_FID)
1390                 id_assign_fid(&lli->lli_id, &body->id1);
1391
1392         if (body->valid & OBD_MD_FLID)
1393                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1394
1395         if (body->valid & OBD_MD_FLGENER)
1396                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1397
1398         spin_lock(&lli->lli_lock);
1399         if (ll_acl_access != NULL) {
1400                 if (lli->lli_acl_access != NULL)
1401                         posix_acl_release(lli->lli_acl_access);
1402                 lli->lli_acl_access = ll_acl_access;
1403         }
1404         spin_unlock(&lli->lli_lock);
1405
1406         if (body->valid & OBD_MD_FLID)
1407                 inode->i_ino = id_ino(&body->id1);
1408         if (body->valid & OBD_MD_FLGENER)
1409                 inode->i_generation = id_gen(&body->id1);
1410         if (body->valid & OBD_MD_FLATIME)
1411                 LTIME_S(inode->i_atime) = body->atime;
1412         if (body->valid & OBD_MD_FLMTIME &&
1413             body->mtime > LTIME_S(inode->i_mtime)) {
1414                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1415                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1416                 LTIME_S(inode->i_mtime) = body->mtime;
1417         }
1418         if (body->valid & OBD_MD_FLCTIME &&
1419             body->ctime > LTIME_S(inode->i_ctime))
1420                 LTIME_S(inode->i_ctime) = body->ctime;
1421         if (body->valid & OBD_MD_FLMODE) {
1422                 inode->i_mode = (inode->i_mode & S_IFMT) |
1423                         (body->mode & ~S_IFMT);
1424         }
1425         if (body->valid & OBD_MD_FLTYPE) {
1426                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1427                         (body->mode & S_IFMT);
1428         }
1429         if (body->valid & OBD_MD_FLUID)
1430                 inode->i_uid = body->uid;
1431         if (body->valid & OBD_MD_FLGID)
1432                 inode->i_gid = body->gid;
1433         if (body->valid & OBD_MD_FLFLAGS)
1434                 inode->i_flags = body->flags;
1435         if (body->valid & OBD_MD_FLNLINK)
1436                 inode->i_nlink = body->nlink;
1437         if (body->valid & OBD_MD_FLRDEV)
1438 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1439                 inode->i_rdev = body->rdev;
1440 #else
1441                 inode->i_rdev = old_decode_dev(body->rdev);
1442 #endif
1443         if (body->valid & OBD_MD_FLSIZE)
1444                 inode->i_size = body->size;
1445         if (body->valid & OBD_MD_FLBLOCKS)
1446                 inode->i_blocks = body->blocks;
1447
1448         if (body->valid & OBD_MD_FLSIZE)
1449                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1450
1451 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1452         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1453 #endif
1454         LASSERT(id_fid(&lli->lli_id) != 0);
1455 }
1456
1457 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1458 static struct backing_dev_info ll_backing_dev_info = {
1459         .ra_pages       = 0,    /* No readahead */
1460         .memory_backed  = 0,    /* Does contribute to dirty memory */
1461 };
1462 #endif
1463
1464 void ll_read_inode2(struct inode *inode, void *opaque)
1465 {
1466         struct lustre_md *md = opaque;
1467         struct ll_inode_info *lli = ll_i2info(inode);
1468         ENTRY;
1469
1470         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1471                inode->i_generation, inode);
1472
1473         ll_lli_init(lli);
1474
1475         LASSERT(!lli->lli_smd);
1476
1477         /* Core attributes from the MDS first.  This is a new inode, and
1478          * the VFS doesn't zero times in the core inode so we have to do
1479          * it ourselves.  They will be overwritten by either MDS or OST
1480          * attributes - we just need to make sure they aren't newer. */
1481         LTIME_S(inode->i_mtime) = 0;
1482         LTIME_S(inode->i_atime) = 0;
1483         LTIME_S(inode->i_ctime) = 0;
1484
1485         inode->i_rdev = 0;
1486         ll_update_inode(inode, md);
1487
1488         /* OIDEBUG(inode); */
1489
1490         if (S_ISREG(inode->i_mode)) {
1491                 inode->i_op = &ll_file_inode_operations;
1492                 inode->i_fop = &ll_file_operations;
1493                 inode->i_mapping->a_ops = &ll_aops;
1494                 EXIT;
1495         } else if (S_ISDIR(inode->i_mode)) {
1496                 inode->i_op = &ll_dir_inode_operations;
1497                 inode->i_fop = &ll_dir_operations;
1498                 inode->i_mapping->a_ops = &ll_dir_aops;
1499                 EXIT;
1500         } else if (S_ISLNK(inode->i_mode)) {
1501                 inode->i_op = &ll_fast_symlink_inode_operations;
1502                 EXIT;
1503         } else {
1504                 inode->i_op = &ll_special_inode_operations;
1505
1506 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1507                 init_special_inode(inode, inode->i_mode,
1508                                    kdev_t_to_nr(inode->i_rdev));
1509
1510                 /* initializing backing dev info. */
1511                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1512 #else
1513                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1514 #endif
1515                 lli->ll_save_ifop = inode->i_fop;
1516
1517                 if (S_ISCHR(inode->i_mode))
1518                         inode->i_fop = &ll_special_chr_inode_fops;
1519                 else if (S_ISBLK(inode->i_mode))
1520                         inode->i_fop = &ll_special_blk_inode_fops;
1521                 else if (S_ISFIFO(inode->i_mode))
1522                         inode->i_fop = &ll_special_fifo_inode_fops;
1523                 else if (S_ISSOCK(inode->i_mode))
1524                         inode->i_fop = &ll_special_sock_inode_fops;
1525
1526                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1527                       inode->i_fop);
1528
1529                 if (lli->ll_save_ifop->owner) {
1530                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1531                               lli->ll_save_ifop->owner);
1532                 }
1533                 EXIT;
1534         }
1535 }
1536
1537 void ll_delete_inode(struct inode *inode)
1538 {
1539         struct ll_sb_info *sbi = ll_i2sbi(inode);
1540         struct lustre_id id;
1541         int rc;
1542         ENTRY;
1543
1544         ll_inode2id(&id, inode);
1545
1546         rc = md_delete_inode(sbi->ll_md_exp, &id);
1547         if (rc) {
1548                 CERROR("md_delete_inode() failed, error %d\n",
1549                        rc);
1550         }
1551
1552         clear_inode(inode);
1553         EXIT;
1554 }
1555
1556 int ll_iocontrol(struct inode *inode, struct file *file,
1557                  unsigned int cmd, unsigned long arg)
1558 {
1559         struct ll_sb_info *sbi = ll_i2sbi(inode);
1560         struct ptlrpc_request *req = NULL;
1561         int rc, flags = 0;
1562         ENTRY;
1563
1564         switch(cmd) {
1565         case EXT3_IOC_GETFLAGS: {
1566                 struct lustre_id id;
1567                 __u64 valid = OBD_MD_FLFLAGS;
1568                 struct mds_body *body;
1569
1570                 ll_inode2id(&id, inode);
1571                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1572                 if (rc) {
1573                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1574                         RETURN(-abs(rc));
1575                 }
1576
1577                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1578
1579                 if (body->flags & S_APPEND)
1580                         flags |= EXT3_APPEND_FL;
1581                 if (body->flags & S_IMMUTABLE)
1582                         flags |= EXT3_IMMUTABLE_FL;
1583                 if (body->flags & S_NOATIME)
1584                         flags |= EXT3_NOATIME_FL;
1585
1586                 ptlrpc_req_finished (req);
1587
1588                 RETURN(put_user(flags, (int *)arg));
1589         }
1590         case EXT3_IOC_SETFLAGS: {
1591                 struct mdc_op_data *op_data;
1592                 struct iattr attr;
1593                 struct obdo *oa;
1594                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1595
1596                 if (get_user(flags, (int *)arg))
1597                         RETURN(-EFAULT);
1598
1599                 oa = obdo_alloc();
1600                 if (!oa)
1601                         RETURN(-ENOMEM);
1602
1603                 OBD_ALLOC(op_data, sizeof(*op_data));
1604                 if (op_data == NULL) {
1605                         obdo_free(oa);
1606                         RETURN(-ENOMEM);
1607                 }
1608                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1609
1610                 memset(&attr, 0x0, sizeof(attr));
1611                 attr.ia_attr_flags = flags;
1612                 attr.ia_valid |= ATTR_ATTR_FLAG;
1613
1614                 rc = md_setattr(sbi->ll_md_exp, op_data,
1615                                 &attr, NULL, 0, NULL, 0, &req);
1616                 OBD_FREE(op_data, sizeof(*op_data));
1617                 if (rc) {
1618                         ptlrpc_req_finished(req);
1619                         if (rc != -EPERM && rc != -EACCES)
1620                                 CERROR("md_setattr fails: rc = %d\n", rc);
1621                         obdo_free(oa);
1622                         RETURN(rc);
1623                 }
1624                 ptlrpc_req_finished(req);
1625
1626                 oa->o_id = lsm->lsm_object_id;
1627                 oa->o_gr = lsm->lsm_object_gr;
1628                 oa->o_flags = flags;
1629                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1630
1631                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1632                 obdo_free(oa);
1633                 if (rc) {
1634                         if (rc != -EPERM && rc != -EACCES)
1635                                 CERROR("md_setattr fails: rc = %d\n", rc);
1636                         RETURN(rc);
1637                 }
1638
1639                 if (flags & EXT3_APPEND_FL)
1640                         inode->i_flags |= S_APPEND;
1641                 else
1642                         inode->i_flags &= ~S_APPEND;
1643                 if (flags & EXT3_IMMUTABLE_FL)
1644                         inode->i_flags |= S_IMMUTABLE;
1645                 else
1646                         inode->i_flags &= ~S_IMMUTABLE;
1647                 if (flags & EXT3_NOATIME_FL)
1648                         inode->i_flags |= S_NOATIME;
1649                 else
1650                         inode->i_flags &= ~S_NOATIME;
1651
1652                 RETURN(0);
1653         }
1654         default:
1655                 RETURN(-ENOSYS);
1656         }
1657
1658         RETURN(0);
1659 }
1660
1661 /* this is only called in the case of forced umount. */
1662 void ll_umount_begin(struct super_block *sb)
1663 {
1664         struct ll_sb_info *sbi = ll_s2sbi(sb);
1665         struct obd_ioctl_data ioc_data = { 0 };
1666         struct obd_device *obd;
1667         ENTRY;
1668
1669         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1670                sb->s_count, atomic_read(&sb->s_active));
1671
1672         obd = class_exp2obd(sbi->ll_md_exp);
1673         if (obd == NULL) {
1674                 CERROR("Invalid MDC connection handle "LPX64"\n",
1675                        sbi->ll_md_exp->exp_handle.h_cookie);
1676                 EXIT;
1677                 return;
1678         }
1679         obd->obd_no_recov = 1;
1680         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1681                       sizeof(ioc_data), &ioc_data, NULL);
1682
1683         obd = class_exp2obd(sbi->ll_dt_exp);
1684         if (obd == NULL) {
1685                 CERROR("Invalid LOV connection handle "LPX64"\n",
1686                        sbi->ll_dt_exp->exp_handle.h_cookie);
1687                 EXIT;
1688                 return;
1689         }
1690
1691         obd->obd_no_recov = 1;
1692         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1693                       sizeof(ioc_data), &ioc_data, NULL);
1694
1695         /*
1696          * really, we'd like to wait until there are no requests outstanding,
1697          * and then continue.  For now, we just invalidate the requests,
1698          * schedule, and hope.
1699          */
1700         schedule();
1701
1702         EXIT;
1703 }
1704
1705 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1706                   struct inode **inode, struct ptlrpc_request *req,
1707                   int offset, struct super_block *sb)
1708 {
1709         struct lustre_md md;
1710         int rc = 0;
1711
1712         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1713         if (rc)
1714                 RETURN(rc);
1715
1716         if (*inode) {
1717                 ll_update_inode(*inode, &md);
1718         } else {
1719                 LASSERT(sb);
1720                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1721                 if (*inode == NULL || is_bad_inode(*inode)) {
1722                         /* free the lsm if we allocated one above */
1723                         if (md.lsm != NULL)
1724                                 obd_free_memmd(dt_exp, &md.lsm);
1725                         if (md.mea != NULL)
1726                                 obd_free_memmd(md_exp,
1727                                                (struct lov_stripe_md**)&md.mea);
1728                         rc = -ENOMEM;
1729                         CERROR("new_inode -fatal: rc %d\n", rc);
1730                 }
1731         }
1732
1733         RETURN(rc);
1734 }
1735
1736 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1737                char *filename, struct lustre_id *ret)
1738 {
1739         struct ptlrpc_request *request = NULL;
1740         struct mds_body *body;
1741         int rc;
1742
1743         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1744                              OBD_MD_FID, 0, &request);
1745         if (rc < 0) {
1746                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1747                        filename, rc);
1748                 return rc;
1749         }
1750
1751         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1752         LASSERT(body != NULL);
1753         LASSERT_REPSWABBED(request, 0);
1754
1755         *ret = body->id1;
1756         ptlrpc_req_finished(request);
1757
1758         return rc;
1759 }