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         int valsize, rc;
 119         ENTRY;
 120
 121         valsize = sizeof(sbi->ll_dt_desc);
 122         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 123         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 124                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 125         RETURN(rc);
 126 }
 127
 128 extern struct dentry_operations ll_d_ops;
 129
 130 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 131                              int async, char *security, __u32 *nllu,
 132                              __u64 *remote)
 133 {
 134         struct ll_sb_info *sbi = ll_s2sbi(sb);
 135         struct ptlrpc_request *request = NULL;
 136         struct lustre_handle dt_conn = {0, };
 137         struct lustre_handle md_conn = {0, };
 138         struct obd_connect_data *data;
 139         struct inode *root = NULL;
 140         struct obd_device *obd;
 141         struct obd_statfs osfs;
 142         struct lustre_md md;
 143         kdev_t devno;
 144         int err;
 145         __u32 valsize;
 146         ENTRY;
 147
 148         obd = class_name2obd(lmv);
 149         if (!obd) {
 150                 CERROR("MDC %s: not setup or attached\n", lmv);
 151                 RETURN(-EINVAL);
 152         }
 153         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 154                      sizeof(async), &async);
 155
 156         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 157             (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 158                 CERROR("wrong remote flag "LPX64"\n", *remote);
 159                 RETURN(-EINVAL);
 160         }
 161
 162         OBD_ALLOC(data, sizeof(*data));
 163         if (!data)
 164                 RETURN(-ENOMEM);
 165
 166         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 167                                               OBD_CONNECT_REMOTE);
 168         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 169
 170         if (security == NULL)
 171                 security = "null";
 172
 173         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 174                            strlen(security), security);
 175         if (err) {
 176                 CERROR("LMV %s: failed to set security %s, err %d\n",
 177                         lmv, security, err);
 178                 OBD_FREE(data, sizeof(*data));
 179                 RETURN(err);
 180         }
 181
 182         if (proc_lustre_fs_root) {
 183                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 184                                                   lov, lmv);
 185                 if (err < 0)
 186                         CERROR("could not register mount in /proc/lustre");
 187         }
 188
 189         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
 190                           OBD_OPT_REAL_CLIENT);
 191         if (err == -EBUSY) {
 192                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 193                        " client is not a part.  Please wait for recovery to "
 194                        "complete, abort, or time out.\n", lmv);
 195                 GOTO(out, err);
 196         } else if (err) {
 197                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 198                 GOTO(out, err);
 199         }
 200         sbi->ll_md_exp = class_conn2export(&md_conn);
 201         err = obd_statfs(obd, &osfs, jiffies - HZ);
 202         if (err)
 203                 GOTO(out_lmv, err);
 204
 205         if (!osfs.os_bsize) {
 206                 CERROR("Invalid block size is detected.");
 207                 GOTO(out_lmv, err);
 208         }
 209
 210         sb->s_magic = LL_SUPER_MAGIC;
 211         sb->s_blocksize = osfs.os_bsize;
 212         sb->s_blocksize_bits = log2(osfs.os_bsize);
 213         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 214
 215         devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 216                              strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 217
 218         sb->s_dev = devno;
 219
 220         /* after statfs, we are supposed to have connected to MDSs,
 221          * so it's ok to check remote flag returned.
 222          */
 223         valsize = sizeof(&sbi->ll_remote);
 224         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 225                            &valsize, &sbi->ll_remote);
 226         if (err) {
 227                 CERROR("fail to obtain remote flag\n");
 228                 GOTO(out, err);
 229         }
 230
 231         obd = class_name2obd(lov);
 232         if (!obd) {
 233                 CERROR("OSC %s: not setup or attached\n", lov);
 234                 GOTO(out_lmv, err);
 235         }
 236         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 237                      sizeof(async), &async);
 238
 239         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data,
 240                           OBD_OPT_REAL_CLIENT);
 241         if (err == -EBUSY) {
 242                 CERROR("An OST (lov %s) is performing recovery, of which this"
 243                        " client is not a part.  Please wait for recovery to "
 244                        "complete, abort, or time out.\n", lov);
 245                 GOTO(out, err);
 246         } else if (err) {
 247                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 248                 GOTO(out_lmv, err);
 249         }
 250         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 251
 252         err = lustre_init_dt_desc(sbi);
 253         if (err == 0) {
 254                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 255                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
 256                                  sbi->ll_dt_desc.ld_tgt_count *
 257                                  sizeof(struct llog_cookie));
 258         }
 259
 260         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 261         if (err) {
 262                 CERROR("cannot mds_connect: rc = %d\n", err);
 263                 GOTO(out_lov, err);
 264         }
 265         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 266
 267         sb->s_op = &lustre_super_operations;
 268
 269         /* make root inode
 270          * XXX: move this to after cbd setup? */
 271         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 272                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
 273                          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, nllu,
 502                                        &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;
 521         struct portals_cfg pcfg;
 522         char *peer = "MDS_PEER_UUID";
 523         struct obd_device *obd;
 524         struct lustre_handle md_conn = {0, };
 525         struct obd_export *exp;
 526         char *name = "mdc_dev";
 527         class_uuid_t uuid;
 528         struct obd_uuid lmv_uuid;
 529         struct llog_ctxt *ctxt;
 530         int rc = 0, err = 0;
 531         ENTRY;
 532
 533         if (lmd_bad_magic(lmd))
 534                 RETURN(-EINVAL);
 535
 536         generate_random_uuid(uuid);
 537         class_uuid_unparse(uuid, &lmv_uuid);
 538
 539         if (lmd->lmd_local_nid) {
 540                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 541                 pcfg.pcfg_nal = lmd->lmd_nal;
 542                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 543                 rc = libcfs_nal_cmd(&pcfg);
 544                 if (rc < 0)
 545                         GOTO(out, rc);
 546         }
 547
 548         if (lmd->lmd_nal == SOCKNAL ||
 549             lmd->lmd_nal == OPENIBNAL ||
 550             lmd->lmd_nal == IIBNAL ||
 551             lmd->lmd_nal == VIBNAL ||
 552             lmd->lmd_nal == RANAL) {
 553                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 554                 pcfg.pcfg_nal     = lmd->lmd_nal;
 555                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 556                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 557                 pcfg.pcfg_misc    = lmd->lmd_port;
 558                 rc = libcfs_nal_cmd(&pcfg);
 559                 if (rc < 0)
 560                         GOTO(out, rc);
 561         }
 562
 563         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
 564         lcfg.lcfg_nid = lmd->lmd_server_nid;
 565         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 566         lcfg.lcfg_inlbuf1 = peer;
 567         lcfg.lcfg_nal = lmd->lmd_nal;
 568         rc = class_process_config(&lcfg);
 569         if (rc < 0)
 570                 GOTO(out_del_conn, rc);
 571
 572         LCFG_INIT(lcfg, LCFG_ATTACH, name);
 573         lcfg.lcfg_inlbuf1 = "mdc";
 574         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 575         lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
 576         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 577         err = class_process_config(&lcfg);
 578         if (rc < 0)
 579                 GOTO(out_del_uuid, rc);
 580
 581         LCFG_INIT(lcfg, LCFG_SETUP, name);
 582         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
 583         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 584         lcfg.lcfg_inlbuf2 = peer;
 585         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 586         rc = class_process_config(&lcfg);
 587         if (rc < 0)
 588                 GOTO(out_detach, rc);
 589
 590         obd = class_name2obd(name);
 591         if (obd == NULL)
 592                 GOTO(out_cleanup, rc = -EINVAL);
 593
 594         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 595                           strlen(lmd->lmd_security), lmd->lmd_security);
 596         if (rc)
 597                 GOTO(out_cleanup, rc);
 598
 599         /* Disable initial recovery on this import */
 600         rc = obd_set_info(obd->obd_self_export,
 601                           strlen("initial_recov"), "initial_recov",
 602                           sizeof(allow_recov), &allow_recov);
 603         if (rc)
 604                 GOTO(out_cleanup, rc);
 605
 606         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 607         if (rc) {
 608                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 609                 GOTO(out_cleanup, rc);
 610         }
 611
 612         exp = class_conn2export(&md_conn);
 613
 614         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 615         rc = class_config_process_llog(ctxt, profile, cfg);
 616         if (rc)
 617                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 618
 619         err = obd_disconnect(exp, 0);
 620
 621         EXIT;
 622 out_cleanup:
 623         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
 624         err = class_process_config(&lcfg);
 625         if (err < 0)
 626                 GOTO(out, err);
 627
 628 out_detach:
 629         LCFG_INIT(lcfg, LCFG_DETACH, name);
 630         err = class_process_config(&lcfg);
 631         if (err < 0)
 632                 GOTO(out, err);
 633
 634 out_del_uuid:
 635         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
 636         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 637         lcfg.lcfg_inlbuf1 = peer;
 638         err = class_process_config(&lcfg);
 639
 640 out_del_conn:
 641         if (lmd->lmd_nal == SOCKNAL ||
 642             lmd->lmd_nal == OPENIBNAL ||
 643             lmd->lmd_nal == IIBNAL ||
 644             lmd->lmd_nal == VIBNAL ||
 645             lmd->lmd_nal == RANAL) {
 646                 int err2;
 647
 648                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 649                 pcfg.pcfg_nal     = lmd->lmd_nal;
 650                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 651                 pcfg.pcfg_flags   = 1;          /* single_share */
 652                 err2 = libcfs_nal_cmd(&pcfg);
 653                 if (err2 && !err)
 654                         err = err2;
 655                 if (err < 0)
 656                         GOTO(out, err);
 657         }
 658 out:
 659         if (rc == 0)
 660                 rc = err;
 661
 662         return rc;
 663 }
 664
 665 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 666 {
 667         struct lustre_cfg lcfg;
 668         struct obd_device *obd;
 669         int next = 0;
 670
 671         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 672         {
 673                 int err;
 674
 675                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
 676                 err = class_process_config(&lcfg);
 677                 if (err) {
 678                         CERROR("cleanup failed: %s\n", obd->obd_name);
 679                         //continue;
 680                 }
 681
 682                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
 683                 err = class_process_config(&lcfg);
 684                 if (err) {
 685                         CERROR("detach failed: %s\n", obd->obd_name);
 686                         //continue;
 687                 }
 688         }
 689
 690         if (sbi->ll_lmd != NULL)
 691                 class_del_profile(sbi->ll_lmd->lmd_profile);
 692 }
 693
 694 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 695 {
 696         struct lustre_mount_data * lmd = data;
 697         char *lov = NULL, *lmv = NULL;
 698         struct ll_sb_info *sbi;
 699         int err;
 700         ENTRY;
 701
 702         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 703         if (lmd_bad_magic(lmd))
 704                 RETURN(-EINVAL);
 705
 706         sbi = lustre_init_sbi(sb);
 707         if (!sbi)
 708                 RETURN(-ENOMEM);
 709
 710         sbi->ll_flags |= LL_SBI_READAHEAD;
 711
 712         if (lmd->lmd_profile) {
 713                 struct lustre_profile *lprof;
 714                 struct config_llog_instance cfg;
 715                 int len;
 716
 717                 if (lmd->lmd_mds[0] == '\0') {
 718                         CERROR("no mds name\n");
 719                         GOTO(out_free, err = -EINVAL);
 720                 }
 721                 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
 722
 723                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 724                 if (sbi->ll_lmd == NULL)
 725                         GOTO(out_free, err = -ENOMEM);
 726                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 727
 728                 /* generate a string unique to this super, let's try
 729                  the address of the super itself.*/
 730                 len = (sizeof(sb) * 2) + 1;
 731                 OBD_ALLOC(sbi->ll_instance, len);
 732                 if (sbi->ll_instance == NULL)
 733                         GOTO(out_free, err = -ENOMEM);
 734                 sprintf(sbi->ll_instance, "%p", sb);
 735
 736                 cfg.cfg_instance = sbi->ll_instance;
 737                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 738                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 739                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 740                 if (err < 0) {
 741                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 742                         GOTO(out_free, err);
 743                 }
 744
 745                 lprof = class_get_profile(lmd->lmd_profile);
 746                 if (lprof == NULL) {
 747                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 748                         GOTO(out_free, err = -EINVAL);
 749                 }
 750                 if (lov)
 751                         OBD_FREE(lov, strlen(lov) + 1);
 752                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
 753                           strlen(sbi->ll_instance) + 2);
 754                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 755
 756                 if (lmv)
 757                         OBD_FREE(lmv, strlen(lmv) + 1);
 758                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
 759                           strlen(sbi->ll_instance) + 2);
 760                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 761         }
 762
 763         if (!lov) {
 764                 CERROR("no osc\n");
 765                 GOTO(out_free, err = -EINVAL);
 766         }
 767
 768         if (!lmv) {
 769                 CERROR("no mdc\n");
 770                 GOTO(out_free, err = -EINVAL);
 771         }
 772
 773         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
 774                                        lmd->lmd_security, &lmd->lmd_nllu,
 775                                        &lmd->lmd_remote_flag);
 776
 777         if (err)
 778                 GOTO(out_free, err);
 779
 780 out_dev:
 781         if (lmv)
 782                 OBD_FREE(lmv, strlen(lmv) + 1);
 783         if (lov)
 784                 OBD_FREE(lov, strlen(lov) + 1);
 785
 786         RETURN(err);
 787
 788 out_free:
 789         if (sbi->ll_lmd) {
 790                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 791                 int err;
 792
 793                 if (sbi->ll_instance != NULL) {
 794                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 795                         struct config_llog_instance cfg;
 796                         char *cl_prof;
 797
 798                         cfg.cfg_instance = sbi->ll_instance;
 799                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 800
 801                         OBD_ALLOC(cl_prof, len);
 802                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 803                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 804                         if (err < 0) {
 805                                 CERROR("Unable to process log: %s\n", cl_prof);
 806                                 lustre_manual_cleanup(sbi);
 807                         }
 808                         OBD_FREE(cl_prof, len);
 809                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 810                 }
 811                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 812         }
 813         lustre_free_sbi(sb);
 814         goto out_dev;
 815 } /* lustre_fill_super */
 816
 817 void lustre_put_super(struct super_block *sb)
 818 {
 819         struct obd_device *obd;
 820         struct ll_sb_info *sbi = ll_s2sbi(sb);
 821         int force_umount = 0;
 822         ENTRY;
 823
 824         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 825         obd = class_exp2obd(sbi->ll_md_exp);
 826         if (obd)
 827                 force_umount = obd->obd_no_recov;
 828         obd = NULL;
 829
 830         lustre_common_put_super(sb);
 831         if (sbi->ll_lmd != NULL) {
 832                 char *cl_prof;
 833                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 834                 int err;
 835                 struct config_llog_instance cfg;
 836
 837                 if (force_umount) {
 838                         CERROR("force umount, doing manual cleanup\n");
 839                         lustre_manual_cleanup(sbi);
 840                         GOTO(free_lmd, 0);
 841                 }
 842
 843                 cfg.cfg_instance = sbi->ll_instance;
 844                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 845
 846                 OBD_ALLOC(cl_prof, len);
 847                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 848                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
 849                 if (err < 0) {
 850                         CERROR("Unable to process log: %s, doing manual cleanup"
 851                                "\n", cl_prof);
 852                         lustre_manual_cleanup(sbi);
 853                 }
 854
 855                 OBD_FREE(cl_prof, len);
 856         free_lmd:
 857                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 858                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 859         }
 860
 861         lustre_free_sbi(sb);
 862
 863         EXIT;
 864 } /* lustre_put_super */
 865
 866 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 867 {
 868         struct obd_export *md_exp = sbi->ll_md_exp;
 869         struct lustre_mount_data *lmd = sbi->ll_lmd;
 870         struct llog_ctxt *ctxt;
 871         struct config_llog_instance cfg;
 872         char *profile = lmd->lmd_profile, *name = NULL;
 873         int rc, namelen =  0, version;
 874         ENTRY;
 875
 876         if (profile == NULL)
 877                 RETURN(0);
 878         if (lmd == NULL) {
 879                 CERROR("Client not mounted with zero-conf; cannot process "
 880                        "update log.\n");
 881                 RETURN(0);
 882         }
 883
 884         rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
 885                                     LDLM_FL_CONFIG_CHANGE, NULL);
 886         if (rc != 0)
 887                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
 888
 889         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
 890                                NULL);
 891         if (rc != 0)
 892                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 893
 894         cfg.cfg_instance = sbi->ll_instance;
 895         cfg.cfg_uuid = sbi->ll_sb_uuid;
 896         cfg.cfg_local_nid = lmd->lmd_local_nid;
 897
 898         namelen = strlen(profile) + 20; /* -clean-######### */
 899         OBD_ALLOC(name, namelen);
 900         if (name == NULL)
 901                 RETURN(-ENOMEM);
 902
 903         if (clean) {
 904                 version = sbi->ll_config_version - 1;
 905                 sprintf(name, "%s-clean-%d", profile, version);
 906         } else {
 907                 version = sbi->ll_config_version + 1;
 908                 sprintf(name, "%s-%d", profile, version);
 909         }
 910
 911         CWARN("Applying configuration log %s\n", name);
 912
 913         ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
 914                                 LLOG_CONFIG_REPL_CTXT);
 915         rc = class_config_process_llog(ctxt, name, &cfg);
 916         if (rc == 0)
 917                 sbi->ll_config_version = version;
 918         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 919
 920         if (rc == 0 && clean == 0) {
 921                 struct lov_desc desc;
 922                 int rc, valsize;
 923                 valsize = sizeof(desc);
 924                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 925                                   "lovdesc", &valsize, &desc);
 926
 927                 rc = obd_init_ea_size(md_exp,
 928                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
 929                                       (desc.ld_tgt_count *
 930                                        sizeof(struct llog_cookie)));
 931         }
 932         OBD_FREE(name, namelen);
 933         RETURN(rc);
 934 }
 935
 936 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 937 {
 938         struct inode *inode = NULL;
 939         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 940         if (lock->l_ast_data) {
 941                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 942                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 943                         inode = igrab(lock->l_ast_data);
 944                 } else {
 945                         inode = lock->l_ast_data;
 946                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
 947                                "l_ast_data %p is bogus: magic %0x8\n",
 948                                lock->l_ast_data, lli->lli_inode_magic);
 949                         inode = NULL;
 950                 }
 951         }
 952         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 953         return inode;
 954 }
 955
 956 int null_if_equal(struct ldlm_lock *lock, void *data)
 957 {
 958         if (data == lock->l_ast_data) {
 959                 lock->l_ast_data = NULL;
 960
 961                 if (lock->l_req_mode != lock->l_granted_mode)
 962                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 963         }
 964
 965         return LDLM_ITER_CONTINUE;
 966 }
 967
 968 void ll_clear_inode(struct inode *inode)
 969 {
 970         struct lustre_id id;
 971         struct ll_inode_info *lli = ll_i2info(inode);
 972         struct ll_sb_info *sbi = ll_i2sbi(inode);
 973         ENTRY;
 974
 975         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 976                inode->i_generation, inode);
 977
 978         lli->lli_inode_magic = LLI_INODE_DEAD;
 979         ll_inode2id(&id, inode);
 980
 981         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 982         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
 983
 984         LASSERT(!lli->lli_open_fd_write_count);
 985         LASSERT(!lli->lli_open_fd_read_count);
 986         LASSERT(!lli->lli_open_fd_exec_count);
 987
 988         if (lli->lli_mds_write_och)
 989                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
 990         if (lli->lli_mds_exec_och)
 991                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
 992         if (lli->lli_mds_read_och)
 993                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
 994
 995         if (lli->lli_smd)
 996                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 997                                   null_if_equal, inode);
 998
 999         if (lli->lli_smd) {
1000                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1001                 lli->lli_smd = NULL;
1002         }
1003
1004         if (lli->lli_mea) {
1005                 obd_free_memmd(sbi->ll_md_exp,
1006                                (struct lov_stripe_md **) &lli->lli_mea);
1007                 lli->lli_mea = NULL;
1008         }
1009
1010         if (lli->lli_symlink_name) {
1011                 OBD_FREE(lli->lli_symlink_name,
1012                          strlen(lli->lli_symlink_name) + 1);
1013                 lli->lli_symlink_name = NULL;
1014         }
1015         lli->lli_inode_magic = LLI_INODE_DEAD;
1016
1017         EXIT;
1018 }
1019
1020 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1021  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1022  * keep these values until such a time that objects are allocated for it.
1023  * We do the MDS operations first, as it is checking permissions for us.
1024  * We don't to the MDS RPC if there is nothing that we want to store there,
1025  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1026  * going to do an RPC anyways.
1027  *
1028  * If we are doing a truncate, we will send the mtime and ctime updates
1029  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1030  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1031  * at the same time.
1032  */
1033 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1034 {
1035         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1036         struct ll_sb_info *sbi = ll_i2sbi(inode);
1037         struct ptlrpc_request *request = NULL;
1038         struct mdc_op_data *op_data;
1039         int ia_valid = attr->ia_valid;
1040         int err, rc = 0;
1041         ENTRY;
1042
1043         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1044         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1045
1046         if (ia_valid & ATTR_SIZE) {
1047                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1048                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1049                                attr->ia_size, ll_file_maxbytes(inode));
1050                         RETURN(-EFBIG);
1051                 }
1052
1053                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1054         }
1055
1056         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1057         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1058                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1059                         RETURN(-EPERM);
1060         }
1061
1062         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1063         if (attr->ia_valid & ATTR_CTIME) {
1064                 attr->ia_ctime = CURRENT_TIME;
1065                 attr->ia_valid |= ATTR_CTIME_SET;
1066         }
1067         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1068                 attr->ia_atime = CURRENT_TIME;
1069                 attr->ia_valid |= ATTR_ATIME_SET;
1070         }
1071         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1072                 attr->ia_mtime = CURRENT_TIME;
1073                 attr->ia_valid |= ATTR_MTIME_SET;
1074         }
1075
1076         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1077                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1078                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1079                        LTIME_S(CURRENT_TIME));
1080
1081         if (lsm)
1082                 attr->ia_valid &= ~ATTR_SIZE;
1083
1084         /* If only OST attributes being set on objects, don't do MDS RPC.
1085          * In that case, we need to check permissions and update the local
1086          * inode ourselves so we can call obdo_from_inode() always. */
1087         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1088                 struct lustre_md md;
1089
1090                 OBD_ALLOC(op_data, sizeof(*op_data));
1091                 if (op_data == NULL)
1092                         RETURN(-ENOMEM);
1093                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1094
1095                 rc = md_setattr(sbi->ll_md_exp, op_data,
1096                                 attr, NULL, 0, NULL, 0, &request);
1097                 OBD_FREE(op_data, sizeof(*op_data));
1098                 if (rc) {
1099                         ptlrpc_req_finished(request);
1100                         if (rc != -EPERM && rc != -EACCES)
1101                                 CERROR("md_setattr fails: rc = %d\n", rc);
1102                         RETURN(rc);
1103                 }
1104
1105                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1106                                        sbi->ll_dt_exp, &md);
1107                 if (rc) {
1108                         ptlrpc_req_finished(request);
1109                         RETURN(rc);
1110                 }
1111
1112                 /* We call inode_setattr to adjust timestamps, but we first
1113                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1114                  *
1115                  * NB: ATTR_SIZE will only be set at this point if the size
1116                  * resides on the MDS, ie, this file has no objects. */
1117                 attr->ia_valid &= ~ATTR_SIZE;
1118
1119                 /*
1120                  * assigning inode_setattr() to @err to disable warning that
1121                  * function's result should be checked by by caller. error is
1122                  * impossible here, as vmtruncate() control path is disabled.
1123                  */
1124                 err = inode_setattr(inode, attr);
1125                 ll_update_inode(inode, &md);
1126                 ptlrpc_req_finished(request);
1127
1128                 if (!lsm || !S_ISREG(inode->i_mode)) {
1129                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1130                         RETURN(0);
1131                 }
1132         } else {
1133                 /* The OST doesn't check permissions, but the alternative is
1134                  * a gratuitous RPC to the MDS.  We already rely on the client
1135                  * to do read/write/truncate permission checks, so is mtime OK?
1136                  */
1137                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1138                         /* from sys_utime() */
1139                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1140                                 if (current->fsuid != inode->i_uid &&
1141                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1142                                         RETURN(rc);
1143                         } else {
1144                                 /* from inode_change_ok() */
1145                                 if (current->fsuid != inode->i_uid &&
1146                                     !capable(CAP_FOWNER))
1147                                         RETURN(-EPERM);
1148                         }
1149                 }
1150
1151                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1152                 err = inode_setattr(inode, attr);
1153                 /*
1154                  * assigning inode_setattr() to @err to disable warning that
1155                  * function's result should be checked by by caller. error is
1156                  * impossible here, as vmtruncate() control path is disabled.
1157                  */
1158         }
1159
1160         /* We really need to get our PW lock before we change inode->i_size.
1161          * If we don't we can race with other i_size updaters on our node, like
1162          * ll_file_read.  We can also race with i_size propogation to other
1163          * nodes through dirtying and writeback of final cached pages.  This
1164          * last one is especially bad for racing o_append users on other
1165          * nodes. */
1166         if (ia_valid & ATTR_SIZE) {
1167                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1168                                                            OBD_OBJECT_EOF } };
1169                 struct lustre_handle lockh = { 0 };
1170                 struct ll_inode_info *lli = ll_i2info(inode);
1171                 int err, ast_flags = 0;
1172                 /* XXX when we fix the AST intents to pass the discard-range
1173                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1174                  * XXX here. */
1175                 if (attr->ia_size == 0)
1176                         ast_flags = LDLM_AST_DISCARD_DATA;
1177
1178                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1179                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1180
1181                 if (rc != 0)
1182                         RETURN(rc);
1183
1184                 down(&lli->lli_size_sem);
1185                 lli->lli_size_pid = current->pid;
1186                 rc = vmtruncate(inode, attr->ia_size);
1187                 if (rc != 0) {
1188                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1189                         lli->lli_size_pid = 0;
1190                         up(&lli->lli_size_sem);
1191                 }
1192
1193                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1194                 if (err) {
1195                         CERROR("ll_extent_unlock failed: %d\n", err);
1196                         if (!rc)
1197                                 rc = err;
1198                 }
1199         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1200                 struct obdo *oa = NULL;
1201
1202                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1203                        inode->i_ino, LTIME_S(attr->ia_mtime));
1204
1205                 oa = obdo_alloc();
1206                 if (oa == NULL)
1207                         RETURN(-ENOMEM);
1208
1209                 oa->o_id = lsm->lsm_object_id;
1210                 oa->o_gr = lsm->lsm_object_gr;
1211                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1212                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1213                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1214                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1215                 obdo_free(oa);
1216                 if (rc)
1217                         CERROR("obd_setattr fails: rc = %d\n", rc);
1218         }
1219         RETURN(rc);
1220 }
1221
1222 int ll_setattr(struct dentry *de, struct iattr *attr)
1223 {
1224         LASSERT(de->d_inode);
1225         return ll_setattr_raw(de->d_inode, attr);
1226 }
1227
1228 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1229                        unsigned long max_age)
1230 {
1231         struct ll_sb_info *sbi = ll_s2sbi(sb);
1232         struct obd_statfs obd_osfs;
1233         int rc;
1234         ENTRY;
1235
1236         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1237         if (rc) {
1238                 CERROR("obd_statfs fails: rc = %d\n", rc);
1239                 RETURN(rc);
1240         }
1241
1242         osfs->os_type = sb->s_magic;
1243
1244         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1245                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1246
1247         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1248         if (rc) {
1249                 CERROR("obd_statfs fails: rc = %d\n", rc);
1250                 RETURN(rc);
1251         }
1252
1253         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1254                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1255                obd_osfs.os_files);
1256
1257         osfs->os_blocks = obd_osfs.os_blocks;
1258         osfs->os_bfree = obd_osfs.os_bfree;
1259         osfs->os_bavail = obd_osfs.os_bavail;
1260
1261         /* If we don't have as many objects free on the OST as inodes
1262          * on the MDS, we reduce the total number of inodes to
1263          * compensate, so that the "inodes in use" number is correct.
1264          */
1265         if (obd_osfs.os_ffree < osfs->os_ffree) {
1266                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1267                         obd_osfs.os_ffree;
1268                 osfs->os_ffree = obd_osfs.os_ffree;
1269         }
1270
1271         RETURN(rc);
1272 }
1273
1274 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1275 {
1276         struct obd_statfs osfs;
1277         int rc;
1278
1279         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1280         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1281
1282         /* For now we will always get up-to-date statfs values, but in the
1283          * future we may allow some amount of caching on the client (e.g.
1284          * from QOS or lprocfs updates). */
1285         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1286         if (rc)
1287                 return rc;
1288
1289         statfs_unpack(sfs, &osfs);
1290
1291         if (sizeof(sfs->f_blocks) == 4) {
1292                 while (osfs.os_blocks > ~0UL) {
1293                         sfs->f_bsize <<= 1;
1294
1295                         osfs.os_blocks >>= 1;
1296                         osfs.os_bfree >>= 1;
1297                         osfs.os_bavail >>= 1;
1298                 }
1299         }
1300
1301         sfs->f_blocks = osfs.os_blocks;
1302         sfs->f_bfree = osfs.os_bfree;
1303         sfs->f_bavail = osfs.os_bavail;
1304
1305         return 0;
1306 }
1307
1308 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1309 {
1310         struct ll_inode_info *lli = ll_i2info(inode);
1311         struct lov_stripe_md *lsm = md->lsm;
1312         struct mds_body *body = md->body;
1313         struct mea *mea = md->mea;
1314         struct posix_acl *ll_acl_access = md->acl_access;
1315         ENTRY;
1316
1317         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1318
1319         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1320                 /* check for default striping info for dir. */
1321                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1322         }
1323
1324         if (lsm != NULL) {
1325                 LASSERT(lsm->lsm_object_gr > 0);
1326                 if (lli->lli_smd == NULL) {
1327                         lli->lli_smd = lsm;
1328                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1329                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1330                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1331                 } else {
1332                         int i;
1333                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1334                                 CERROR("lsm mismatch for inode %ld\n",
1335                                        inode->i_ino);
1336                                 CERROR("lli_smd:\n");
1337                                 dump_lsm(D_ERROR, lli->lli_smd);
1338                                 CERROR("lsm:\n");
1339                                 dump_lsm(D_ERROR, lsm);
1340                                 LBUG();
1341                         }
1342                         /* XXX FIXME -- We should decide on a safer (atomic) and
1343                          * more elegant way to update the lsm */
1344                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1345                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1346                                         lsm->lsm_oinfo[i].loi_id;
1347                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1348                                         lsm->lsm_oinfo[i].loi_gr;
1349                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1350                                         lsm->lsm_oinfo[i].loi_ost_idx;
1351                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1352                                         lsm->lsm_oinfo[i].loi_ost_gen;
1353                         }
1354                 }
1355                 /* bug 2844 - limit i_blksize for broken user-space apps */
1356                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1357                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1358                 if (lli->lli_smd != lsm)
1359                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1360         }
1361
1362         if (mea != NULL) {
1363                 if (lli->lli_mea == NULL) {
1364                         lli->lli_mea = mea;
1365                 } else {
1366                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1367                                 CERROR("mea mismatch for inode %lu\n",
1368                                         inode->i_ino);
1369                                 LBUG();
1370                         }
1371                 }
1372                 if (lli->lli_mea != mea)
1373                         obd_free_memmd(ll_i2mdexp(inode),
1374                                        (struct lov_stripe_md **) &mea);
1375         }
1376
1377         if (body->valid & OBD_MD_FID)
1378                 id_assign_fid(&lli->lli_id, &body->id1);
1379
1380         if (body->valid & OBD_MD_FLID)
1381                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1382
1383         if (body->valid & OBD_MD_FLGENER)
1384                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1385
1386         spin_lock(&lli->lli_lock);
1387         if (ll_acl_access != NULL) {
1388                 if (lli->lli_acl_access != NULL)
1389                         posix_acl_release(lli->lli_acl_access);
1390                 lli->lli_acl_access = ll_acl_access;
1391         }
1392         spin_unlock(&lli->lli_lock);
1393
1394         if (body->valid & OBD_MD_FLID)
1395                 inode->i_ino = id_ino(&body->id1);
1396         if (body->valid & OBD_MD_FLGENER)
1397                 inode->i_generation = id_gen(&body->id1);
1398         if (body->valid & OBD_MD_FLATIME)
1399                 LTIME_S(inode->i_atime) = body->atime;
1400         if (body->valid & OBD_MD_FLMTIME &&
1401             body->mtime > LTIME_S(inode->i_mtime)) {
1402                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1403                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1404                 LTIME_S(inode->i_mtime) = body->mtime;
1405         }
1406         if (body->valid & OBD_MD_FLCTIME &&
1407             body->ctime > LTIME_S(inode->i_ctime))
1408                 LTIME_S(inode->i_ctime) = body->ctime;
1409         if (body->valid & OBD_MD_FLMODE) {
1410                 inode->i_mode = (inode->i_mode & S_IFMT) |
1411                         (body->mode & ~S_IFMT);
1412         }
1413         if (body->valid & OBD_MD_FLTYPE) {
1414                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1415                         (body->mode & S_IFMT);
1416         }
1417         if (body->valid & OBD_MD_FLUID)
1418                 inode->i_uid = body->uid;
1419         if (body->valid & OBD_MD_FLGID)
1420                 inode->i_gid = body->gid;
1421         if (body->valid & OBD_MD_FLFLAGS)
1422                 inode->i_flags = body->flags;
1423         if (body->valid & OBD_MD_FLNLINK)
1424                 inode->i_nlink = body->nlink;
1425         if (body->valid & OBD_MD_FLRDEV)
1426 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1427                 inode->i_rdev = body->rdev;
1428 #else
1429                 inode->i_rdev = old_decode_dev(body->rdev);
1430 #endif
1431         if (body->valid & OBD_MD_FLSIZE)
1432                 inode->i_size = body->size;
1433         if (body->valid & OBD_MD_FLBLOCKS)
1434                 inode->i_blocks = body->blocks;
1435
1436         if (body->valid & OBD_MD_FLSIZE)
1437                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1438
1439 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1440         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1441 #endif
1442         LASSERT(id_fid(&lli->lli_id) != 0);
1443 }
1444
1445 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1446 static struct backing_dev_info ll_backing_dev_info = {
1447         .ra_pages       = 0,    /* No readahead */
1448         .memory_backed  = 0,    /* Does contribute to dirty memory */
1449 };
1450 #endif
1451
1452 void ll_read_inode2(struct inode *inode, void *opaque)
1453 {
1454         struct lustre_md *md = opaque;
1455         struct ll_inode_info *lli = ll_i2info(inode);
1456         ENTRY;
1457
1458         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1459                inode->i_generation, inode);
1460
1461         ll_lli_init(lli);
1462
1463         LASSERT(!lli->lli_smd);
1464
1465         /* Core attributes from the MDS first.  This is a new inode, and
1466          * the VFS doesn't zero times in the core inode so we have to do
1467          * it ourselves.  They will be overwritten by either MDS or OST
1468          * attributes - we just need to make sure they aren't newer. */
1469         LTIME_S(inode->i_mtime) = 0;
1470         LTIME_S(inode->i_atime) = 0;
1471         LTIME_S(inode->i_ctime) = 0;
1472
1473         inode->i_rdev = 0;
1474         ll_update_inode(inode, md);
1475
1476         /* OIDEBUG(inode); */
1477
1478         if (S_ISREG(inode->i_mode)) {
1479                 inode->i_op = &ll_file_inode_operations;
1480                 inode->i_fop = &ll_file_operations;
1481                 inode->i_mapping->a_ops = &ll_aops;
1482                 EXIT;
1483         } else if (S_ISDIR(inode->i_mode)) {
1484                 inode->i_op = &ll_dir_inode_operations;
1485                 inode->i_fop = &ll_dir_operations;
1486                 inode->i_mapping->a_ops = &ll_dir_aops;
1487                 EXIT;
1488         } else if (S_ISLNK(inode->i_mode)) {
1489                 inode->i_op = &ll_fast_symlink_inode_operations;
1490                 EXIT;
1491         } else {
1492                 inode->i_op = &ll_special_inode_operations;
1493
1494 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1495                 init_special_inode(inode, inode->i_mode,
1496                                    kdev_t_to_nr(inode->i_rdev));
1497
1498                 /* initializing backing dev info. */
1499                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1500 #else
1501                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1502 #endif
1503                 lli->ll_save_ifop = inode->i_fop;
1504
1505                 if (S_ISCHR(inode->i_mode))
1506                         inode->i_fop = &ll_special_chr_inode_fops;
1507                 else if (S_ISBLK(inode->i_mode))
1508                         inode->i_fop = &ll_special_blk_inode_fops;
1509                 else if (S_ISFIFO(inode->i_mode))
1510                         inode->i_fop = &ll_special_fifo_inode_fops;
1511                 else if (S_ISSOCK(inode->i_mode))
1512                         inode->i_fop = &ll_special_sock_inode_fops;
1513
1514                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1515                       inode->i_fop);
1516
1517                 if (lli->ll_save_ifop->owner) {
1518                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1519                               lli->ll_save_ifop->owner);
1520                 }
1521                 EXIT;
1522         }
1523 }
1524
1525 void ll_delete_inode(struct inode *inode)
1526 {
1527         struct ll_sb_info *sbi = ll_i2sbi(inode);
1528         struct lustre_id id;
1529         int rc;
1530         ENTRY;
1531
1532         ll_inode2id(&id, inode);
1533
1534         rc = md_delete_inode(sbi->ll_md_exp, &id);
1535         if (rc) {
1536                 CERROR("md_delete_inode() failed, error %d\n",
1537                        rc);
1538         }
1539
1540         clear_inode(inode);
1541         EXIT;
1542 }
1543
1544 int ll_iocontrol(struct inode *inode, struct file *file,
1545                  unsigned int cmd, unsigned long arg)
1546 {
1547         struct ll_sb_info *sbi = ll_i2sbi(inode);
1548         struct ptlrpc_request *req = NULL;
1549         int rc, flags = 0;
1550         ENTRY;
1551
1552         switch(cmd) {
1553         case EXT3_IOC_GETFLAGS: {
1554                 struct lustre_id id;
1555                 __u64 valid = OBD_MD_FLFLAGS;
1556                 struct mds_body *body;
1557
1558                 ll_inode2id(&id, inode);
1559                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1560                 if (rc) {
1561                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1562                         RETURN(-abs(rc));
1563                 }
1564
1565                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1566
1567                 if (body->flags & S_APPEND)
1568                         flags |= EXT3_APPEND_FL;
1569                 if (body->flags & S_IMMUTABLE)
1570                         flags |= EXT3_IMMUTABLE_FL;
1571                 if (body->flags & S_NOATIME)
1572                         flags |= EXT3_NOATIME_FL;
1573
1574                 ptlrpc_req_finished (req);
1575
1576                 RETURN(put_user(flags, (int *)arg));
1577         }
1578         case EXT3_IOC_SETFLAGS: {
1579                 struct mdc_op_data *op_data;
1580                 struct iattr attr;
1581                 struct obdo *oa;
1582                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1583
1584                 if (get_user(flags, (int *)arg))
1585                         RETURN(-EFAULT);
1586
1587                 oa = obdo_alloc();
1588                 if (!oa)
1589                         RETURN(-ENOMEM);
1590
1591                 OBD_ALLOC(op_data, sizeof(*op_data));
1592                 if (op_data == NULL) {
1593                         obdo_free(oa);
1594                         RETURN(-ENOMEM);
1595                 }
1596                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1597
1598                 memset(&attr, 0x0, sizeof(attr));
1599                 attr.ia_attr_flags = flags;
1600                 attr.ia_valid |= ATTR_ATTR_FLAG;
1601
1602                 rc = md_setattr(sbi->ll_md_exp, op_data,
1603                                 &attr, NULL, 0, NULL, 0, &req);
1604                 OBD_FREE(op_data, sizeof(*op_data));
1605                 if (rc) {
1606                         ptlrpc_req_finished(req);
1607                         if (rc != -EPERM && rc != -EACCES)
1608                                 CERROR("md_setattr fails: rc = %d\n", rc);
1609                         obdo_free(oa);
1610                         RETURN(rc);
1611                 }
1612                 ptlrpc_req_finished(req);
1613
1614                 oa->o_id = lsm->lsm_object_id;
1615                 oa->o_gr = lsm->lsm_object_gr;
1616                 oa->o_flags = flags;
1617                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1618
1619                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1620                 obdo_free(oa);
1621                 if (rc) {
1622                         if (rc != -EPERM && rc != -EACCES)
1623                                 CERROR("md_setattr fails: rc = %d\n", rc);
1624                         RETURN(rc);
1625                 }
1626
1627                 if (flags & EXT3_APPEND_FL)
1628                         inode->i_flags |= S_APPEND;
1629                 else
1630                         inode->i_flags &= ~S_APPEND;
1631                 if (flags & EXT3_IMMUTABLE_FL)
1632                         inode->i_flags |= S_IMMUTABLE;
1633                 else
1634                         inode->i_flags &= ~S_IMMUTABLE;
1635                 if (flags & EXT3_NOATIME_FL)
1636                         inode->i_flags |= S_NOATIME;
1637                 else
1638                         inode->i_flags &= ~S_NOATIME;
1639
1640                 RETURN(0);
1641         }
1642         default:
1643                 RETURN(-ENOSYS);
1644         }
1645
1646         RETURN(0);
1647 }
1648
1649 /* this is only called in the case of forced umount. */
1650 void ll_umount_begin(struct super_block *sb)
1651 {
1652         struct ll_sb_info *sbi = ll_s2sbi(sb);
1653         struct obd_ioctl_data ioc_data = { 0 };
1654         struct obd_device *obd;
1655         ENTRY;
1656
1657         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1658                sb->s_count, atomic_read(&sb->s_active));
1659
1660         obd = class_exp2obd(sbi->ll_md_exp);
1661         if (obd == NULL) {
1662                 CERROR("Invalid MDC connection handle "LPX64"\n",
1663                        sbi->ll_md_exp->exp_handle.h_cookie);
1664                 EXIT;
1665                 return;
1666         }
1667         obd->obd_no_recov = 1;
1668         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1669                       sizeof(ioc_data), &ioc_data, NULL);
1670
1671         obd = class_exp2obd(sbi->ll_dt_exp);
1672         if (obd == NULL) {
1673                 CERROR("Invalid LOV connection handle "LPX64"\n",
1674                        sbi->ll_dt_exp->exp_handle.h_cookie);
1675                 EXIT;
1676                 return;
1677         }
1678
1679         obd->obd_no_recov = 1;
1680         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1681                       sizeof(ioc_data), &ioc_data, NULL);
1682
1683         /*
1684          * really, we'd like to wait until there are no requests outstanding,
1685          * and then continue.  For now, we just invalidate the requests,
1686          * schedule, and hope.
1687          */
1688         schedule();
1689
1690         EXIT;
1691 }
1692
1693 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1694                   struct inode **inode, struct ptlrpc_request *req,
1695                   int offset, struct super_block *sb)
1696 {
1697         struct lustre_md md;
1698         int rc = 0;
1699
1700         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1701         if (rc)
1702                 RETURN(rc);
1703
1704         if (*inode) {
1705                 ll_update_inode(*inode, &md);
1706         } else {
1707                 LASSERT(sb);
1708                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1709                 if (*inode == NULL || is_bad_inode(*inode)) {
1710                         /* free the lsm if we allocated one above */
1711                         if (md.lsm != NULL)
1712                                 obd_free_memmd(dt_exp, &md.lsm);
1713                         if (md.mea != NULL)
1714                                 obd_free_memmd(md_exp,
1715                                                (struct lov_stripe_md**)&md.mea);
1716                         rc = -ENOMEM;
1717                         CERROR("new_inode -fatal: rc %d\n", rc);
1718                 }
1719         }
1720
1721         RETURN(rc);
1722 }
1723
1724 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1725                char *filename, struct lustre_id *ret)
1726 {
1727         struct ptlrpc_request *request = NULL;
1728         struct mds_body *body;
1729         int rc;
1730
1731         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1732                              OBD_MD_FID, 0, &request);
1733         if (rc < 0) {
1734                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1735                        filename, rc);
1736                 return rc;
1737         }
1738
1739         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1740         LASSERT(body != NULL);
1741         LASSERT_REPSWABBED(request, 0);
1742
1743         *ret = body->id1;
1744         ptlrpc_req_finished(request);
1745
1746         return rc;
1747 }