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         atomic_set(&sbi->ll_gns_enabled, 1);
  87         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  88         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  89
  90         sbi->ll_gns_timer.data = (unsigned long)sbi;
  91         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  92         init_timer(&sbi->ll_gns_timer);
  93
  94         ll_set_sbi(sb, sbi);
  95
  96         generate_random_uuid(uuid);
  97         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  98         RETURN(sbi);
  99 }
 100
 101 void lustre_free_sbi(struct super_block *sb)
 102 {
 103         struct ll_sb_info *sbi = ll_s2sbi(sb);
 104         ENTRY;
 105
 106         if (sbi != NULL) {
 107                 list_del(&sbi->ll_gns_sbi_head);
 108                 del_timer(&sbi->ll_gns_timer);
 109                 OBD_FREE(sbi, sizeof(*sbi));
 110         }
 111         ll_set_sbi(sb, NULL);
 112         EXIT;
 113 }
 114
 115 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 116 {
 117         int valsize, rc;
 118         ENTRY;
 119
 120         valsize = sizeof(sbi->ll_dt_desc);
 121         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 122         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 123                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 124         RETURN(rc);
 125 }
 126
 127 extern struct dentry_operations ll_d_ops;
 128
 129 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 130                              int async, char *security, __u32 *nllu,
 131                              __u64 *remote)
 132 {
 133         struct ll_sb_info *sbi = ll_s2sbi(sb);
 134         struct ptlrpc_request *request = NULL;
 135         struct lustre_handle dt_conn = {0, };
 136         struct lustre_handle md_conn = {0, };
 137         struct obd_connect_data *data;
 138         struct inode *root = NULL;
 139         struct obd_device *obd;
 140         struct obd_statfs osfs;
 141         struct lustre_md md;
 142         kdev_t devno;
 143         int err;
 144         __u32 valsize;
 145         ENTRY;
 146
 147         obd = class_name2obd(lmv);
 148         if (!obd) {
 149                 CERROR("MDC %s: not setup or attached\n", lmv);
 150                 RETURN(-EINVAL);
 151         }
 152         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 153                      sizeof(async), &async);
 154
 155         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 156             (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 157                 CERROR("wrong remote flag "LPX64"\n", *remote);
 158                 RETURN(-EINVAL);
 159         }
 160
 161         OBD_ALLOC(data, sizeof(*data));
 162         if (!data)
 163                 RETURN(-ENOMEM);
 164
 165         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 166                                               OBD_CONNECT_REMOTE);
 167         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 168
 169         if (security == NULL)
 170                 security = "null";
 171
 172         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 173                            strlen(security), security);
 174         if (err) {
 175                 CERROR("LMV %s: failed to set security %s, err %d\n",
 176                         lmv, security, err);
 177                 OBD_FREE(data, sizeof(*data));
 178                 RETURN(err);
 179         }
 180
 181         if (proc_lustre_fs_root) {
 182                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 183                                                   lov, lmv);
 184                 if (err < 0)
 185                         CERROR("could not register mount in /proc/lustre");
 186         }
 187
 188         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
 189                           OBD_OPT_REAL_CLIENT);
 190         if (err == -EBUSY) {
 191                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 192                        " client is not a part.  Please wait for recovery to "
 193                        "complete, abort, or time out.\n", lmv);
 194                 GOTO(out, err);
 195         } else if (err) {
 196                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 197                 GOTO(out, err);
 198         }
 199         sbi->ll_md_exp = class_conn2export(&md_conn);
 200         err = obd_statfs(obd, &osfs, jiffies - HZ);
 201         if (err)
 202                 GOTO(out_lmv, err);
 203
 204         if (!osfs.os_bsize) {
 205                 CERROR("Invalid block size is detected.");
 206                 GOTO(out_lmv, err);
 207         }
 208
 209         sb->s_magic = LL_SUPER_MAGIC;
 210         sb->s_blocksize = osfs.os_bsize;
 211         sb->s_blocksize_bits = log2(osfs.os_bsize);
 212         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 213
 214         devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 215                              strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 216
 217         sb->s_dev = devno;
 218
 219         /* after statfs, we are supposed to have connected to MDSs,
 220          * so it's ok to check remote flag returned.
 221          */
 222         valsize = sizeof(&sbi->ll_remote);
 223         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 224                            &valsize, &sbi->ll_remote);
 225         if (err) {
 226                 CERROR("fail to obtain remote flag\n");
 227                 GOTO(out, err);
 228         }
 229
 230         obd = class_name2obd(lov);
 231         if (!obd) {
 232                 CERROR("OSC %s: not setup or attached\n", lov);
 233                 GOTO(out_lmv, err);
 234         }
 235         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 236                      sizeof(async), &async);
 237
 238         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
 239                           OBD_OPT_REAL_CLIENT);
 240         if (err == -EBUSY) {
 241                 CERROR("An OST (lov %s) is performing recovery, of which this"
 242                        " client is not a part.  Please wait for recovery to "
 243                        "complete, abort, or time out.\n", lov);
 244                 GOTO(out, err);
 245         } else if (err) {
 246                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 247                 GOTO(out_lmv, err);
 248         }
 249         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 250
 251         err = lustre_init_dt_desc(sbi);
 252         if (err == 0) {
 253                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 254                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
 255                                  sbi->ll_dt_desc.ld_tgt_count *
 256                                  sizeof(struct llog_cookie));
 257         }
 258
 259         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 260         if (err) {
 261                 CERROR("cannot mds_connect: rc = %d\n", err);
 262                 GOTO(out_lov, err);
 263         }
 264         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 265
 266         sb->s_op = &lustre_super_operations;
 267
 268         /* make root inode
 269          * XXX: move this to after cbd setup? */
 270         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 271                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
 272                          0, &request);
 273         if (err) {
 274                 CERROR("md_getattr failed for root: rc = %d\n", err);
 275                 GOTO(out_lov, err);
 276         }
 277
 278         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 279                                 sbi->ll_dt_exp, &md);
 280         if (err) {
 281                 CERROR("failed to understand root inode md: rc = %d\n", err);
 282                 ptlrpc_req_finished(request);
 283                 GOTO(out_lov, err);
 284         }
 285
 286         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 287         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 288
 289         ptlrpc_req_finished(request);
 290
 291         if (root == NULL || is_bad_inode(root)) {
 292                 if (md.lsm != NULL)
 293                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 294                 if (md.mea != NULL)
 295                     obd_free_memmd(sbi->ll_md_exp,
 296                                    (struct lov_stripe_md**)&md.mea);
 297                 CERROR("lustre_lite: bad iget4 for root\n");
 298                 GOTO(out_root, err = -EBADF);
 299         }
 300
 301         err = ll_close_thread_start(&sbi->ll_lcq);
 302         if (err) {
 303                 CERROR("cannot start close thread: rc %d\n", err);
 304                 GOTO(out_root, err);
 305         }
 306
 307         ll_gns_add_timer(sbi);
 308
 309         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 310            backing dev info assigned to inode mapping is used for
 311            determining maximal readahead. */
 312 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 313     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 314         /* bug 2805 - set VM readahead to zero */
 315         vm_max_readahead = vm_min_readahead = 0;
 316 #endif
 317
 318         sb->s_root = d_alloc_root(root);
 319         sb->s_root->d_op = &ll_d_ops;
 320
 321         sb->s_flags |= MS_POSIXACL;
 322 #ifdef S_PDIROPS
 323         CWARN("Enabling PDIROPS\n");
 324         sb->s_flags |= S_PDIROPS;
 325 #endif
 326
 327         if (data != NULL)
 328                 OBD_FREE(data, sizeof(*data));
 329         RETURN(err);
 330 out_root:
 331         if (root)
 332                 iput(root);
 333 out_lov:
 334         obd_disconnect(sbi->ll_dt_exp, 0);
 335 out_lmv:
 336         obd_disconnect(sbi->ll_md_exp, 0);
 337 out:
 338         if (data != NULL)
 339                 OBD_FREE(data, sizeof(*data));
 340         lprocfs_unregister_mountpoint(sbi);
 341         return err;
 342 }
 343
 344 void lustre_common_put_super(struct super_block *sb)
 345 {
 346         struct ll_sb_info *sbi = ll_s2sbi(sb);
 347         struct hlist_node *tmp, *next;
 348         ENTRY;
 349
 350         ll_gns_del_timer(sbi);
 351         ll_close_thread_shutdown(sbi->ll_lcq);
 352
 353         list_del(&sbi->ll_conn_chain);
 354         obd_disconnect(sbi->ll_dt_exp, 0);
 355
 356         lprocfs_unregister_mountpoint(sbi);
 357         if (sbi->ll_proc_root) {
 358                 lprocfs_remove(sbi->ll_proc_root);
 359                 sbi->ll_proc_root = NULL;
 360         }
 361
 362         obd_disconnect(sbi->ll_md_exp, 0);
 363
 364         // We do this to get rid of orphaned dentries. That is not really trw.
 365         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 366                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 367                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 368                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 369                 shrink_dcache_parent(dentry);
 370         }
 371         EXIT;
 372 }
 373
 374 char *ll_read_opt(const char *opt, char *data)
 375 {
 376         char *value;
 377         char *retval;
 378         ENTRY;
 379
 380         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 381         if (strncmp(opt, data, strlen(opt)))
 382                 RETURN(NULL);
 383         if ((value = strchr(data, '=')) == NULL)
 384                 RETURN(NULL);
 385
 386         value++;
 387         OBD_ALLOC(retval, strlen(value) + 1);
 388         if (!retval) {
 389                 CERROR("out of memory!\n");
 390                 RETURN(NULL);
 391         }
 392
 393         memcpy(retval, value, strlen(value)+1);
 394         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 395         RETURN(retval);
 396 }
 397
 398 int ll_set_opt(const char *opt, char *data, int fl)
 399 {
 400         ENTRY;
 401
 402         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 403         if (strncmp(opt, data, strlen(opt)))
 404                 RETURN(0);
 405         else
 406                 RETURN(fl);
 407 }
 408
 409 void ll_options(char *options, char **lov, char **lmv, char **sec,
 410                 int *async, int *flags)
 411 {
 412         char *this_char;
 413 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 414         char *opt_ptr = options;
 415 #endif
 416         ENTRY;
 417
 418         if (!options) {
 419                 EXIT;
 420                 return;
 421         }
 422
 423         *async = 0;
 424 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 425         for (this_char = strtok (options, ",");
 426              this_char != NULL;
 427              this_char = strtok (NULL, ",")) {
 428 #else
 429         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 430 #endif
 431                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 432                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 433                         continue;
 434                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 435                         continue;
 436                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 437                         *async = 1;
 438                         continue;
 439                 }
 440                 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
 441                         continue;
 442                 if (!(*flags & LL_SBI_NOLCK) &&
 443                     ((*flags) = (*flags) |
 444                                 ll_set_opt("nolock", this_char,
 445                                            LL_SBI_NOLCK)))
 446                         continue;
 447         }
 448
 449         EXIT;
 450 }
 451
 452 void ll_lli_init(struct ll_inode_info *lli)
 453 {
 454         sema_init(&lli->lli_open_sem, 1);
 455         sema_init(&lli->lli_size_sem, 1);
 456         lli->lli_flags = 0;
 457         lli->lli_size_pid = 0;
 458         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 459         spin_lock_init(&lli->lli_lock);
 460         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 461         lli->lli_inode_magic = LLI_INODE_MAGIC;
 462         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 463         sema_init(&lli->lli_och_sem, 1);
 464         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 465         lli->lli_mds_exec_och = NULL;
 466         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 467         lli->lli_open_fd_exec_count = 0;
 468 }
 469
 470 int ll_fill_super(struct super_block *sb, void *data, int silent)
 471 {
 472         struct ll_sb_info *sbi;
 473         char *lov = NULL;
 474         char *lmv = NULL;
 475         int async, err;
 476         char *sec = NULL;
 477         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 478         __u64 remote_flag = 0;
 479         ENTRY;
 480
 481         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 482
 483         sbi = lustre_init_sbi(sb);
 484         if (!sbi)
 485                 RETURN(-ENOMEM);
 486
 487         sbi->ll_flags |= LL_SBI_READAHEAD;
 488         ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
 489
 490         if (!lov) {
 491                 CERROR("no osc\n");
 492                 GOTO(out, err = -EINVAL);
 493         }
 494
 495         if (!lmv) {
 496                 CERROR("no mdc\n");
 497                 GOTO(out, err = -EINVAL);
 498         }
 499
 500         err = lustre_common_fill_super(sb, lmv, lov, async, sec, nllu,
 501                                        &remote_flag);
 502         EXIT;
 503 out:
 504         if (err)
 505                 lustre_free_sbi(sb);
 506
 507         if (sec)
 508                 OBD_FREE(sec, strlen(sec) + 1);
 509         if (lmv)
 510                 OBD_FREE(lmv, strlen(lmv) + 1);
 511         if (lov)
 512                 OBD_FREE(lov, strlen(lov) + 1);
 513         return err;
 514 } /* ll_read_super */
 515
 516 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 517                               struct config_llog_instance *cfg, int allow_recov)
 518 {
 519         struct lustre_cfg lcfg;
 520         struct portals_cfg pcfg;
 521         char *peer = "MDS_PEER_UUID";
 522         struct obd_device *obd;
 523         struct lustre_handle md_conn = {0, };
 524         struct obd_export *exp;
 525         char *name = "mdc_dev";
 526         class_uuid_t uuid;
 527         struct obd_uuid lmv_uuid;
 528         struct llog_ctxt *ctxt;
 529         int rc = 0, err = 0;
 530         ENTRY;
 531
 532         if (lmd_bad_magic(lmd))
 533                 RETURN(-EINVAL);
 534
 535         generate_random_uuid(uuid);
 536         class_uuid_unparse(uuid, &lmv_uuid);
 537
 538         if (lmd->lmd_local_nid) {
 539                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 540                 pcfg.pcfg_nal = lmd->lmd_nal;
 541                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 542                 rc = libcfs_nal_cmd(&pcfg);
 543                 if (rc < 0)
 544                         GOTO(out, rc);
 545         }
 546
 547         if (lmd->lmd_nal == SOCKNAL ||
 548             lmd->lmd_nal == OPENIBNAL ||
 549             lmd->lmd_nal == IIBNAL ||
 550             lmd->lmd_nal == VIBNAL ||
 551             lmd->lmd_nal == RANAL) {
 552                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 553                 pcfg.pcfg_nal     = lmd->lmd_nal;
 554                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 555                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 556                 pcfg.pcfg_misc    = lmd->lmd_port;
 557                 rc = libcfs_nal_cmd(&pcfg);
 558                 if (rc < 0)
 559                         GOTO(out, rc);
 560         }
 561
 562         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
 563         lcfg.lcfg_nid = lmd->lmd_server_nid;
 564         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 565         lcfg.lcfg_inlbuf1 = peer;
 566         lcfg.lcfg_nal = lmd->lmd_nal;
 567         rc = class_process_config(&lcfg);
 568         if (rc < 0)
 569                 GOTO(out_del_conn, rc);
 570
 571         LCFG_INIT(lcfg, LCFG_ATTACH, name);
 572         lcfg.lcfg_inlbuf1 = "mdc";
 573         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 574         lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
 575         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 576         err = class_process_config(&lcfg);
 577         if (rc < 0)
 578                 GOTO(out_del_uuid, rc);
 579
 580         LCFG_INIT(lcfg, LCFG_SETUP, name);
 581         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
 582         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 583         lcfg.lcfg_inlbuf2 = peer;
 584         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 585         rc = class_process_config(&lcfg);
 586         if (rc < 0)
 587                 GOTO(out_detach, rc);
 588
 589         obd = class_name2obd(name);
 590         if (obd == NULL)
 591                 GOTO(out_cleanup, rc = -EINVAL);
 592
 593         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 594                           strlen(lmd->lmd_security), lmd->lmd_security);
 595         if (rc)
 596                 GOTO(out_cleanup, rc);
 597
 598         /* Disable initial recovery on this import */
 599         rc = obd_set_info(obd->obd_self_export,
 600                           strlen("initial_recov"), "initial_recov",
 601                           sizeof(allow_recov), &allow_recov);
 602         if (rc)
 603                 GOTO(out_cleanup, rc);
 604
 605         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 606         if (rc) {
 607                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 608                 GOTO(out_cleanup, rc);
 609         }
 610
 611         exp = class_conn2export(&md_conn);
 612
 613         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 614         rc = class_config_process_llog(ctxt, profile, cfg);
 615         if (rc)
 616                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 617
 618         err = obd_disconnect(exp, 0);
 619
 620         EXIT;
 621 out_cleanup:
 622         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
 623         err = class_process_config(&lcfg);
 624         if (err < 0)
 625                 GOTO(out, err);
 626
 627 out_detach:
 628         LCFG_INIT(lcfg, LCFG_DETACH, name);
 629         err = class_process_config(&lcfg);
 630         if (err < 0)
 631                 GOTO(out, err);
 632
 633 out_del_uuid:
 634         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
 635         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 636         lcfg.lcfg_inlbuf1 = peer;
 637         err = class_process_config(&lcfg);
 638
 639 out_del_conn:
 640         if (lmd->lmd_nal == SOCKNAL ||
 641             lmd->lmd_nal == OPENIBNAL ||
 642             lmd->lmd_nal == IIBNAL ||
 643             lmd->lmd_nal == VIBNAL ||
 644             lmd->lmd_nal == RANAL) {
 645                 int err2;
 646
 647                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 648                 pcfg.pcfg_nal     = lmd->lmd_nal;
 649                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 650                 pcfg.pcfg_flags   = 1;          /* single_share */
 651                 err2 = libcfs_nal_cmd(&pcfg);
 652                 if (err2 && !err)
 653                         err = err2;
 654                 if (err < 0)
 655                         GOTO(out, err);
 656         }
 657 out:
 658         if (rc == 0)
 659                 rc = err;
 660
 661         return rc;
 662 }
 663
 664 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 665 {
 666         struct lustre_cfg lcfg;
 667         struct obd_device *obd;
 668         int next = 0;
 669
 670         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 671         {
 672                 int err;
 673
 674                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
 675                 err = class_process_config(&lcfg);
 676                 if (err) {
 677                         CERROR("cleanup failed: %s\n", obd->obd_name);
 678                         //continue;
 679                 }
 680
 681                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
 682                 err = class_process_config(&lcfg);
 683                 if (err) {
 684                         CERROR("detach failed: %s\n", obd->obd_name);
 685                         //continue;
 686                 }
 687         }
 688
 689         if (sbi->ll_lmd != NULL)
 690                 class_del_profile(sbi->ll_lmd->lmd_profile);
 691 }
 692
 693 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 694 {
 695         struct lustre_mount_data * lmd = data;
 696         char *lov = NULL, *lmv = NULL;
 697         struct ll_sb_info *sbi;
 698         int err;
 699         ENTRY;
 700
 701         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 702         if (lmd_bad_magic(lmd))
 703                 RETURN(-EINVAL);
 704
 705         sbi = lustre_init_sbi(sb);
 706         if (!sbi)
 707                 RETURN(-ENOMEM);
 708
 709         sbi->ll_flags |= LL_SBI_READAHEAD;
 710
 711         if (lmd->lmd_profile) {
 712                 struct lustre_profile *lprof;
 713                 struct config_llog_instance cfg;
 714                 int len;
 715
 716                 if (lmd->lmd_mds[0] == '\0') {
 717                         CERROR("no mds name\n");
 718                         GOTO(out_free, err = -EINVAL);
 719                 }
 720                 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
 721
 722                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 723                 if (sbi->ll_lmd == NULL)
 724                         GOTO(out_free, err = -ENOMEM);
 725                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 726
 727                 /* generate a string unique to this super, let's try
 728                  the address of the super itself.*/
 729                 len = (sizeof(sb) * 2) + 1;
 730                 OBD_ALLOC(sbi->ll_instance, len);
 731                 if (sbi->ll_instance == NULL)
 732                         GOTO(out_free, err = -ENOMEM);
 733                 sprintf(sbi->ll_instance, "%p", sb);
 734
 735                 cfg.cfg_instance = sbi->ll_instance;
 736                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 737                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 738                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 739                 if (err < 0) {
 740                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 741                         GOTO(out_free, err);
 742                 }
 743
 744                 lprof = class_get_profile(lmd->lmd_profile);
 745                 if (lprof == NULL) {
 746                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 747                         GOTO(out_free, err = -EINVAL);
 748                 }
 749                 if (lov)
 750                         OBD_FREE(lov, strlen(lov) + 1);
 751                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
 752                           strlen(sbi->ll_instance) + 2);
 753                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 754
 755                 if (lmv)
 756                         OBD_FREE(lmv, strlen(lmv) + 1);
 757                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
 758                           strlen(sbi->ll_instance) + 2);
 759                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 760         }
 761
 762         if (!lov) {
 763                 CERROR("no osc\n");
 764                 GOTO(out_free, err = -EINVAL);
 765         }
 766
 767         if (!lmv) {
 768                 CERROR("no mdc\n");
 769                 GOTO(out_free, err = -EINVAL);
 770         }
 771
 772         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
 773                                        lmd->lmd_security, &lmd->lmd_nllu,
 774                                        &lmd->lmd_remote_flag);
 775
 776         if (err)
 777                 GOTO(out_free, err);
 778
 779 out_dev:
 780         if (lmv)
 781                 OBD_FREE(lmv, strlen(lmv) + 1);
 782         if (lov)
 783                 OBD_FREE(lov, strlen(lov) + 1);
 784
 785         RETURN(err);
 786
 787 out_free:
 788         if (sbi->ll_lmd) {
 789                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 790                 int err;
 791
 792                 if (sbi->ll_instance != NULL) {
 793                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 794                         struct config_llog_instance cfg;
 795                         char *cl_prof;
 796
 797                         cfg.cfg_instance = sbi->ll_instance;
 798                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 799
 800                         OBD_ALLOC(cl_prof, len);
 801                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 802                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 803                         if (err < 0) {
 804                                 CERROR("Unable to process log: %s\n", cl_prof);
 805                                 lustre_manual_cleanup(sbi);
 806                         }
 807                         OBD_FREE(cl_prof, len);
 808                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 809                 }
 810                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 811         }
 812         lustre_free_sbi(sb);
 813         goto out_dev;
 814 } /* lustre_fill_super */
 815
 816 void lustre_put_super(struct super_block *sb)
 817 {
 818         struct obd_device *obd;
 819         struct ll_sb_info *sbi = ll_s2sbi(sb);
 820         int force_umount = 0;
 821         ENTRY;
 822
 823         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 824         obd = class_exp2obd(sbi->ll_md_exp);
 825         if (obd)
 826                 force_umount = obd->obd_no_recov;
 827         obd = NULL;
 828
 829         lustre_common_put_super(sb);
 830         if (sbi->ll_lmd != NULL) {
 831                 char *cl_prof;
 832                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 833                 int err;
 834                 struct config_llog_instance cfg;
 835
 836                 if (force_umount) {
 837                         CERROR("force umount, doing manual cleanup\n");
 838                         lustre_manual_cleanup(sbi);
 839                         GOTO(free_lmd, 0);
 840                 }
 841
 842                 cfg.cfg_instance = sbi->ll_instance;
 843                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 844
 845                 OBD_ALLOC(cl_prof, len);
 846                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 847                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
 848                 if (err < 0) {
 849                         CERROR("Unable to process log: %s, doing manual cleanup"
 850                                "\n", cl_prof);
 851                         lustre_manual_cleanup(sbi);
 852                 }
 853
 854                 OBD_FREE(cl_prof, len);
 855         free_lmd:
 856                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 857                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 858         }
 859
 860         lustre_free_sbi(sb);
 861
 862         EXIT;
 863 } /* lustre_put_super */
 864
 865 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 866 {
 867         struct obd_export *md_exp = sbi->ll_md_exp;
 868         struct lustre_mount_data *lmd = sbi->ll_lmd;
 869         struct llog_ctxt *ctxt;
 870         struct config_llog_instance cfg;
 871         char *profile = lmd->lmd_profile, *name = NULL;
 872         int rc, namelen =  0, version;
 873         ENTRY;
 874
 875         if (profile == NULL)
 876                 RETURN(0);
 877         if (lmd == NULL) {
 878                 CERROR("Client not mounted with zero-conf; cannot process "
 879                        "update log.\n");
 880                 RETURN(0);
 881         }
 882
 883         rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
 884                                     LDLM_FL_CONFIG_CHANGE, NULL);
 885         if (rc != 0)
 886                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
 887
 888         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
 889                                NULL);
 890         if (rc != 0)
 891                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 892
 893         cfg.cfg_instance = sbi->ll_instance;
 894         cfg.cfg_uuid = sbi->ll_sb_uuid;
 895         cfg.cfg_local_nid = lmd->lmd_local_nid;
 896
 897         namelen = strlen(profile) + 20; /* -clean-######### */
 898         OBD_ALLOC(name, namelen);
 899         if (name == NULL)
 900                 RETURN(-ENOMEM);
 901
 902         if (clean) {
 903                 version = sbi->ll_config_version - 1;
 904                 sprintf(name, "%s-clean-%d", profile, version);
 905         } else {
 906                 version = sbi->ll_config_version + 1;
 907                 sprintf(name, "%s-%d", profile, version);
 908         }
 909
 910         CWARN("Applying configuration log %s\n", name);
 911
 912         ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
 913                                 LLOG_CONFIG_REPL_CTXT);
 914         rc = class_config_process_llog(ctxt, name, &cfg);
 915         if (rc == 0)
 916                 sbi->ll_config_version = version;
 917         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 918
 919         if (rc == 0 && clean == 0) {
 920                 struct lov_desc desc;
 921                 int rc, valsize;
 922                 valsize = sizeof(desc);
 923                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 924                                   "lovdesc", &valsize, &desc);
 925
 926                 rc = obd_init_ea_size(md_exp,
 927                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
 928                                       (desc.ld_tgt_count *
 929                                        sizeof(struct llog_cookie)));
 930         }
 931         OBD_FREE(name, namelen);
 932         RETURN(rc);
 933 }
 934
 935 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 936 {
 937         struct inode *inode = NULL;
 938         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 939         if (lock->l_ast_data) {
 940                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 941                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 942                         inode = igrab(lock->l_ast_data);
 943                 } else {
 944                         inode = lock->l_ast_data;
 945                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
 946                                "l_ast_data %p is bogus: magic %0x8\n",
 947                                lock->l_ast_data, lli->lli_inode_magic);
 948                         inode = NULL;
 949                 }
 950         }
 951         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 952         return inode;
 953 }
 954
 955 int null_if_equal(struct ldlm_lock *lock, void *data)
 956 {
 957         if (data == lock->l_ast_data) {
 958                 lock->l_ast_data = NULL;
 959
 960                 if (lock->l_req_mode != lock->l_granted_mode)
 961                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 962         }
 963
 964         return LDLM_ITER_CONTINUE;
 965 }
 966
 967 void ll_clear_inode(struct inode *inode)
 968 {
 969         struct lustre_id id;
 970         struct ll_inode_info *lli = ll_i2info(inode);
 971         struct ll_sb_info *sbi = ll_i2sbi(inode);
 972         ENTRY;
 973
 974         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 975                inode->i_generation, inode);
 976
 977         lli->lli_inode_magic = LLI_INODE_DEAD;
 978         ll_inode2id(&id, inode);
 979
 980         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 981         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
 982
 983         LASSERT(!lli->lli_open_fd_write_count);
 984         LASSERT(!lli->lli_open_fd_read_count);
 985         LASSERT(!lli->lli_open_fd_exec_count);
 986
 987         if (lli->lli_mds_write_och)
 988                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
 989         if (lli->lli_mds_exec_och)
 990                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
 991         if (lli->lli_mds_read_och)
 992                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
 993
 994         if (lli->lli_smd)
 995                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 996                                   null_if_equal, inode);
 997
 998         if (lli->lli_smd) {
 999                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1000                 lli->lli_smd = NULL;
1001         }
1002
1003         if (lli->lli_mea) {
1004                 obd_free_memmd(sbi->ll_md_exp,
1005                                (struct lov_stripe_md **) &lli->lli_mea);
1006                 lli->lli_mea = NULL;
1007         }
1008
1009         if (lli->lli_symlink_name) {
1010                 OBD_FREE(lli->lli_symlink_name,
1011                          strlen(lli->lli_symlink_name) + 1);
1012                 lli->lli_symlink_name = NULL;
1013         }
1014         lli->lli_inode_magic = LLI_INODE_DEAD;
1015
1016         EXIT;
1017 }
1018
1019 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1020  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1021  * keep these values until such a time that objects are allocated for it.
1022  * We do the MDS operations first, as it is checking permissions for us.
1023  * We don't to the MDS RPC if there is nothing that we want to store there,
1024  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1025  * going to do an RPC anyways.
1026  *
1027  * If we are doing a truncate, we will send the mtime and ctime updates
1028  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1029  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1030  * at the same time.
1031  */
1032 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1033 {
1034         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1035         struct ll_sb_info *sbi = ll_i2sbi(inode);
1036         struct ptlrpc_request *request = NULL;
1037         struct mdc_op_data *op_data;
1038         int ia_valid = attr->ia_valid;
1039         int err, rc = 0;
1040         ENTRY;
1041
1042         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1043         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1044
1045         if (ia_valid & ATTR_SIZE) {
1046                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1047                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1048                                attr->ia_size, ll_file_maxbytes(inode));
1049                         RETURN(-EFBIG);
1050                 }
1051
1052                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1053         }
1054
1055         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1056         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1057                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1058                         RETURN(-EPERM);
1059         }
1060
1061         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1062         if (attr->ia_valid & ATTR_CTIME) {
1063                 attr->ia_ctime = CURRENT_TIME;
1064                 attr->ia_valid |= ATTR_CTIME_SET;
1065         }
1066         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1067                 attr->ia_atime = CURRENT_TIME;
1068                 attr->ia_valid |= ATTR_ATIME_SET;
1069         }
1070         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1071                 attr->ia_mtime = CURRENT_TIME;
1072                 attr->ia_valid |= ATTR_MTIME_SET;
1073         }
1074
1075         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1076                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1077                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1078                        LTIME_S(CURRENT_TIME));
1079
1080         if (lsm)
1081                 attr->ia_valid &= ~ATTR_SIZE;
1082
1083         /* If only OST attributes being set on objects, don't do MDS RPC.
1084          * In that case, we need to check permissions and update the local
1085          * inode ourselves so we can call obdo_from_inode() always. */
1086         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1087                 struct lustre_md md;
1088
1089                 OBD_ALLOC(op_data, sizeof(*op_data));
1090                 if (op_data == NULL)
1091                         RETURN(-ENOMEM);
1092                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1093
1094                 rc = md_setattr(sbi->ll_md_exp, op_data,
1095                                 attr, NULL, 0, NULL, 0, &request);
1096                 OBD_FREE(op_data, sizeof(*op_data));
1097                 if (rc) {
1098                         ptlrpc_req_finished(request);
1099                         if (rc != -EPERM && rc != -EACCES)
1100                                 CERROR("md_setattr fails: rc = %d\n", rc);
1101                         RETURN(rc);
1102                 }
1103
1104                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1105                                        sbi->ll_dt_exp, &md);
1106                 if (rc) {
1107                         ptlrpc_req_finished(request);
1108                         RETURN(rc);
1109                 }
1110
1111                 /* We call inode_setattr to adjust timestamps, but we first
1112                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1113                  *
1114                  * NB: ATTR_SIZE will only be set at this point if the size
1115                  * resides on the MDS, ie, this file has no objects. */
1116                 attr->ia_valid &= ~ATTR_SIZE;
1117
1118                 /*
1119                  * assigning inode_setattr() to @err to disable warning that
1120                  * function's result should be checked by by caller. error is
1121                  * impossible here, as vmtruncate() control path is disabled.
1122                  */
1123                 err = inode_setattr(inode, attr);
1124                 ll_update_inode(inode, &md);
1125                 ptlrpc_req_finished(request);
1126
1127                 if (!lsm || !S_ISREG(inode->i_mode)) {
1128                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1129                         RETURN(0);
1130                 }
1131         } else {
1132                 /* The OST doesn't check permissions, but the alternative is
1133                  * a gratuitous RPC to the MDS.  We already rely on the client
1134                  * to do read/write/truncate permission checks, so is mtime OK?
1135                  */
1136                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1137                         /* from sys_utime() */
1138                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1139                                 if (current->fsuid != inode->i_uid &&
1140                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1141                                         RETURN(rc);
1142                         } else {
1143                                 /* from inode_change_ok() */
1144                                 if (current->fsuid != inode->i_uid &&
1145                                     !capable(CAP_FOWNER))
1146                                         RETURN(-EPERM);
1147                         }
1148                 }
1149
1150                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1151                 err = inode_setattr(inode, attr);
1152                 /*
1153                  * assigning inode_setattr() to @err to disable warning that
1154                  * function's result should be checked by by caller. error is
1155                  * impossible here, as vmtruncate() control path is disabled.
1156                  */
1157         }
1158
1159         /* We really need to get our PW lock before we change inode->i_size.
1160          * If we don't we can race with other i_size updaters on our node, like
1161          * ll_file_read.  We can also race with i_size propogation to other
1162          * nodes through dirtying and writeback of final cached pages.  This
1163          * last one is especially bad for racing o_append users on other
1164          * nodes. */
1165         if (ia_valid & ATTR_SIZE) {
1166                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1167                                                            OBD_OBJECT_EOF } };
1168                 struct lustre_handle lockh = { 0 };
1169                 struct ll_inode_info *lli = ll_i2info(inode);
1170                 int err, ast_flags = 0;
1171                 /* XXX when we fix the AST intents to pass the discard-range
1172                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1173                  * XXX here. */
1174                 if (attr->ia_size == 0)
1175                         ast_flags = LDLM_AST_DISCARD_DATA;
1176
1177                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1178                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1179
1180                 if (rc != 0)
1181                         RETURN(rc);
1182
1183                 down(&lli->lli_size_sem);
1184                 lli->lli_size_pid = current->pid;
1185                 rc = vmtruncate(inode, attr->ia_size);
1186                 if (rc != 0) {
1187                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1188                         lli->lli_size_pid = 0;
1189                         up(&lli->lli_size_sem);
1190                 }
1191
1192                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1193                 if (err) {
1194                         CERROR("ll_extent_unlock failed: %d\n", err);
1195                         if (!rc)
1196                                 rc = err;
1197                 }
1198         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1199                 struct obdo *oa = NULL;
1200
1201                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1202                        inode->i_ino, LTIME_S(attr->ia_mtime));
1203
1204                 oa = obdo_alloc();
1205                 if (oa == NULL)
1206                         RETURN(-ENOMEM);
1207
1208                 oa->o_id = lsm->lsm_object_id;
1209                 oa->o_gr = lsm->lsm_object_gr;
1210                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1211                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1212                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1213                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1214                 obdo_free(oa);
1215                 if (rc)
1216                         CERROR("obd_setattr fails: rc = %d\n", rc);
1217         }
1218         RETURN(rc);
1219 }
1220
1221 int ll_setattr(struct dentry *de, struct iattr *attr)
1222 {
1223         LASSERT(de->d_inode);
1224         return ll_setattr_raw(de->d_inode, attr);
1225 }
1226
1227 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1228                        unsigned long max_age)
1229 {
1230         struct ll_sb_info *sbi = ll_s2sbi(sb);
1231         struct obd_statfs obd_osfs;
1232         int rc;
1233         ENTRY;
1234
1235         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1236         if (rc) {
1237                 CERROR("obd_statfs fails: rc = %d\n", rc);
1238                 RETURN(rc);
1239         }
1240
1241         osfs->os_type = sb->s_magic;
1242
1243         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1244                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1245
1246         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1247         if (rc) {
1248                 CERROR("obd_statfs fails: rc = %d\n", rc);
1249                 RETURN(rc);
1250         }
1251
1252         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1253                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1254                obd_osfs.os_files);
1255
1256         osfs->os_blocks = obd_osfs.os_blocks;
1257         osfs->os_bfree = obd_osfs.os_bfree;
1258         osfs->os_bavail = obd_osfs.os_bavail;
1259
1260         /* If we don't have as many objects free on the OST as inodes
1261          * on the MDS, we reduce the total number of inodes to
1262          * compensate, so that the "inodes in use" number is correct.
1263          */
1264         if (obd_osfs.os_ffree < osfs->os_ffree) {
1265                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1266                         obd_osfs.os_ffree;
1267                 osfs->os_ffree = obd_osfs.os_ffree;
1268         }
1269
1270         RETURN(rc);
1271 }
1272
1273 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1274 {
1275         struct obd_statfs osfs;
1276         int rc;
1277
1278         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1279         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1280
1281         /* For now we will always get up-to-date statfs values, but in the
1282          * future we may allow some amount of caching on the client (e.g.
1283          * from QOS or lprocfs updates). */
1284         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1285         if (rc)
1286                 return rc;
1287
1288         statfs_unpack(sfs, &osfs);
1289
1290         if (sizeof(sfs->f_blocks) == 4) {
1291                 while (osfs.os_blocks > ~0UL) {
1292                         sfs->f_bsize <<= 1;
1293
1294                         osfs.os_blocks >>= 1;
1295                         osfs.os_bfree >>= 1;
1296                         osfs.os_bavail >>= 1;
1297                 }
1298         }
1299
1300         sfs->f_blocks = osfs.os_blocks;
1301         sfs->f_bfree = osfs.os_bfree;
1302         sfs->f_bavail = osfs.os_bavail;
1303
1304         return 0;
1305 }
1306
1307 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1308 {
1309         struct ll_inode_info *lli = ll_i2info(inode);
1310         struct lov_stripe_md *lsm = md->lsm;
1311         struct mds_body *body = md->body;
1312         struct mea *mea = md->mea;
1313         struct posix_acl *ll_acl_access = md->acl_access;
1314         ENTRY;
1315
1316         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1317
1318         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1319                 /* check for default striping info for dir. */
1320                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1321         }
1322
1323         if (lsm != NULL) {
1324                 LASSERT(lsm->lsm_object_gr > 0);
1325                 if (lli->lli_smd == NULL) {
1326                         lli->lli_smd = lsm;
1327                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1328                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1329                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1330                 } else {
1331                         int i;
1332                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1333                                 CERROR("lsm mismatch for inode %ld\n",
1334                                        inode->i_ino);
1335                                 CERROR("lli_smd:\n");
1336                                 dump_lsm(D_ERROR, lli->lli_smd);
1337                                 CERROR("lsm:\n");
1338                                 dump_lsm(D_ERROR, lsm);
1339                                 LBUG();
1340                         }
1341                         /* XXX FIXME -- We should decide on a safer (atomic) and
1342                          * more elegant way to update the lsm */
1343                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1344                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1345                                         lsm->lsm_oinfo[i].loi_id;
1346                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1347                                         lsm->lsm_oinfo[i].loi_gr;
1348                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1349                                         lsm->lsm_oinfo[i].loi_ost_idx;
1350                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1351                                         lsm->lsm_oinfo[i].loi_ost_gen;
1352                         }
1353                 }
1354                 /* bug 2844 - limit i_blksize for broken user-space apps */
1355                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1356                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1357                 if (lli->lli_smd != lsm)
1358                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1359         }
1360
1361         if (mea != NULL) {
1362                 if (lli->lli_mea == NULL) {
1363                         lli->lli_mea = mea;
1364                 } else {
1365                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1366                                 CERROR("mea mismatch for inode %lu\n",
1367                                         inode->i_ino);
1368                                 LBUG();
1369                         }
1370                 }
1371                 if (lli->lli_mea != mea)
1372                         obd_free_memmd(ll_i2mdexp(inode),
1373                                        (struct lov_stripe_md **) &mea);
1374         }
1375
1376         if (body->valid & OBD_MD_FID)
1377                 id_assign_fid(&lli->lli_id, &body->id1);
1378
1379         if (body->valid & OBD_MD_FLID)
1380                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1381
1382         if (body->valid & OBD_MD_FLGENER)
1383                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1384
1385         spin_lock(&lli->lli_lock);
1386         if (ll_acl_access != NULL) {
1387                 if (lli->lli_acl_access != NULL)
1388                         posix_acl_release(lli->lli_acl_access);
1389                 lli->lli_acl_access = ll_acl_access;
1390         }
1391         spin_unlock(&lli->lli_lock);
1392
1393         if (body->valid & OBD_MD_FLID)
1394                 inode->i_ino = id_ino(&body->id1);
1395         if (body->valid & OBD_MD_FLGENER)
1396                 inode->i_generation = id_gen(&body->id1);
1397         if (body->valid & OBD_MD_FLATIME)
1398                 LTIME_S(inode->i_atime) = body->atime;
1399         if (body->valid & OBD_MD_FLMTIME &&
1400             body->mtime > LTIME_S(inode->i_mtime)) {
1401                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1402                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1403                 LTIME_S(inode->i_mtime) = body->mtime;
1404         }
1405         if (body->valid & OBD_MD_FLCTIME &&
1406             body->ctime > LTIME_S(inode->i_ctime))
1407                 LTIME_S(inode->i_ctime) = body->ctime;
1408         if (body->valid & OBD_MD_FLMODE) {
1409                 inode->i_mode = (inode->i_mode & S_IFMT) |
1410                         (body->mode & ~S_IFMT);
1411         }
1412         if (body->valid & OBD_MD_FLTYPE) {
1413                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1414                         (body->mode & S_IFMT);
1415         }
1416         if (body->valid & OBD_MD_FLUID)
1417                 inode->i_uid = body->uid;
1418         if (body->valid & OBD_MD_FLGID)
1419                 inode->i_gid = body->gid;
1420         if (body->valid & OBD_MD_FLFLAGS)
1421                 inode->i_flags = body->flags;
1422         if (body->valid & OBD_MD_FLNLINK)
1423                 inode->i_nlink = body->nlink;
1424         if (body->valid & OBD_MD_FLRDEV)
1425 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1426                 inode->i_rdev = body->rdev;
1427 #else
1428                 inode->i_rdev = old_decode_dev(body->rdev);
1429 #endif
1430         if (body->valid & OBD_MD_FLSIZE)
1431                 inode->i_size = body->size;
1432         if (body->valid & OBD_MD_FLBLOCKS)
1433                 inode->i_blocks = body->blocks;
1434
1435         if (body->valid & OBD_MD_FLSIZE)
1436                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1437
1438 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1439         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1440 #endif
1441         LASSERT(id_fid(&lli->lli_id) != 0);
1442 }
1443
1444 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1445 static struct backing_dev_info ll_backing_dev_info = {
1446         .ra_pages       = 0,    /* No readahead */
1447         .memory_backed  = 0,    /* Does contribute to dirty memory */
1448 };
1449 #endif
1450
1451 void ll_read_inode2(struct inode *inode, void *opaque)
1452 {
1453         struct lustre_md *md = opaque;
1454         struct ll_inode_info *lli = ll_i2info(inode);
1455         ENTRY;
1456
1457         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1458                inode->i_generation, inode);
1459
1460         ll_lli_init(lli);
1461
1462         LASSERT(!lli->lli_smd);
1463
1464         /* Core attributes from the MDS first.  This is a new inode, and
1465          * the VFS doesn't zero times in the core inode so we have to do
1466          * it ourselves.  They will be overwritten by either MDS or OST
1467          * attributes - we just need to make sure they aren't newer. */
1468         LTIME_S(inode->i_mtime) = 0;
1469         LTIME_S(inode->i_atime) = 0;
1470         LTIME_S(inode->i_ctime) = 0;
1471
1472         inode->i_rdev = 0;
1473         ll_update_inode(inode, md);
1474
1475         /* OIDEBUG(inode); */
1476
1477         if (S_ISREG(inode->i_mode)) {
1478                 inode->i_op = &ll_file_inode_operations;
1479                 inode->i_fop = &ll_file_operations;
1480                 inode->i_mapping->a_ops = &ll_aops;
1481                 EXIT;
1482         } else if (S_ISDIR(inode->i_mode)) {
1483                 inode->i_op = &ll_dir_inode_operations;
1484                 inode->i_fop = &ll_dir_operations;
1485                 inode->i_mapping->a_ops = &ll_dir_aops;
1486                 EXIT;
1487         } else if (S_ISLNK(inode->i_mode)) {
1488                 inode->i_op = &ll_fast_symlink_inode_operations;
1489                 EXIT;
1490         } else {
1491                 inode->i_op = &ll_special_inode_operations;
1492
1493 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1494                 init_special_inode(inode, inode->i_mode,
1495                                    kdev_t_to_nr(inode->i_rdev));
1496
1497                 /* initializing backing dev info. */
1498                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1499 #else
1500                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1501 #endif
1502                 lli->ll_save_ifop = inode->i_fop;
1503
1504                 if (S_ISCHR(inode->i_mode))
1505                         inode->i_fop = &ll_special_chr_inode_fops;
1506                 else if (S_ISBLK(inode->i_mode))
1507                         inode->i_fop = &ll_special_blk_inode_fops;
1508                 else if (S_ISFIFO(inode->i_mode))
1509                         inode->i_fop = &ll_special_fifo_inode_fops;
1510                 else if (S_ISSOCK(inode->i_mode))
1511                         inode->i_fop = &ll_special_sock_inode_fops;
1512
1513                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1514                       inode->i_fop);
1515
1516                 if (lli->ll_save_ifop->owner) {
1517                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1518                               lli->ll_save_ifop->owner);
1519                 }
1520                 EXIT;
1521         }
1522 }
1523
1524 void ll_delete_inode(struct inode *inode)
1525 {
1526         struct ll_sb_info *sbi = ll_i2sbi(inode);
1527         struct lustre_id id;
1528         int rc;
1529         ENTRY;
1530
1531         ll_inode2id(&id, inode);
1532
1533         rc = md_delete_inode(sbi->ll_md_exp, &id);
1534         if (rc) {
1535                 CERROR("md_delete_inode() failed, error %d\n",
1536                        rc);
1537         }
1538
1539         clear_inode(inode);
1540         EXIT;
1541 }
1542
1543 int ll_iocontrol(struct inode *inode, struct file *file,
1544                  unsigned int cmd, unsigned long arg)
1545 {
1546         struct ll_sb_info *sbi = ll_i2sbi(inode);
1547         struct ptlrpc_request *req = NULL;
1548         int rc, flags = 0;
1549         ENTRY;
1550
1551         switch(cmd) {
1552         case EXT3_IOC_GETFLAGS: {
1553                 struct lustre_id id;
1554                 __u64 valid = OBD_MD_FLFLAGS;
1555                 struct mds_body *body;
1556
1557                 ll_inode2id(&id, inode);
1558                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1559                 if (rc) {
1560                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1561                         RETURN(-abs(rc));
1562                 }
1563
1564                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1565
1566                 if (body->flags & S_APPEND)
1567                         flags |= EXT3_APPEND_FL;
1568                 if (body->flags & S_IMMUTABLE)
1569                         flags |= EXT3_IMMUTABLE_FL;
1570                 if (body->flags & S_NOATIME)
1571                         flags |= EXT3_NOATIME_FL;
1572
1573                 ptlrpc_req_finished (req);
1574
1575                 RETURN(put_user(flags, (int *)arg));
1576         }
1577         case EXT3_IOC_SETFLAGS: {
1578                 struct mdc_op_data *op_data;
1579                 struct iattr attr;
1580                 struct obdo *oa;
1581                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1582
1583                 if (get_user(flags, (int *)arg))
1584                         RETURN(-EFAULT);
1585
1586                 oa = obdo_alloc();
1587                 if (!oa)
1588                         RETURN(-ENOMEM);
1589
1590                 OBD_ALLOC(op_data, sizeof(*op_data));
1591                 if (op_data == NULL) {
1592                         obdo_free(oa);
1593                         RETURN(-ENOMEM);
1594                 }
1595                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1596
1597                 memset(&attr, 0x0, sizeof(attr));
1598                 attr.ia_attr_flags = flags;
1599                 attr.ia_valid |= ATTR_ATTR_FLAG;
1600
1601                 rc = md_setattr(sbi->ll_md_exp, op_data,
1602                                 &attr, NULL, 0, NULL, 0, &req);
1603                 OBD_FREE(op_data, sizeof(*op_data));
1604                 if (rc) {
1605                         ptlrpc_req_finished(req);
1606                         if (rc != -EPERM && rc != -EACCES)
1607                                 CERROR("md_setattr fails: rc = %d\n", rc);
1608                         obdo_free(oa);
1609                         RETURN(rc);
1610                 }
1611                 ptlrpc_req_finished(req);
1612
1613                 oa->o_id = lsm->lsm_object_id;
1614                 oa->o_gr = lsm->lsm_object_gr;
1615                 oa->o_flags = flags;
1616                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1617
1618                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1619                 obdo_free(oa);
1620                 if (rc) {
1621                         if (rc != -EPERM && rc != -EACCES)
1622                                 CERROR("md_setattr fails: rc = %d\n", rc);
1623                         RETURN(rc);
1624                 }
1625
1626                 if (flags & EXT3_APPEND_FL)
1627                         inode->i_flags |= S_APPEND;
1628                 else
1629                         inode->i_flags &= ~S_APPEND;
1630                 if (flags & EXT3_IMMUTABLE_FL)
1631                         inode->i_flags |= S_IMMUTABLE;
1632                 else
1633                         inode->i_flags &= ~S_IMMUTABLE;
1634                 if (flags & EXT3_NOATIME_FL)
1635                         inode->i_flags |= S_NOATIME;
1636                 else
1637                         inode->i_flags &= ~S_NOATIME;
1638
1639                 RETURN(0);
1640         }
1641         default:
1642                 RETURN(-ENOSYS);
1643         }
1644
1645         RETURN(0);
1646 }
1647
1648 /* this is only called in the case of forced umount. */
1649 void ll_umount_begin(struct super_block *sb)
1650 {
1651         struct ll_sb_info *sbi = ll_s2sbi(sb);
1652         struct obd_ioctl_data ioc_data = { 0 };
1653         struct obd_device *obd;
1654         ENTRY;
1655
1656         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1657                sb->s_count, atomic_read(&sb->s_active));
1658
1659         obd = class_exp2obd(sbi->ll_md_exp);
1660         if (obd == NULL) {
1661                 CERROR("Invalid MDC connection handle "LPX64"\n",
1662                        sbi->ll_md_exp->exp_handle.h_cookie);
1663                 EXIT;
1664                 return;
1665         }
1666         obd->obd_no_recov = 1;
1667         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1668                       sizeof(ioc_data), &ioc_data, NULL);
1669
1670         obd = class_exp2obd(sbi->ll_dt_exp);
1671         if (obd == NULL) {
1672                 CERROR("Invalid LOV connection handle "LPX64"\n",
1673                        sbi->ll_dt_exp->exp_handle.h_cookie);
1674                 EXIT;
1675                 return;
1676         }
1677
1678         obd->obd_no_recov = 1;
1679         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1680                       sizeof(ioc_data), &ioc_data, NULL);
1681
1682         /*
1683          * really, we'd like to wait until there are no requests outstanding,
1684          * and then continue.  For now, we just invalidate the requests,
1685          * schedule, and hope.
1686          */
1687         schedule();
1688
1689         EXIT;
1690 }
1691
1692 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1693                   struct inode **inode, struct ptlrpc_request *req,
1694                   int offset, struct super_block *sb)
1695 {
1696         struct lustre_md md;
1697         int rc = 0;
1698
1699         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1700         if (rc)
1701                 RETURN(rc);
1702
1703         if (*inode) {
1704                 ll_update_inode(*inode, &md);
1705         } else {
1706                 LASSERT(sb);
1707                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1708                 if (*inode == NULL || is_bad_inode(*inode)) {
1709                         /* free the lsm if we allocated one above */
1710                         if (md.lsm != NULL)
1711                                 obd_free_memmd(dt_exp, &md.lsm);
1712                         if (md.mea != NULL)
1713                                 obd_free_memmd(md_exp,
1714                                                (struct lov_stripe_md**)&md.mea);
1715                         rc = -ENOMEM;
1716                         CERROR("new_inode -fatal: rc %d\n", rc);
1717                 }
1718         }
1719
1720         RETURN(rc);
1721 }
1722
1723 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1724                char *filename, struct lustre_id *ret)
1725 {
1726         struct ptlrpc_request *request = NULL;
1727         struct mds_body *body;
1728         int rc;
1729
1730         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1731                              OBD_MD_FID, 0, &request);
1732         if (rc < 0) {
1733                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1734                        filename, rc);
1735                 return rc;
1736         }
1737
1738         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1739         LASSERT(body != NULL);
1740         LASSERT_REPSWABBED(request, 0);
1741
1742         *ret = body->id1;
1743         ptlrpc_req_finished(request);
1744
1745         return rc;
1746 }