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 #include <linux/seq_file.h>
  31
  32 #include <linux/lustre_lite.h>
  33 #include <linux/lustre_ha.h>
  34 #include <linux/lustre_dlm.h>
  35 #include <linux/lprocfs_status.h>
  36 #include <linux/lustre_acl.h>
  37 #include <linux/lustre_audit.h>
  38 #include <linux/lustre_gs.h>
  39 #include <linux/lustre_sec.h>
  40 #include "llite_internal.h"
  41
  42 kmem_cache_t *ll_file_data_slab;
  43 kmem_cache_t *ll_intent_slab;
  44
  45 extern struct address_space_operations ll_aops;
  46 extern struct address_space_operations ll_dir_aops;
  47
  48 #ifndef log2
  49 #define log2(n) ffz(~(n))
  50 #endif
  51
  52 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
  53 {
  54         struct ll_sb_info *sbi = NULL;
  55         class_uuid_t uuid;
  56         ENTRY;
  57
  58         OBD_ALLOC(sbi, sizeof(*sbi));
  59         if (!sbi)
  60                 RETURN(NULL);
  61
  62         spin_lock_init(&sbi->ll_lock);
  63         INIT_LIST_HEAD(&sbi->ll_pglist);
  64         sbi->ll_pglist_gen = 0;
  65         if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
  66                 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
  67         else
  68                 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
  69         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  70         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  71         INIT_LIST_HEAD(&sbi->ll_mnt_list);
  72
  73         sema_init(&sbi->ll_gns_sem, 1);
  74         spin_lock_init(&sbi->ll_gns_lock);
  75         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
  76         init_waitqueue_head(&sbi->ll_gns_waitq);
  77         init_completion(&sbi->ll_gns_mount_finished);
  78
  79         /* this later may be reset via /proc/fs/... */
  80         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
  81         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
  82
  83         /* this later may be reset via /proc/fs/... */
  84         memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
  85                strlen("/usr/sbin/gns_upcall"));
  86         sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
  87
  88         /* default values, may be changed via /proc/fs/... */
  89         sbi->ll_gns_state = LL_GNS_IDLE;
  90         sbi->ll_gns_pending_dentry = NULL;
  91         atomic_set(&sbi->ll_gns_enabled, 1);
  92         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  93         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  94
  95         sbi->ll_gns_timer.data = (unsigned long)sbi;
  96         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  97         init_timer(&sbi->ll_gns_timer);
  98         //audit mask
  99         sbi->ll_audit_mask = AUDIT_OFF;
 100         ll_set_sbi(sb, sbi);
 101
 102         generate_random_uuid(uuid);
 103         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 104         RETURN(sbi);
 105 }
 106
 107 void lustre_free_sbi(struct super_block *sb)
 108 {
 109         struct ll_sb_info *sbi = ll_s2sbi(sb);
 110         ENTRY;
 111
 112         if (sbi != NULL) {
 113                 list_del(&sbi->ll_gns_sbi_head);
 114                 del_timer(&sbi->ll_gns_timer);
 115                 OBD_FREE(sbi, sizeof(*sbi));
 116         }
 117         ll_set_sbi(sb, NULL);
 118         EXIT;
 119 }
 120
 121 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 122 {
 123         __u32 valsize;
 124         int rc = 0;
 125         ENTRY;
 126
 127         valsize = sizeof(sbi->ll_dt_desc);
 128         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 129         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 130                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 131         RETURN(rc);
 132 }
 133
 134 static int lustre_connect_mds(struct super_block *sb, char *lmv,
 135                               struct obd_connect_data *data,
 136                               char *mds_security, int async, int pag)
 137 {
 138         struct ll_sb_info *sbi = ll_s2sbi(sb);
 139         struct lustre_handle md_conn = {0, };
 140         struct obd_device *md_obd;
 141         struct obd_statfs osfs;
 142         unsigned long sec_flags;
 143         __u32 valsize;
 144         int    err = 0;
 145         ENTRY;
 146
 147         md_obd = class_name2obd(lmv);
 148         if (!md_obd) {
 149                 CERROR("MDC %s: not setup or attached\n", lmv);
 150                 RETURN(-EINVAL);
 151         }
 152
 153         obd_set_info(md_obd->obd_self_export, strlen("async"), "async",
 154                      sizeof(async), &async);
 155
 156         err = obd_set_info(md_obd->obd_self_export, strlen("sec"), "sec",
 157                            strlen(mds_security), mds_security);
 158
 159         if (err) {
 160                 CERROR("LMV %s: failed to set security %s, err %d\n",
 161                         lmv, mds_security, err);
 162                 RETURN(err);
 163         }
 164
 165         if (pag) {
 166                 sec_flags = PTLRPC_SEC_FL_PAG;
 167                 err = obd_set_info(md_obd->obd_self_export,
 168                                    strlen("sec_flags"), "sec_flags",
 169                                    sizeof(sec_flags), &sec_flags);
 170                 if (err) {
 171                         OBD_FREE(data, sizeof(*data));
 172                         RETURN(err);
 173                 }
 174         }
 175
 176         err = obd_connect(&md_conn, md_obd, &sbi->ll_sb_uuid, data,
 177                           OBD_OPT_REAL_CLIENT);
 178         if (err == -EBUSY) {
 179                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 180                        " client is not a part.  Please wait for recovery to "
 181                        "complete, abort, or time out.\n", lmv);
 182                 GOTO(out, err);
 183         } else if (err) {
 184                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 185                 GOTO(out, err);
 186         }
 187
 188         sbi->ll_md_exp = class_conn2export(&md_conn);
 189
 190         err = obd_statfs(md_obd, &osfs, jiffies - HZ);
 191         if (err)
 192                 GOTO(out_disconnect, err);
 193
 194         if (!osfs.os_bsize) {
 195                 CERROR("Invalid block size is detected.");
 196                 GOTO(out_disconnect, err);
 197         }
 198
 199         sb->s_magic = LL_SUPER_MAGIC;
 200         sb->s_blocksize = osfs.os_bsize;
 201         sb->s_blocksize_bits = log2(osfs.os_bsize);
 202         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 203
 204         /* in 2.6.x FS is not allowed to form s_dev */
 205 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 206         {
 207                 kdev_t devno;
 208
 209                 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 210                                      strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 211
 212                 sb->s_dev = devno;
 213         }
 214 #endif
 215
 216         /* after statfs, we are supposed to have connected to MDSs,
 217          * so it's ok to check remote flag returned.
 218          */
 219         valsize = sizeof(&sbi->ll_remote);
 220         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 221                            &valsize, &sbi->ll_remote);
 222         if (err) {
 223                 CERROR("fail to obtain remote flag\n");
 224                 GOTO(out_disconnect, err);
 225         }
 226
 227 out_disconnect:
 228         if (err)
 229                 obd_disconnect(sbi->ll_md_exp, 0);
 230 out:
 231         RETURN(err);
 232 }
 233 static int lustre_connect_ost(struct super_block *sb, char *lov,
 234                               struct obd_connect_data *data,
 235                               char *oss_security, int async, int pag)
 236 {
 237         struct ll_sb_info *sbi = ll_s2sbi(sb);
 238         struct lustre_handle dt_conn = {0, };
 239         struct obd_device *obd = NULL;
 240         unsigned long sec_flags;
 241         int err, mdsize;
 242
 243         obd = class_name2obd(lov);
 244         if (!obd) {
 245                 CERROR("OSC %s: not setup or attached\n", lov);
 246                 GOTO(out, err = -EINVAL);
 247         }
 248         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 249                      sizeof(async), &async);
 250
 251        if (oss_security == NULL)
 252                 oss_security = "null";
 253
 254         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 255                            strlen(oss_security), oss_security);
 256         if (err) {
 257                 CERROR("LOV %s: failed to set security %s, err %d\n",
 258                         lov, oss_security, err);
 259                 RETURN(err);
 260         }
 261
 262         if (pag) {
 263                 sec_flags = PTLRPC_SEC_FL_PAG;
 264                 err = obd_set_info(obd->obd_self_export,
 265                                    strlen("sec_flags"), "sec_flags",
 266                                    sizeof(sec_flags), &sec_flags);
 267                 if (err) {
 268                         OBD_FREE(data, sizeof(*data));
 269                         RETURN(err);
 270                 }
 271         }
 272
 273         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
 274         if (err == -EBUSY) {
 275                 CERROR("An OST (lov %s) is performing recovery, of which this"
 276                        " client is not a part.  Please wait for recovery to "
 277                        "complete, abort, or time out.\n", lov);
 278                 GOTO(out, err);
 279         } else if (err) {
 280                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 281                 GOTO(out, err);
 282         }
 283         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 284
 285         err = lustre_init_dt_desc(sbi);
 286
 287         if (err) {
 288                 CWARN("init dt_desc error %d \n", err);
 289                 GOTO(out, err = 0);
 290         }
 291         mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 292         obd_init_ea_size(sbi->ll_md_exp, mdsize, sbi->ll_dt_desc.ld_tgt_count *
 293                          sizeof(struct llog_cookie));
 294 out:
 295         RETURN(err);
 296 }
 297
 298 extern struct dentry_operations ll_d_ops;
 299
 300 static int lustre_init_root_inode(struct super_block *sb)
 301 {
 302         struct ll_sb_info *sbi = ll_s2sbi(sb);
 303         struct ptlrpc_request *request = NULL;
 304         struct inode *root = NULL;
 305         struct lustre_md md;
 306         int err = 0;
 307         ENTRY;
 308
 309         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 310         if (err) {
 311                 CERROR("cannot mds_connect: rc = %d\n", err);
 312                 GOTO(out, err);
 313         }
 314         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 315
 316         sb->s_op = &lustre_super_operations;
 317
 318         /* make root inode */
 319         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 320                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
 321                          NULL, NULL, 0, 0, NULL, &request);
 322         if (err) {
 323                 CERROR("md_getattr failed for root: rc = %d\n", err);
 324                 GOTO(out, err);
 325         }
 326
 327         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 328                                 sbi->ll_dt_exp, &md);
 329         if (err) {
 330                 CERROR("failed to understand root inode md: rc = %d\n", err);
 331                 ptlrpc_req_finished(request);
 332                 GOTO(out, err);
 333         }
 334
 335         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 336         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 337
 338         ptlrpc_req_finished(request);
 339
 340         if (root == NULL || is_bad_inode(root)) {
 341                 if (md.lsm != NULL)
 342                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 343                 if (md.mea != NULL)
 344                     obd_free_memmd(sbi->ll_md_exp,
 345                                    (struct lov_stripe_md**)&md.mea);
 346                 CERROR("lustre_lite: bad iget4 for root\n");
 347                 GOTO(out_root, err = -EBADF);
 348         }
 349         sb->s_root = d_alloc_root(root);
 350         sb->s_root->d_op = &ll_d_ops;
 351 out_root:
 352         if (err)
 353                 iput(root);
 354 out:
 355         RETURN(err);
 356 }
 357
 358 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 359                              char *gkc, int async,  char *mds_security,
 360                              char *oss_security, __u32 *nllu, int pag,
 361                              __u64 *remote)
 362 {
 363         struct ll_sb_info *sbi = ll_s2sbi(sb);
 364         struct obd_connect_data *data;
 365         int err;
 366         ENTRY;
 367
 368         /*process the connect flags*/
 369         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 370                        (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 371                 CERROR("wrong remote flag "LPX64"\n", *remote);
 372                 RETURN(-EINVAL);
 373         }
 374
 375         OBD_ALLOC(data, sizeof(*data));
 376         if (!data)
 377                 RETURN(-ENOMEM);
 378
 379         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 380                                               OBD_CONNECT_REMOTE);
 381         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 382
 383         if (proc_lustre_fs_root) {
 384                 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
 385                                                   sb, lov, lmv);
 386                 if (err < 0)
 387                         CERROR("could not register mount in /proc/lustre");
 388         }
 389
 390         /*connect mds */
 391         err = lustre_connect_mds(sb, lmv, data, mds_security, async, pag);
 392         if (err)
 393                 GOTO(out, err);
 394
 395         /*connect OST*/
 396         err = lustre_connect_ost(sb, lov, data, oss_security, async, pag);
 397         if (err)
 398                 GOTO(out_lmv, err);
 399
 400         err = lustre_init_crypto(sb, gkc, data, async);
 401         if (err) {
 402                 CERROR("Could not connect to GSS err %d\n", err);
 403                 err = 0;
 404         }
 405         /*connect GSS*/
 406         err = lustre_init_root_inode(sb);
 407         if (err)
 408                 GOTO(out_gks, err);
 409
 410         err = ll_close_thread_start(&sbi->ll_lcq);
 411         if (err) {
 412                 CERROR("cannot start close thread: rc %d\n", err);
 413                 GOTO(out_root, err);
 414         }
 415
 416         ll_gns_add_timer(sbi);
 417
 418         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 419            backing dev info assigned to inode mapping is used for
 420            determining maximal readahead. */
 421 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 422     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 423         /* bug 2805 - set VM readahead to zero */
 424         vm_max_readahead = vm_min_readahead = 0;
 425 #endif
 426         sb->s_flags |= MS_POSIXACL;
 427 #ifdef S_PDIROPS
 428         CWARN("Enabling PDIROPS\n");
 429         sb->s_flags |= S_PDIROPS;
 430 #endif
 431         if (data != NULL)
 432                 OBD_FREE(data, sizeof(*data));
 433         RETURN(err);
 434 out_root:
 435         if (sb->s_root)
 436                 dput(sb->s_root);
 437 out_gks:
 438         lustre_destroy_crypto(sb);
 439 out_lmv:
 440         obd_disconnect(sbi->ll_md_exp, 0);
 441 out:
 442         if (data != NULL)
 443                 OBD_FREE(data, sizeof(*data));
 444         lprocfs_unregister_mountpoint(sbi);
 445         RETURN(err);
 446 }
 447
 448 void lustre_common_put_super(struct super_block *sb)
 449 {
 450         struct ll_sb_info *sbi = ll_s2sbi(sb);
 451         struct hlist_node *tmp, *next;
 452         ENTRY;
 453
 454         ll_gns_del_timer(sbi);
 455         ll_close_thread_stop(sbi->ll_lcq);
 456
 457         lustre_destroy_crypto(sb);
 458
 459         list_del(&sbi->ll_conn_chain);
 460         obd_disconnect(sbi->ll_dt_exp, 0);
 461
 462         lprocfs_unregister_mountpoint(sbi);
 463         if (sbi->ll_proc_root) {
 464                 lprocfs_remove(sbi->ll_proc_root);
 465                 sbi->ll_proc_root = NULL;
 466         }
 467
 468         obd_disconnect(sbi->ll_md_exp, 0);
 469
 470         // We do this to get rid of orphaned dentries. That is not really trw.
 471         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 472                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 473                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 474                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 475                 shrink_dcache_parent(dentry);
 476         }
 477         EXIT;
 478 }
 479
 480 char *ll_read_opt(const char *opt, char *data)
 481 {
 482         char *value;
 483         char *retval;
 484         ENTRY;
 485
 486         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 487         if (strncmp(opt, data, strlen(opt)))
 488                 RETURN(NULL);
 489         if ((value = strchr(data, '=')) == NULL)
 490                 RETURN(NULL);
 491
 492         value++;
 493         OBD_ALLOC(retval, strlen(value) + 1);
 494         if (!retval) {
 495                 CERROR("out of memory!\n");
 496                 RETURN(NULL);
 497         }
 498
 499         memcpy(retval, value, strlen(value)+1);
 500         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 501         RETURN(retval);
 502 }
 503
 504 int ll_set_opt(const char *opt, char *data, int fl)
 505 {
 506         ENTRY;
 507
 508         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 509         if (strncmp(opt, data, strlen(opt)))
 510                 RETURN(0);
 511         else
 512                 RETURN(fl);
 513 }
 514
 515 void ll_options(char *options, char **lov, char **lmv, char **gkc,
 516                 char **mds_sec, char **oss_sec, int *async, int *flags)
 517 {
 518         char *this_char;
 519 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 520         char *opt_ptr = options;
 521 #endif
 522         ENTRY;
 523
 524         if (!options) {
 525                 EXIT;
 526                 return;
 527         }
 528
 529         *async = 0;
 530 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 531         for (this_char = strtok (options, ",");
 532              this_char != NULL;
 533              this_char = strtok (NULL, ",")) {
 534 #else
 535         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 536 #endif
 537                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 538                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 539                         continue;
 540                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 541                         continue;
 542                 if (!*gkc && (*gkc = ll_read_opt("gkc", this_char)))
 543                         continue;
 544                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 545                         *async = 1;
 546                         continue;
 547                 }
 548                 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
 549                         continue;
 550                 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
 551                         continue;
 552                 if (!(*flags & LL_SBI_NOLCK) &&
 553                     ((*flags) = (*flags) |
 554                                 ll_set_opt("nolock", this_char,
 555                                            LL_SBI_NOLCK)))
 556                         continue;
 557         }
 558
 559         EXIT;
 560 }
 561
 562 void ll_lli_init(struct ll_inode_info *lli)
 563 {
 564         sema_init(&lli->lli_open_sem, 1);
 565         sema_init(&lli->lli_size_sem, 1);
 566         lli->lli_flags = 0;
 567         lli->lli_size_pid = 0;
 568         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 569         spin_lock_init(&lli->lli_lock);
 570         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 571         INIT_LIST_HEAD(&lli->lli_close_item);
 572         lli->lli_inode_magic = LLI_INODE_MAGIC;
 573         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 574         sema_init(&lli->lli_och_sem, 1);
 575         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 576         lli->lli_mds_exec_och = NULL;
 577         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 578         lli->lli_open_fd_exec_count = 0;
 579         lli->lli_audit_mask = AUDIT_OFF;
 580         lli->lli_key_info = NULL;
 581 }
 582
 583 int ll_fill_super(struct super_block *sb, void *data, int silent)
 584 {
 585         struct ll_sb_info *sbi;
 586         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 587         char *mds_sec = NULL;
 588         char *oss_sec = NULL;
 589         int async, err;
 590         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 591         __u64 remote_flag = 0;
 592         ENTRY;
 593
 594         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 595
 596         sbi = lustre_init_sbi(sb);
 597         if (!sbi)
 598                 RETURN(-ENOMEM);
 599
 600         sbi->ll_flags |= LL_SBI_READAHEAD;
 601         ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
 602                    &async, &sbi->ll_flags);
 603
 604         if (!lov || !lmv) {
 605                 CERROR("no osc %p or no mdc %p\n", lov, lmv);
 606                 GOTO(out, err = -EINVAL);
 607         }
 608
 609         err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
 610                                        oss_sec, nllu, 0, &remote_flag);
 611         EXIT;
 612 out:
 613         if (err)
 614                 lustre_free_sbi(sb);
 615
 616         if (lmv)
 617                 OBD_FREE(lmv, strlen(lmv) + 1);
 618         if (lov)
 619                 OBD_FREE(lov, strlen(lov) + 1);
 620         if (mds_sec)
 621                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 622         if (oss_sec)
 623                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 624         if (gkc)
 625                 OBD_FREE(gkc, strlen(gkc) + 1);
 626
 627         return err;
 628 } /* ll_read_super */
 629
 630 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 631                               struct config_llog_instance *cfg, int allow_recov)
 632 {
 633         struct lustre_cfg *lcfg = NULL;
 634         struct lustre_cfg_bufs bufs;
 635         struct portals_cfg pcfg;
 636         char *peer = "MDS_PEER_UUID";
 637         struct obd_device *obd;
 638         struct lustre_handle md_conn = {0, };
 639         struct obd_export *exp;
 640         char *name = "mdc_dev";
 641         class_uuid_t uuid;
 642         struct obd_uuid lmv_uuid;
 643         struct llog_ctxt *ctxt;
 644         int rc = 0, err = 0;
 645         ENTRY;
 646
 647         if (lmd_bad_magic(lmd))
 648                 RETURN(-EINVAL);
 649
 650         generate_random_uuid(uuid);
 651         class_uuid_unparse(uuid, &lmv_uuid);
 652
 653         if (lmd->lmd_local_nid) {
 654                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 655                 pcfg.pcfg_nal = lmd->lmd_nal;
 656                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 657                 rc = libcfs_nal_cmd(&pcfg);
 658                 if (rc < 0)
 659                         GOTO(out, rc);
 660         }
 661
 662         if (lmd->lmd_nal == SOCKNAL ||
 663             lmd->lmd_nal == OPENIBNAL ||
 664             lmd->lmd_nal == IIBNAL ||
 665             lmd->lmd_nal == VIBNAL ||
 666             lmd->lmd_nal == RANAL) {
 667                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 668                 pcfg.pcfg_nal     = lmd->lmd_nal;
 669                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 670                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 671                 pcfg.pcfg_misc    = lmd->lmd_port;
 672                 rc = libcfs_nal_cmd(&pcfg);
 673                 if (rc < 0)
 674                         GOTO(out, rc);
 675         }
 676         lustre_cfg_bufs_reset(&bufs, name);
 677         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 678
 679         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 680         lcfg->lcfg_nal = lmd->lmd_nal;
 681         lcfg->lcfg_nid = lmd->lmd_server_nid;
 682         LASSERT(lcfg->lcfg_nal);
 683         LASSERT(lcfg->lcfg_nid);
 684         err = class_process_config(lcfg);
 685         lustre_cfg_free(lcfg);
 686         if (err < 0)
 687                 GOTO(out_del_conn, err);
 688
 689         lustre_cfg_bufs_reset(&bufs, name);
 690         lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
 691         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 692
 693         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 694         err = class_process_config(lcfg);
 695         lustre_cfg_free(lcfg);
 696         if (err < 0)
 697                 GOTO(out_del_uuid, err);
 698
 699         lustre_cfg_bufs_reset(&bufs, name);
 700         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 701         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 702
 703         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 704         err = class_process_config(lcfg);
 705         lustre_cfg_free(lcfg);
 706         if (err < 0)
 707                 GOTO(out_detach, err);
 708
 709         obd = class_name2obd(name);
 710         if (obd == NULL)
 711                 GOTO(out_cleanup, rc = -EINVAL);
 712
 713         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 714                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 715         if (rc)
 716                 GOTO(out_cleanup, rc);
 717
 718         if (lmd->lmd_pag) {
 719                 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
 720                 rc = obd_set_info(obd->obd_self_export,
 721                                   strlen("sec_flags"), "sec_flags",
 722                                   sizeof(sec_flags), &sec_flags);
 723                 if (rc)
 724                         GOTO(out_cleanup, rc);
 725         }
 726
 727         /* Disable initial recovery on this import */
 728         rc = obd_set_info(obd->obd_self_export,
 729                           strlen("initial_recov"), "initial_recov",
 730                           sizeof(allow_recov), &allow_recov);
 731         if (rc)
 732                 GOTO(out_cleanup, rc);
 733
 734         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 735         if (rc) {
 736                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 737                 GOTO(out_cleanup, rc);
 738         }
 739
 740         exp = class_conn2export(&md_conn);
 741
 742         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 743         rc = class_config_process_llog(ctxt, profile, cfg);
 744         if (rc)
 745                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 746
 747         err = obd_disconnect(exp, 0);
 748
 749         EXIT;
 750 out_cleanup:
 751         lustre_cfg_bufs_reset(&bufs, name);
 752         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 753         err = class_process_config(lcfg);
 754         lustre_cfg_free(lcfg);
 755         if (err < 0)
 756                 GOTO(out, err);
 757 out_detach:
 758         lustre_cfg_bufs_reset(&bufs, name);
 759         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 760         err = class_process_config(lcfg);
 761         lustre_cfg_free(lcfg);
 762         if (err < 0)
 763                 GOTO(out, err);
 764
 765 out_del_uuid:
 766         lustre_cfg_bufs_reset(&bufs, name);
 767         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 768         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 769         err = class_process_config(lcfg);
 770         lustre_cfg_free(lcfg);
 771
 772 out_del_conn:
 773         if (lmd->lmd_nal == SOCKNAL ||
 774             lmd->lmd_nal == OPENIBNAL ||
 775             lmd->lmd_nal == IIBNAL ||
 776             lmd->lmd_nal == VIBNAL ||
 777             lmd->lmd_nal == RANAL) {
 778                 int err2;
 779
 780                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 781                 pcfg.pcfg_nal     = lmd->lmd_nal;
 782                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 783                 pcfg.pcfg_flags   = 1;          /* single_share */
 784                 err2 = libcfs_nal_cmd(&pcfg);
 785                 if (err2 && !err)
 786                         err = err2;
 787                 if (err < 0)
 788                         GOTO(out, err);
 789         }
 790 out:
 791         if (rc == 0)
 792                 rc = err;
 793
 794         return rc;
 795 }
 796
 797 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 798 {
 799         struct lustre_cfg *lcfg;
 800         struct lustre_cfg_bufs bufs;
 801         struct obd_device *obd;
 802         int next = 0;
 803
 804         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
 805                 int err;
 806
 807                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 808                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 809                 err = class_process_config(lcfg);
 810                 if (err) {
 811                         CERROR("cleanup failed: %s\n", obd->obd_name);
 812                         //continue;
 813                 }
 814
 815                 lcfg->lcfg_command = LCFG_DETACH;
 816                 err = class_process_config(lcfg);
 817                 lustre_cfg_free(lcfg);
 818                 if (err) {
 819                         CERROR("detach failed: %s\n", obd->obd_name);
 820                         //continue;
 821                 }
 822         }
 823
 824         if (sbi->ll_lmd != NULL)
 825                 class_del_profile(sbi->ll_lmd->lmd_profile);
 826 }
 827
 828 static int lustre_process_profile(struct super_block *sb,
 829                                   struct lustre_mount_data *lmd,
 830                                   char **lov, char **lmv, char **gkc)
 831 {
 832         struct ll_sb_info *sbi = ll_s2sbi(sb);
 833         struct lustre_profile *lprof;
 834         struct config_llog_instance cfg;
 835         int len, err = 0;
 836
 837         ENTRY;
 838
 839         if (!lmd->lmd_profile)
 840                 RETURN(0);
 841
 842         if (lmd->lmd_mds[0] == '\0') {
 843                 CERROR("no mds name\n");
 844                 GOTO(out, err = -EINVAL);
 845         }
 846         lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
 847         lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
 848
 849         OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 850         if (sbi->ll_lmd == NULL)
 851                 GOTO(out, err = -ENOMEM);
 852         memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 853
 854         /* generate a string unique to this super, let's try
 855          the address of the super itself.*/
 856         len = (sizeof(sb) * 2) + 1;
 857         OBD_ALLOC(sbi->ll_instance, len);
 858         if (sbi->ll_instance == NULL)
 859                 GOTO(out, err = -ENOMEM);
 860         sprintf(sbi->ll_instance, "%p", sb);
 861
 862         cfg.cfg_instance = sbi->ll_instance;
 863         cfg.cfg_uuid = sbi->ll_sb_uuid;
 864         cfg.cfg_local_nid = lmd->lmd_local_nid;
 865         err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 866         if (err < 0) {
 867                 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 868                 GOTO(out, err);
 869         }
 870
 871         lprof = class_get_profile(lmd->lmd_profile);
 872         if (lprof == NULL) {
 873                 CERROR("No profile found: %s\n", lmd->lmd_profile);
 874                 GOTO(out, err = -EINVAL);
 875         }
 876
 877         OBD_ALLOC(*lov, strlen(lprof->lp_lov) +
 878                   strlen(sbi->ll_instance) + 2);
 879         sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 880
 881         OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
 882                   strlen(sbi->ll_instance) + 2);
 883         sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 884
 885         if (lprof->lp_gkc) {
 886                OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
 887                                strlen(sbi->ll_instance) + 2);
 888                sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
 889         }
 890 out:
 891         RETURN(err);
 892 }
 893
 894 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
 895 {
 896         struct lustre_mount_data *lmd = sbi->ll_lmd;
 897         struct config_llog_instance cfg;
 898         char *cl_prof;
 899         int len, err = 0;
 900         ENTRY;
 901
 902         if (!lmd)
 903                 RETURN(err);
 904
 905         len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
 906
 907         if (force_umount) {
 908                 CERROR("force umount, doing manual cleanup\n");
 909                 lustre_manual_cleanup(sbi);
 910                 GOTO(free_lmd, 0);
 911
 912         }
 913         if (sbi->ll_instance != NULL) {
 914                 cfg.cfg_instance = sbi->ll_instance;
 915                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 916
 917                 OBD_ALLOC(cl_prof, len);
 918                 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 919                 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 920                 if (err < 0) {
 921                         CERROR("Unable to process log: %s\n", cl_prof);
 922                         lustre_manual_cleanup(sbi);
 923                 }
 924                 OBD_FREE(cl_prof, len);
 925         }
 926 free_lmd:
 927         if (sbi->ll_instance)
 928                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 929         OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 930         RETURN(err);
 931 }
 932
 933 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 934 {
 935         struct lustre_mount_data * lmd = data;
 936         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 937         struct ll_sb_info *sbi;
 938         int err;
 939         ENTRY;
 940
 941         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 942         if (lmd_bad_magic(lmd))
 943                 RETURN(-EINVAL);
 944
 945         sbi = lustre_init_sbi(sb);
 946         if (!sbi)
 947                 RETURN(-ENOMEM);
 948
 949         sbi->ll_flags |= LL_SBI_READAHEAD;
 950
 951         err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
 952         if (err) {
 953                 CERROR("Can not process the profile err %d \n", err);
 954                 GOTO(out_free, err);
 955         }
 956         if (!lov || !lmv) {
 957                 CERROR("no osc %p or no mdc %p \n", lov, lmv);
 958                 GOTO(out_free, err = -EINVAL);
 959         }
 960
 961         err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
 962                                        lmd->lmd_mds_security,
 963                                        lmd->lmd_oss_security,
 964                                        &lmd->lmd_nllu, lmd->lmd_pag,
 965                                        &lmd->lmd_remote_flag);
 966
 967         if (err)
 968                 GOTO(out_free, err);
 969
 970 out_dev:
 971         if (lmv)
 972                 OBD_FREE(lmv, strlen(lmv) + 1);
 973         if (lov)
 974                 OBD_FREE(lov, strlen(lov) + 1);
 975         if (gkc)
 976                 OBD_FREE(gkc, strlen(gkc) + 1);
 977
 978         RETURN(err);
 979 out_free:
 980         lustre_clean_profile(sbi, 0);
 981         lustre_free_sbi(sb);
 982         GOTO(out_dev, err);
 983
 984 } /* lustre_fill_super */
 985
 986 void lustre_put_super(struct super_block *sb)
 987 {
 988         struct obd_device *obd;
 989         struct ll_sb_info *sbi = ll_s2sbi(sb);
 990         int force_umount = 0;
 991         ENTRY;
 992
 993         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 994         obd = class_exp2obd(sbi->ll_md_exp);
 995         if (obd)
 996                 force_umount = obd->obd_no_recov;
 997         obd = NULL;
 998
 999         lustre_common_put_super(sb);
1000         lustre_clean_profile(sbi, force_umount);
1001         lustre_free_sbi(sb);
1002
1003         EXIT;
1004 } /* lustre_put_super */
1005
1006 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1007 {
1008         struct lustre_mount_data *lmd = sbi->ll_lmd;
1009         char *profile = lmd->lmd_profile, *name = NULL;
1010         struct config_llog_instance cfg;
1011         int rc, namelen =  0, version;
1012         struct llog_ctxt *ctxt;
1013         ENTRY;
1014
1015         if (profile == NULL)
1016                 RETURN(0);
1017         if (lmd == NULL) {
1018                 CERROR("Client not mounted with zero-conf; cannot "
1019                        "process update log.\n");
1020                 RETURN(0);
1021         }
1022
1023         rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
1024                                LDLM_FL_CONFIG_CHANGE, NULL);
1025         if (rc != 0)
1026                 CWARN("obd_cancel_unused(mdc): %d\n", rc);
1027
1028         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
1029                                LDLM_FL_CONFIG_CHANGE, NULL);
1030         if (rc != 0)
1031                 CWARN("obd_cancel_unused(lov): %d\n", rc);
1032
1033         cfg.cfg_instance = sbi->ll_instance;
1034         cfg.cfg_uuid = sbi->ll_sb_uuid;
1035         cfg.cfg_local_nid = lmd->lmd_local_nid;
1036
1037         namelen = strlen(profile) + 20; /* -clean-######### */
1038         OBD_ALLOC(name, namelen);
1039         if (name == NULL)
1040                 RETURN(-ENOMEM);
1041
1042         if (clean) {
1043                 version = sbi->ll_config_version - 1;
1044                 sprintf(name, "%s-clean-%d", profile, version);
1045         } else {
1046                 version = sbi->ll_config_version + 1;
1047                 sprintf(name, "%s-%d", profile, version);
1048         }
1049
1050         CWARN("Applying configuration log %s\n", name);
1051
1052         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1053                                 LLOG_CONFIG_REPL_CTXT);
1054         rc = class_config_process_llog(ctxt, name, &cfg);
1055         if (rc == 0)
1056                 sbi->ll_config_version = version;
1057         CWARN("Finished applying configuration log %s: %d\n", name, rc);
1058
1059         if (rc == 0 && clean == 0) {
1060                 struct lov_desc desc;
1061                 __u32 valsize;
1062                 int rc = 0;
1063
1064                 valsize = sizeof(desc);
1065                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1066                                   "lovdesc", &valsize, &desc);
1067
1068                 rc = obd_init_ea_size(sbi->ll_md_exp,
1069                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
1070                                       (desc.ld_tgt_count *
1071                                        sizeof(struct llog_cookie)));
1072         }
1073         OBD_FREE(name, namelen);
1074         RETURN(rc);
1075 }
1076
1077 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1078 {
1079         struct inode *inode = NULL;
1080
1081         /* NOTE: we depend on atomic igrab() -bzzz */
1082         lock_res_and_lock(lock);
1083         if (lock->l_ast_data) {
1084                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1085                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1086                         inode = igrab(lock->l_ast_data);
1087                 } else {
1088                         struct timeval now;
1089                         do_gettimeofday(&now);
1090                         inode = lock->l_ast_data;
1091                         LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1092                                    lock->l_enqueued_time.tv_sec,
1093                                    lock->l_enqueued_time.tv_usec,
1094                                    now.tv_sec, now.tv_usec);
1095                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1096                                "l_ast_data %p is bogus: magic %0x8\n",
1097                                lock->l_ast_data, lli->lli_inode_magic);
1098                         CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1099                                inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1100                         inode = NULL;
1101                         unlock_res_and_lock(lock);
1102                         LBUG();
1103                 }
1104         }
1105         unlock_res_and_lock(lock);
1106         return inode;
1107 }
1108
1109 int null_if_equal(struct ldlm_lock *lock, void *data)
1110 {
1111         if (data == lock->l_ast_data) {
1112                 lock->l_ast_data = NULL;
1113
1114                 if (lock->l_req_mode != lock->l_granted_mode)
1115                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1116         }
1117
1118         return LDLM_ITER_CONTINUE;
1119 }
1120
1121 static void remote_acl_free(struct remote_acl *racl);
1122
1123 void ll_clear_inode(struct inode *inode)
1124 {
1125         struct lustre_id id;
1126         struct ll_inode_info *lli = ll_i2info(inode);
1127         struct ll_sb_info *sbi = ll_i2sbi(inode);
1128         ENTRY;
1129
1130         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1131                inode->i_generation, inode);
1132
1133         ll_inode2id(&id, inode);
1134
1135         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1136         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1137
1138         LASSERT(!lli->lli_open_fd_write_count);
1139         LASSERT(!lli->lli_open_fd_read_count);
1140         LASSERT(!lli->lli_open_fd_exec_count);
1141         if (lli->lli_mds_write_och)
1142                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1143         if (lli->lli_mds_exec_och)
1144                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1145         if (lli->lli_mds_read_och)
1146                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1147         if (lli->lli_smd)
1148                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1149                                   null_if_equal, inode);
1150
1151         if (lli->lli_smd) {
1152                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1153                 lli->lli_smd = NULL;
1154         }
1155
1156         if (lli->lli_mea) {
1157                 obd_free_memmd(sbi->ll_md_exp,
1158                                (struct lov_stripe_md **) &lli->lli_mea);
1159                 lli->lli_mea = NULL;
1160         }
1161         ll_crypto_destroy_inode_key(inode);
1162         if (lli->lli_symlink_name) {
1163                 OBD_FREE(lli->lli_symlink_name,
1164                          strlen(lli->lli_symlink_name) + 1);
1165                 lli->lli_symlink_name = NULL;
1166         }
1167
1168         if (lli->lli_posix_acl) {
1169                 LASSERT(lli->lli_remote_acl == NULL);
1170                 posix_acl_release(lli->lli_posix_acl);
1171                 lli->lli_posix_acl = NULL;
1172         }
1173
1174         if (lli->lli_remote_acl) {
1175                 LASSERT(lli->lli_posix_acl == NULL);
1176                 remote_acl_free(lli->lli_remote_acl);
1177                 lli->lli_remote_acl = NULL;
1178         }
1179
1180         if (lli->lli_trunc_capa) {
1181                 OBD_FREE(lli->lli_trunc_capa, sizeof(struct lustre_capa));
1182                 lli->lli_trunc_capa = NULL;
1183         }
1184
1185         lli->lli_inode_magic = LLI_INODE_DEAD;
1186         EXIT;
1187 }
1188
1189 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1190  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1191  * keep these values until such a time that objects are allocated for it.
1192  * We do the MDS operations first, as it is checking permissions for us.
1193  * We don't to the MDS RPC if there is nothing that we want to store there,
1194  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1195  * going to do an RPC anyways.
1196  *
1197  * If we are doing a truncate, we will send the mtime and ctime updates
1198  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1199  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1200  * at the same time.
1201  */
1202 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1203 {
1204         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1205         struct ll_inode_info *lli = ll_i2info(inode);
1206         struct ll_sb_info *sbi = ll_i2sbi(inode);
1207         struct ptlrpc_request *request = NULL;
1208         struct mdc_op_data *op_data;
1209         struct lustre_capa *trunc_capa = NULL;
1210         int ia_valid = attr->ia_valid;
1211         int err, rc = 0;
1212         ENTRY;
1213
1214         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1215         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1216
1217         if (ia_valid & ATTR_SIZE) {
1218                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1219                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1220                                attr->ia_size, ll_file_maxbytes(inode));
1221                         RETURN(-EFBIG);
1222                 }
1223
1224                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1225         }
1226
1227         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1228         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1229                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1230                         RETURN(-EPERM);
1231         }
1232
1233         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1234         if (attr->ia_valid & ATTR_CTIME) {
1235                 attr->ia_ctime = CURRENT_TIME;
1236                 attr->ia_valid |= ATTR_CTIME_SET;
1237         }
1238         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1239                 attr->ia_atime = CURRENT_TIME;
1240                 attr->ia_valid |= ATTR_ATIME_SET;
1241         }
1242         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1243                 attr->ia_mtime = CURRENT_TIME;
1244                 attr->ia_valid |= ATTR_MTIME_SET;
1245         }
1246
1247         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1248                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1249                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1250                        LTIME_S(CURRENT_TIME));
1251
1252         /* If only OST attributes being set on objects, don't do MDS RPC.
1253          * In that case, we need to check permissions and update the local
1254          * inode ourselves so we can call obdo_from_inode() always. */
1255         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1256                 struct lustre_md md;
1257                 void *key = NULL;
1258                 int  key_size = 0;
1259
1260                 OBD_ALLOC(op_data, sizeof(*op_data));
1261                 if (op_data == NULL)
1262                         RETURN(-ENOMEM);
1263                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1264
1265                 if (ia_valid & (ATTR_UID | ATTR_GID)) {
1266                         rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1267                                                &key_size);
1268                 }
1269                 rc = md_setattr(sbi->ll_md_exp, op_data,
1270                                 attr, key, key_size, NULL, 0, NULL,
1271                                 0, &request);
1272                 OBD_FREE(op_data, sizeof(*op_data));
1273
1274                 if (key && key_size)
1275                         OBD_FREE(key, key_size);
1276                 if (rc) {
1277                         ptlrpc_req_finished(request);
1278                         if (rc != -EPERM && rc != -EACCES)
1279                                 CERROR("md_setattr fails: rc = %d\n", rc);
1280                         RETURN(rc);
1281                 }
1282                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1283                                        sbi->ll_dt_exp, &md);
1284                 if (rc) {
1285                         ptlrpc_req_finished(request);
1286                         RETURN(rc);
1287                 }
1288
1289                 if (attr->ia_valid & ATTR_SIZE) {
1290                         /* XXX: hack for truncate capa */
1291                         rc = mdc_req2lustre_capa(request, 0, &trunc_capa);
1292                         if (rc) {
1293                                 ptlrpc_req_finished(request);
1294                                 RETURN(rc);
1295                         }
1296
1297                         spin_lock(&lli->lli_lock);
1298                         if (trunc_capa) {
1299                                 if (lli->lli_trunc_capa)
1300                                         OBD_FREE(lli->lli_trunc_capa,
1301                                                  sizeof(*trunc_capa));
1302                                 lli->lli_trunc_capa = trunc_capa;
1303                         }
1304                         spin_unlock(&lli->lli_lock);
1305                 }
1306
1307                 /* We call inode_setattr to adjust timestamps, but we first
1308                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1309                  *
1310                  * NB: ATTR_SIZE will only be set at this point if the size
1311                  * resides on the MDS, ie, this file has no objects. */
1312                 attr->ia_valid &= ~ATTR_SIZE;
1313
1314                 /*
1315                  * assigning inode_setattr() to @err to disable warning that
1316                  * function's result should be checked by by caller. error is
1317                  * impossible here, as vmtruncate() control path is disabled.
1318                  */
1319                 err = inode_setattr(inode, attr);
1320                 ll_update_inode(inode, &md);
1321                 ptlrpc_req_finished(request);
1322
1323                 if (!lsm || !S_ISREG(inode->i_mode)) {
1324                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1325                         RETURN(0);
1326                 }
1327         } else {
1328                 /* The OST doesn't check permissions, but the alternative is
1329                  * a gratuitous RPC to the MDS.  We already rely on the client
1330                  * to do read/write/truncate permission checks, so is mtime OK?
1331                  */
1332                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1333                         /* from sys_utime() */
1334                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1335                                 if (current->fsuid != inode->i_uid &&
1336                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1337                                         RETURN(rc);
1338                         } else {
1339                                 /* from inode_change_ok() */
1340                                 if (current->fsuid != inode->i_uid &&
1341                                     !capable(CAP_FOWNER))
1342                                         RETURN(-EPERM);
1343                         }
1344                 }
1345
1346                 if (lsm)
1347                         attr->ia_valid &= ~ATTR_SIZE;
1348
1349                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1350                 err = inode_setattr(inode, attr);
1351                 /*
1352                  * assigning inode_setattr() to @err to disable warning that
1353                  * function's result should be checked by by caller. error is
1354                  * impossible here, as vmtruncate() control path is disabled.
1355                  */
1356         }
1357
1358         /* We really need to get our PW lock before we change inode->i_size.
1359          * If we don't we can race with other i_size updaters on our node, like
1360          * ll_file_read.  We can also race with i_size propogation to other
1361          * nodes through dirtying and writeback of final cached pages.  This
1362          * last one is especially bad for racing o_append users on other
1363          * nodes. */
1364         if (ia_valid & ATTR_SIZE) {
1365                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1366                                                            OBD_OBJECT_EOF } };
1367                 struct lustre_handle lockh = { 0 };
1368                 int err, ast_flags = 0;
1369                 /* XXX when we fix the AST intents to pass the discard-range
1370                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1371                  * XXX here. */
1372                 if (attr->ia_size == 0)
1373                         ast_flags = LDLM_AST_DISCARD_DATA;
1374
1375                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1376                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1377
1378                 if (rc != 0)
1379                         RETURN(rc);
1380
1381                 down(&lli->lli_size_sem);
1382                 lli->lli_size_pid = current->pid;
1383                 rc = vmtruncate(inode, attr->ia_size);
1384                 if (rc != 0) {
1385                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1386                         lli->lli_size_pid = 0;
1387                         up(&lli->lli_size_sem);
1388                 }
1389
1390                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1391                 if (err) {
1392                         CERROR("ll_extent_unlock failed: %d\n", err);
1393                         if (!rc)
1394                                 rc = err;
1395                 }
1396         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1397                 struct obdo *oa = NULL;
1398
1399                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1400                        inode->i_ino, LTIME_S(attr->ia_mtime));
1401
1402                 oa = obdo_alloc();
1403                 if (oa == NULL)
1404                         RETURN(-ENOMEM);
1405
1406                 oa->o_id = lsm->lsm_object_id;
1407                 oa->o_gr = lsm->lsm_object_gr;
1408                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1409
1410                 /* adding uid and gid, needed for quota */
1411                 if (ia_valid & ATTR_UID) {
1412                         oa->o_uid = inode->i_uid;
1413                         oa->o_valid |= OBD_MD_FLUID;
1414                 }
1415
1416                 if (ia_valid & ATTR_GID) {
1417                         oa->o_gid = inode->i_gid;
1418                         oa->o_valid |= OBD_MD_FLGID;
1419                 }
1420
1421                 *(obdo_id(oa)) = lli->lli_id;
1422                 oa->o_valid |= OBD_MD_FLIFID;
1423
1424                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1425                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1426                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1427                 obdo_free(oa);
1428                 if (rc)
1429                         CERROR("obd_setattr fails: rc = %d\n", rc);
1430         }
1431
1432         RETURN(rc);
1433 }
1434
1435 int ll_setattr(struct dentry *de, struct iattr *attr)
1436 {
1437         LASSERT(de->d_inode);
1438         return ll_setattr_raw(de->d_inode, attr);
1439 }
1440
1441 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1442                        unsigned long max_age)
1443 {
1444         struct ll_sb_info *sbi = ll_s2sbi(sb);
1445         struct obd_statfs obd_osfs;
1446         int rc;
1447         ENTRY;
1448
1449         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1450         if (rc) {
1451                 CERROR("obd_statfs fails: rc = %d\n", rc);
1452                 RETURN(rc);
1453         }
1454
1455         osfs->os_type = sb->s_magic;
1456
1457         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1458                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1459
1460         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1461         if (rc) {
1462                 CERROR("obd_statfs fails: rc = %d\n", rc);
1463                 RETURN(rc);
1464         }
1465
1466         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1467                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1468                obd_osfs.os_files);
1469
1470         osfs->os_blocks = obd_osfs.os_blocks;
1471         osfs->os_bfree = obd_osfs.os_bfree;
1472         osfs->os_bavail = obd_osfs.os_bavail;
1473
1474         /* If we don't have as many objects free on the OST as inodes
1475          * on the MDS, we reduce the total number of inodes to
1476          * compensate, so that the "inodes in use" number is correct.
1477          */
1478         if (obd_osfs.os_ffree < osfs->os_ffree) {
1479                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1480                         obd_osfs.os_ffree;
1481                 osfs->os_ffree = obd_osfs.os_ffree;
1482         }
1483
1484         RETURN(rc);
1485 }
1486
1487 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1488 {
1489         struct obd_statfs osfs;
1490         int rc;
1491
1492         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1493         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1494
1495         /* For now we will always get up-to-date statfs values, but in the
1496          * future we may allow some amount of caching on the client (e.g.
1497          * from QOS or lprocfs updates). */
1498         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1499         if (rc)
1500                 return rc;
1501
1502         statfs_unpack(sfs, &osfs);
1503
1504         if (sizeof(sfs->f_blocks) == 4) {
1505                 while (osfs.os_blocks > ~0UL) {
1506                         sfs->f_bsize <<= 1;
1507
1508                         osfs.os_blocks >>= 1;
1509                         osfs.os_bfree >>= 1;
1510                         osfs.os_bavail >>= 1;
1511                 }
1512         }
1513
1514         sfs->f_blocks = osfs.os_blocks;
1515         sfs->f_bfree = osfs.os_bfree;
1516         sfs->f_bavail = osfs.os_bavail;
1517
1518         return 0;
1519 }
1520
1521
1522 /********************************
1523  * remote acl                   *
1524  ********************************/
1525
1526 static struct remote_acl *remote_acl_alloc(void)
1527 {
1528         struct remote_acl *racl;
1529         int i;
1530
1531         OBD_ALLOC(racl, sizeof(*racl));
1532         if (!racl)
1533                 return NULL;
1534
1535         spin_lock_init(&racl->ra_lock);
1536         init_MUTEX(&racl->ra_update_sem);
1537
1538         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1539                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1540
1541         return racl;
1542 }
1543
1544 /*
1545  * caller should guarantee no race here.
1546  */
1547 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1548 {
1549         struct remote_perm_setxid *xperm;
1550
1551         while (!list_empty(&perm->lrp_setxid_perms)) {
1552                 xperm = list_entry(perm->lrp_setxid_perms.next,
1553                                    struct remote_perm_setxid,
1554                                    list);
1555                 list_del(&xperm->list);
1556                 OBD_FREE(xperm, sizeof(*xperm));
1557         }
1558 }
1559
1560 /*
1561  * caller should guarantee no race here.
1562  */
1563 static void remote_acl_flush(struct remote_acl *racl)
1564 {
1565         struct list_head *head;
1566         struct lustre_remote_perm *perm, *tmp;
1567         int i;
1568
1569         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1570                 head = &racl->ra_perm_cache[i];
1571
1572                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1573                         remote_perm_flush_xperms(perm);
1574                         list_del(&perm->lrp_list);
1575                         OBD_FREE(perm, sizeof(*perm));
1576                 }
1577         }
1578 }
1579
1580 static void remote_acl_free(struct remote_acl *racl)
1581 {
1582         if (!racl)
1583                 return;
1584
1585         down(&racl->ra_update_sem);
1586         spin_lock(&racl->ra_lock);
1587         remote_acl_flush(racl);
1588         spin_unlock(&racl->ra_lock);
1589         up(&racl->ra_update_sem);
1590
1591         OBD_FREE(racl, sizeof(*racl));
1592 }
1593
1594 static inline int remote_acl_hashfunc(__u32 id)
1595 {
1596         return (id & (REMOTE_ACL_HASHSIZE - 1));
1597 }
1598
1599 static
1600 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1601 {
1602         struct list_head *head;
1603         struct lustre_remote_perm *lperm;
1604         struct remote_perm_setxid *xperm;
1605         int found = 0, rc = -ENOENT;
1606
1607         LASSERT(racl);
1608         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1609         spin_lock(&racl->ra_lock);
1610
1611         list_for_each_entry(lperm, head, lrp_list) {
1612                 if (lperm->lrp_auth_uid == current->uid) {
1613                         found = 1;
1614                         break;
1615                 }
1616         }
1617
1618         if (!found)
1619                 goto out;
1620
1621         if (lperm->lrp_auth_uid == current->fsuid &&
1622             lperm->lrp_auth_gid == current->fsgid) {
1623                 if (lperm->lrp_valid) {
1624                         *perm = lperm->lrp_perm;
1625                         rc = 0;
1626                 }
1627                 goto out;
1628         } else if ((!lperm->lrp_setuid &&
1629                     lperm->lrp_auth_uid != current->fsuid) ||
1630                    (!lperm->lrp_setgid &&
1631                     lperm->lrp_auth_gid != current->fsgid))  {
1632                 *perm = 0;
1633                 rc = 0;
1634                 goto out;
1635         }
1636
1637         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1638                 if (xperm->uid == current->fsuid &&
1639                     xperm->gid == current->fsgid) {
1640                         *perm = xperm->perm;
1641                         rc = 0;
1642                         goto out;
1643                 }
1644         }
1645
1646 out:
1647         spin_unlock(&racl->ra_lock);
1648         return rc;
1649 }
1650
1651 static
1652 int __remote_acl_update(struct remote_acl *racl,
1653                         struct mds_remote_perm *mperm,
1654                         struct lustre_remote_perm *lperm,
1655                         struct remote_perm_setxid *xperm)
1656 {
1657         struct list_head *head;
1658         struct lustre_remote_perm *lp;
1659         struct remote_perm_setxid *xp;
1660         int found = 0, setuid = 0, setgid = 0;
1661
1662         LASSERT(racl);
1663         LASSERT(mperm);
1664         LASSERT(lperm);
1665         LASSERT(current->uid == mperm->mrp_auth_uid);
1666
1667         if (current->fsuid != mperm->mrp_auth_uid)
1668                 setuid = 1;
1669         if (current->fsgid != mperm->mrp_auth_gid)
1670                 setgid = 1;
1671
1672         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1673         spin_lock(&racl->ra_lock);
1674
1675         list_for_each_entry(lp, head, lrp_list) {
1676                 if (lp->lrp_auth_uid == current->uid) {
1677                         found = 1;
1678                         break;
1679                 }
1680         }
1681
1682         if (found) {
1683                 OBD_FREE(lperm, sizeof(*lperm));
1684
1685                 if (!lp->lrp_valid && !setuid && !setgid) {
1686                         lp->lrp_perm = mperm->mrp_perm;
1687                         lp->lrp_valid = 1;
1688                 }
1689
1690                 /* sanity check for changes of setxid rules */
1691                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1692                         CWARN("setuid changes: %d => %d\n",
1693                               (lp->lrp_setuid != 0),
1694                               (mperm->mrp_allow_setuid != 0));
1695                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1696                 }
1697
1698                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1699                         CWARN("setgid changes: %d => %d\n",
1700                               (lp->lrp_setgid != 0),
1701                               (mperm->mrp_allow_setgid != 0));
1702                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1703                 }
1704
1705                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1706                     !list_empty(&lp->lrp_setxid_perms)) {
1707                         remote_perm_flush_xperms(lp);
1708                 }
1709         } else {
1710                 /* initialize lperm and linked into hashtable
1711                  */
1712                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1713                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1714                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1715                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1716                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1717                 list_add(&lperm->lrp_list, head);
1718
1719                 if (!setuid && !setgid) {
1720                         /* in this case, i'm the authenticated user,
1721                          * and mrp_perm is for me.
1722                          */
1723                         lperm->lrp_perm = mperm->mrp_perm;
1724                         lperm->lrp_valid = 1;
1725                         spin_unlock(&racl->ra_lock);
1726
1727                         if (xperm)
1728                                 OBD_FREE(xperm, sizeof(*xperm));
1729                         return 0;
1730                 }
1731
1732                 lp = lperm;
1733                 /* fall through */
1734         }
1735
1736         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1737                 list_empty(&lp->lrp_setxid_perms));
1738
1739         /* if no xperm supplied, we are all done here */
1740         if (!xperm) {
1741                 spin_unlock(&racl->ra_lock);
1742                 return 0;
1743         }
1744
1745         /* whether we allow setuid/setgid */
1746         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1747                 OBD_FREE(xperm, sizeof(*xperm));
1748                 spin_unlock(&racl->ra_lock);
1749                 return 0;
1750         }
1751
1752         /* traverse xperm list */
1753         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1754                 if (xp->uid == current->fsuid &&
1755                     xp->gid == current->fsgid) {
1756                         if (xp->perm != mperm->mrp_perm) {
1757                                 /* actually this should not happen */
1758                                 CWARN("perm changed: %o => %o\n",
1759                                       xp->perm, mperm->mrp_perm);
1760                                 xp->perm = mperm->mrp_perm;
1761                         }
1762                         OBD_FREE(xperm, sizeof(*xperm));
1763                         spin_unlock(&racl->ra_lock);
1764                         return 0;
1765                 }
1766         }
1767
1768         /* finally insert this xperm */
1769         xperm->uid = current->fsuid;
1770         xperm->gid = current->fsgid;
1771         xperm->perm = mperm->mrp_perm;
1772         list_add(&xperm->list, &lp->lrp_setxid_perms);
1773
1774         spin_unlock(&racl->ra_lock);
1775         return 0;
1776 }
1777
1778 /*
1779  * remote_acl semaphore must be held by caller
1780  */
1781 static
1782 int remote_acl_update_locked(struct remote_acl *racl,
1783                              struct mds_remote_perm *mperm)
1784 {
1785         struct lustre_remote_perm *lperm;
1786         struct remote_perm_setxid *xperm;
1787         int setuid = 0, setgid = 0;
1788
1789         might_sleep();
1790
1791         if (current->uid != mperm->mrp_auth_uid) {
1792                 CERROR("current uid %u while authenticated as %u\n",
1793                        current->uid, mperm->mrp_auth_uid);
1794                 return -EINVAL;
1795         }
1796
1797         if (current->fsuid != mperm->mrp_auth_uid)
1798                 setuid = 1;
1799         if (current->fsgid == mperm->mrp_auth_gid)
1800                 setgid = 1;
1801
1802         OBD_ALLOC(lperm, sizeof(*lperm));
1803         if (!lperm)
1804                 return -ENOMEM;
1805
1806         if ((setuid || setgid) &&
1807             !(setuid && !mperm->mrp_allow_setuid) &&
1808             !(setgid && !mperm->mrp_allow_setgid)) {
1809                 OBD_ALLOC(xperm, sizeof(*xperm));
1810                 if (!xperm) {
1811                         OBD_FREE(lperm, sizeof(*lperm));
1812                         return -ENOMEM;
1813                 }
1814         } else
1815                 xperm = NULL;
1816
1817         return __remote_acl_update(racl, mperm, lperm, xperm);
1818 }
1819
1820 /*
1821  * return -EACCES at any error cases
1822  */
1823 int ll_remote_acl_permission(struct inode *inode, int mode)
1824 {
1825         struct ll_sb_info *sbi = ll_i2sbi(inode);
1826         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1827         struct ptlrpc_request *req = NULL;
1828         struct lustre_id id;
1829         struct mds_remote_perm *mperm;
1830         int rc = -EACCES, perm;
1831
1832         if (!racl)
1833                 return -EACCES;
1834
1835         if (__remote_acl_check(racl, &perm) == 0) {
1836                 return ((perm & mode) == mode ? 0 : -EACCES);
1837         }
1838
1839         might_sleep();
1840
1841         /* doing update
1842          */
1843         down(&racl->ra_update_sem);
1844
1845         /* we might lose the race when obtain semaphore,
1846          * so check again.
1847          */
1848         if (__remote_acl_check(racl, &perm) == 0) {
1849                 if ((perm & mode) == mode)
1850                         rc = 0;
1851                 goto out;
1852         }
1853
1854         /* really fetch from mds
1855          */
1856         ll_inode2id(&id, inode);
1857         if (md_access_check(sbi->ll_md_exp, &id, &req))
1858                 goto out;
1859
1860         /* status non-zero indicate there's more apparent error
1861          * detected by mds, e.g. didn't allow this user at all.
1862          * we simply ignore and didn't cache it.
1863          */
1864         if (req->rq_repmsg->status)
1865                 goto out;
1866
1867         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1868                                    lustre_swab_remote_perm);
1869         LASSERT(mperm);
1870         LASSERT_REPSWABBED(req, 1);
1871
1872         if ((mperm->mrp_perm & mode) == mode)
1873                 rc = 0;
1874
1875         remote_acl_update_locked(racl, mperm);
1876 out:
1877         if (req)
1878                 ptlrpc_req_finished(req);
1879
1880         up(&racl->ra_update_sem);
1881         return rc;
1882 }
1883
1884 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1885 {
1886         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1887         int rc;
1888
1889         LASSERT(perm);
1890
1891         if (!racl)
1892                 return -EACCES;
1893
1894         down(&racl->ra_update_sem);
1895         rc = remote_acl_update_locked(racl, perm);
1896         up(&racl->ra_update_sem);
1897
1898         return rc;
1899 }
1900
1901 void ll_inode_invalidate_acl(struct inode *inode)
1902 {
1903         struct ll_sb_info *sbi = ll_i2sbi(inode);
1904         struct ll_inode_info *lli = ll_i2info(inode);
1905
1906         if (sbi->ll_remote) {
1907                 struct remote_acl *racl = lli->lli_remote_acl;
1908
1909                 LASSERT(!lli->lli_posix_acl);
1910                 if (racl) {
1911                         down(&racl->ra_update_sem);
1912                         spin_lock(&racl->ra_lock);
1913                         remote_acl_flush(lli->lli_remote_acl);
1914                         spin_unlock(&racl->ra_lock);
1915                         up(&racl->ra_update_sem);
1916                 }
1917         } else {
1918                 /* we can't invalide acl here: suppose we touch a new file
1919                  * under a dir, blocking ast on dir will lead to open failure
1920                  * on client, although succeed on mds. it's kind of weird,
1921                  * the real fix i think is improve client-vfs interaction.
1922                  *
1923                  * currently we just do nothing here.
1924                  */
1925                 return;
1926
1927                 LASSERT(!lli->lli_remote_acl);
1928                 spin_lock(&lli->lli_lock);
1929                 posix_acl_release(lli->lli_posix_acl);
1930                 lli->lli_posix_acl = NULL;
1931                 spin_unlock(&lli->lli_lock);
1932         }
1933 }
1934
1935 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1936 {
1937         struct ll_inode_info *lli = ll_i2info(inode);
1938         struct lov_stripe_md *lsm = md->lsm;
1939         struct mds_body *body = md->body;
1940         struct mea *mea = md->mea;
1941         struct posix_acl *posix_acl = md->posix_acl;
1942         struct ll_sb_info *sbi = ll_i2sbi(inode);
1943         struct lustre_key *mkey = md->key;
1944         ENTRY;
1945
1946         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1947
1948         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1949                 /* check for default striping info for dir. */
1950                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1951         }
1952
1953         if (lsm != NULL) {
1954                 LASSERT(lsm->lsm_object_gr > 0);
1955                 if (lli->lli_smd == NULL) {
1956                         lli->lli_smd = lsm;
1957                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1958                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1959                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1960                 } else {
1961                         int i;
1962                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1963                                 CERROR("lsm mismatch for inode %ld\n",
1964                                        inode->i_ino);
1965                                 CERROR("lli_smd:\n");
1966                                 dump_lsm(D_ERROR, lli->lli_smd);
1967                                 CERROR("lsm:\n");
1968                                 dump_lsm(D_ERROR, lsm);
1969                                 LBUG();
1970                         }
1971                         /* XXX FIXME -- We should decide on a safer (atomic) and
1972                          * more elegant way to update the lsm */
1973                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1974                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1975                                         lsm->lsm_oinfo[i].loi_id;
1976                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1977                                         lsm->lsm_oinfo[i].loi_gr;
1978                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1979                                         lsm->lsm_oinfo[i].loi_ost_idx;
1980                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1981                                         lsm->lsm_oinfo[i].loi_ost_gen;
1982                         }
1983                 }
1984                 /* bug 2844 - limit i_blksize for broken user-space apps */
1985                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1986                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1987                 if (lli->lli_smd != lsm)
1988                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1989         }
1990
1991         if (mea != NULL) {
1992                 if (lli->lli_mea == NULL) {
1993                         lli->lli_mea = mea;
1994                 } else {
1995                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1996                                 CERROR("mea mismatch for inode %lu\n",
1997                                         inode->i_ino);
1998                                 LBUG();
1999                         }
2000                 }
2001                 if (lli->lli_mea != mea)
2002                         obd_free_memmd(ll_i2mdexp(inode),
2003                                        (struct lov_stripe_md **) &mea);
2004         }
2005
2006         if (body->valid & OBD_MD_FID)
2007                 id_assign_fid(&lli->lli_id, &body->id1);
2008
2009         if (body->valid & OBD_MD_FLID)
2010                 id_ino(&lli->lli_id) = id_ino(&body->id1);
2011
2012         if (body->valid & OBD_MD_FLGENER)
2013                 id_gen(&lli->lli_id) = id_gen(&body->id1);
2014
2015         /* local/remote ACL */
2016         if (sbi->ll_remote) {
2017                 LASSERT(md->posix_acl == NULL);
2018                 if (md->remote_perm) {
2019                         ll_remote_acl_update(inode, md->remote_perm);
2020                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2021                         md->remote_perm = NULL;
2022                 }
2023         } else {
2024                 LASSERT(md->remote_perm == NULL);
2025                 spin_lock(&lli->lli_lock);
2026                 if (posix_acl != NULL) {
2027                         if (lli->lli_posix_acl != NULL)
2028                                 posix_acl_release(lli->lli_posix_acl);
2029                         lli->lli_posix_acl = posix_acl;
2030                 }
2031                 spin_unlock(&lli->lli_lock);
2032         }
2033
2034         if (body->valid & OBD_MD_FLID)
2035                 inode->i_ino = id_ino(&body->id1);
2036         if (body->valid & OBD_MD_FLGENER)
2037                 inode->i_generation = id_gen(&body->id1);
2038         if (body->valid & OBD_MD_FLATIME)
2039                 LTIME_S(inode->i_atime) = body->atime;
2040         if (body->valid & OBD_MD_FLMTIME &&
2041             body->mtime > LTIME_S(inode->i_mtime)) {
2042                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2043                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2044                 LTIME_S(inode->i_mtime) = body->mtime;
2045         }
2046         if (body->valid & OBD_MD_FLCTIME &&
2047             body->ctime > LTIME_S(inode->i_ctime))
2048                 LTIME_S(inode->i_ctime) = body->ctime;
2049         if (body->valid & OBD_MD_FLMODE) {
2050                 inode->i_mode = (inode->i_mode & S_IFMT) |
2051                         (body->mode & ~S_IFMT);
2052         }
2053         if (body->valid & OBD_MD_FLTYPE) {
2054                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2055                         (body->mode & S_IFMT);
2056         }
2057         if (body->valid & OBD_MD_FLUID)
2058                 inode->i_uid = body->uid;
2059         if (body->valid & OBD_MD_FLGID)
2060                 inode->i_gid = body->gid;
2061         if (body->valid & OBD_MD_FLFLAGS)
2062                 inode->i_flags = body->flags;
2063         if (body->valid & OBD_MD_FLNLINK)
2064                 inode->i_nlink = body->nlink;
2065         if (body->valid & OBD_MD_FLRDEV)
2066 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2067                 inode->i_rdev = body->rdev;
2068 #else
2069                 inode->i_rdev = old_decode_dev(body->rdev);
2070 #endif
2071         if (body->valid & OBD_MD_FLSIZE)
2072                 inode->i_size = body->size;
2073         if (body->valid & OBD_MD_FLBLOCKS)
2074                 inode->i_blocks = body->blocks;
2075
2076         if (body->valid & OBD_MD_FLSIZE)
2077                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2078
2079         if (body->valid & OBD_MD_FLAUDIT) {
2080                 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2081                 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2082                         sbi->ll_audit_mask = body->audit;
2083                 else
2084                         lli->lli_audit_mask = body->audit;
2085         }
2086
2087         if (mkey != NULL) {
2088                 LASSERT(body->valid & OBD_MD_FLKEY);
2089                 ll_crypto_init_inode_key(inode, mkey);
2090         }
2091
2092 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2093         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2094 #endif
2095         LASSERT(id_fid(&lli->lli_id) != 0);
2096 }
2097
2098 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2099 static struct backing_dev_info ll_backing_dev_info = {
2100         .ra_pages       = 0,    /* No readahead */
2101         .memory_backed  = 0,    /* Does contribute to dirty memory */
2102 };
2103 #endif
2104
2105 void ll_read_inode2(struct inode *inode, void *opaque)
2106 {
2107         struct lustre_md *md = opaque;
2108         struct ll_inode_info *lli = ll_i2info(inode);
2109         ENTRY;
2110
2111         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2112                inode->i_generation, inode);
2113
2114         ll_lli_init(lli);
2115
2116         LASSERT(!lli->lli_smd);
2117
2118         if (ll_i2sbi(inode)->ll_remote) {
2119                 lli->lli_remote_acl = remote_acl_alloc();
2120                 /* if failed alloc, nobody will be able to access this inode */
2121         }
2122
2123         /* Core attributes from the MDS first.  This is a new inode, and
2124          * the VFS doesn't zero times in the core inode so we have to do
2125          * it ourselves.  They will be overwritten by either MDS or OST
2126          * attributes - we just need to make sure they aren't newer. */
2127         LTIME_S(inode->i_mtime) = 0;
2128         LTIME_S(inode->i_atime) = 0;
2129         LTIME_S(inode->i_ctime) = 0;
2130
2131         inode->i_rdev = 0;
2132         ll_update_inode(inode, md);
2133
2134         /* OIDEBUG(inode); */
2135
2136         if (S_ISREG(inode->i_mode)) {
2137                 inode->i_op = &ll_file_inode_operations;
2138                 inode->i_fop = &ll_file_operations;
2139                 inode->i_mapping->a_ops = &ll_aops;
2140                 EXIT;
2141         } else if (S_ISDIR(inode->i_mode)) {
2142                 inode->i_op = &ll_dir_inode_operations;
2143                 inode->i_fop = &ll_dir_operations;
2144                 inode->i_mapping->a_ops = &ll_dir_aops;
2145                 EXIT;
2146         } else if (S_ISLNK(inode->i_mode)) {
2147                 inode->i_op = &ll_fast_symlink_inode_operations;
2148                 EXIT;
2149         } else {
2150                 inode->i_op = &ll_special_inode_operations;
2151
2152 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2153                 init_special_inode(inode, inode->i_mode,
2154                                    kdev_t_to_nr(inode->i_rdev));
2155
2156                 /* initializing backing dev info. */
2157                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2158 #else
2159                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2160 #endif
2161                 lli->ll_save_ifop = inode->i_fop;
2162
2163                 if (S_ISCHR(inode->i_mode))
2164                         inode->i_fop = &ll_special_chr_inode_fops;
2165                 else if (S_ISBLK(inode->i_mode))
2166                         inode->i_fop = &ll_special_blk_inode_fops;
2167                 else if (S_ISFIFO(inode->i_mode))
2168                         inode->i_fop = &ll_special_fifo_inode_fops;
2169                 else if (S_ISSOCK(inode->i_mode))
2170                         inode->i_fop = &ll_special_sock_inode_fops;
2171
2172                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2173                       inode->i_fop);
2174
2175                 if (lli->ll_save_ifop->owner) {
2176                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2177                               lli->ll_save_ifop->owner);
2178                 }
2179                 EXIT;
2180         }
2181 }
2182
2183 void ll_delete_inode(struct inode *inode)
2184 {
2185         struct ll_sb_info *sbi = ll_i2sbi(inode);
2186         struct lustre_id id;
2187         int rc;
2188         ENTRY;
2189
2190         ll_inode2id(&id, inode);
2191
2192         rc = md_delete_inode(sbi->ll_md_exp, &id);
2193         if (rc) {
2194                 CERROR("md_delete_inode() failed, error %d\n",
2195                        rc);
2196         }
2197
2198         clear_inode(inode);
2199         EXIT;
2200 }
2201
2202 int ll_iocontrol(struct inode *inode, struct file *file,
2203                  unsigned int cmd, unsigned long arg)
2204 {
2205         struct ll_sb_info *sbi = ll_i2sbi(inode);
2206         struct ptlrpc_request *req = NULL;
2207         int rc, flags = 0;
2208         ENTRY;
2209
2210         switch(cmd) {
2211         case EXT3_IOC_GETFLAGS: {
2212                 struct lustre_id id;
2213                 __u64 valid = OBD_MD_FLFLAGS;
2214                 struct mds_body *body;
2215
2216                 ll_inode2id(&id, inode);
2217                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2218                                 0, 0, NULL, &req);
2219                 if (rc) {
2220                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2221                         RETURN(-abs(rc));
2222                 }
2223
2224                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2225
2226                 if (body->flags & S_APPEND)
2227                         flags |= EXT3_APPEND_FL;
2228                 if (body->flags & S_IMMUTABLE)
2229                         flags |= EXT3_IMMUTABLE_FL;
2230                 if (body->flags & S_NOATIME)
2231                         flags |= EXT3_NOATIME_FL;
2232
2233                 ptlrpc_req_finished (req);
2234
2235                 RETURN(put_user(flags, (int *)arg));
2236         }
2237         case EXT3_IOC_SETFLAGS: {
2238                 struct mdc_op_data *op_data;
2239                 struct iattr attr;
2240                 struct obdo *oa;
2241                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2242
2243                 if (get_user(flags, (int *)arg))
2244                         RETURN(-EFAULT);
2245
2246                 oa = obdo_alloc();
2247                 if (!oa)
2248                         RETURN(-ENOMEM);
2249
2250                 OBD_ALLOC(op_data, sizeof(*op_data));
2251                 if (op_data == NULL) {
2252                         obdo_free(oa);
2253                         RETURN(-ENOMEM);
2254                 }
2255                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
2256
2257                 memset(&attr, 0x0, sizeof(attr));
2258                 attr.ia_attr_flags = flags;
2259                 attr.ia_valid |= ATTR_ATTR_FLAG;
2260
2261                 rc = md_setattr(sbi->ll_md_exp, op_data,
2262                                 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2263                 OBD_FREE(op_data, sizeof(*op_data));
2264                 if (rc) {
2265                         ptlrpc_req_finished(req);
2266                         if (rc != -EPERM && rc != -EACCES)
2267                                 CERROR("md_setattr fails: rc = %d\n", rc);
2268                         obdo_free(oa);
2269                         RETURN(rc);
2270                 }
2271                 ptlrpc_req_finished(req);
2272
2273                 oa->o_id = lsm->lsm_object_id;
2274                 oa->o_gr = lsm->lsm_object_gr;
2275                 oa->o_flags = flags;
2276                 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2277                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2278                               | OBD_MD_FLIFID;
2279
2280                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
2281                 obdo_free(oa);
2282                 if (rc) {
2283                         if (rc != -EPERM && rc != -EACCES)
2284                                 CERROR("md_setattr fails: rc = %d\n", rc);
2285                         RETURN(rc);
2286                 }
2287
2288                 if (flags & EXT3_APPEND_FL)
2289                         inode->i_flags |= S_APPEND;
2290                 else
2291                         inode->i_flags &= ~S_APPEND;
2292                 if (flags & EXT3_IMMUTABLE_FL)
2293                         inode->i_flags |= S_IMMUTABLE;
2294                 else
2295                         inode->i_flags &= ~S_IMMUTABLE;
2296                 if (flags & EXT3_NOATIME_FL)
2297                         inode->i_flags |= S_NOATIME;
2298                 else
2299                         inode->i_flags &= ~S_NOATIME;
2300
2301                 RETURN(0);
2302         }
2303         default:
2304                 RETURN(-ENOSYS);
2305         }
2306
2307         RETURN(0);
2308 }
2309
2310 /* this is only called in the case of forced umount. */
2311 void ll_umount_begin(struct super_block *sb)
2312 {
2313         struct ll_sb_info *sbi = ll_s2sbi(sb);
2314         struct obd_ioctl_data ioc_data = { 0 };
2315         struct obd_device *obd;
2316         ENTRY;
2317
2318         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2319                sb->s_count, atomic_read(&sb->s_active));
2320
2321         obd = class_exp2obd(sbi->ll_md_exp);
2322         if (obd == NULL) {
2323                 CERROR("Invalid MDC connection handle "LPX64"\n",
2324                        sbi->ll_md_exp->exp_handle.h_cookie);
2325                 EXIT;
2326                 return;
2327         }
2328         obd->obd_no_recov = 1;
2329         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2330                       sizeof(ioc_data), &ioc_data, NULL);
2331
2332         obd = class_exp2obd(sbi->ll_dt_exp);
2333         if (obd == NULL) {
2334                 CERROR("Invalid LOV connection handle "LPX64"\n",
2335                        sbi->ll_dt_exp->exp_handle.h_cookie);
2336                 EXIT;
2337                 return;
2338         }
2339
2340         obd->obd_no_recov = 1;
2341         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2342                       sizeof(ioc_data), &ioc_data, NULL);
2343
2344         /*
2345          * really, we'd like to wait until there are no requests outstanding,
2346          * and then continue.  For now, we just invalidate the requests,
2347          * schedule, and hope.
2348          */
2349         schedule();
2350
2351         EXIT;
2352 }
2353
2354 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2355                   struct inode **inode, struct ptlrpc_request *req,
2356                   int offset, struct super_block *sb)
2357 {
2358         struct lustre_md md;
2359         int rc = 0;
2360
2361         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2362         if (rc)
2363                 RETURN(rc);
2364
2365         if (*inode) {
2366                 ll_update_inode(*inode, &md);
2367         } else {
2368                 LASSERT(sb);
2369                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2370                 if (*inode == NULL || is_bad_inode(*inode)) {
2371                         /* free the lsm if we allocated one above */
2372                         if (md.lsm != NULL)
2373                                 obd_free_memmd(dt_exp, &md.lsm);
2374                         if (md.mea != NULL)
2375                                 obd_free_memmd(md_exp,
2376                                                (struct lov_stripe_md**)&md.mea);
2377                         rc = -ENOMEM;
2378                         CERROR("new_inode -fatal: rc %d\n", rc);
2379                 }
2380         }
2381
2382         RETURN(rc);
2383 }
2384
2385 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2386 {
2387         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2388         struct lustre_mount_data *lmd = sbi->ll_lmd;
2389
2390         if (lmd) {
2391                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2392                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2393         }
2394         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2395         if (sbi->ll_remote && lmd)
2396                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2397
2398         if (lmd && lmd->lmd_pag)
2399                 seq_printf(m, ",pag");
2400
2401         return 0;
2402 }
2403
2404 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2405                char *filename, struct lustre_id *ret)
2406 {
2407         struct ptlrpc_request *request = NULL;
2408         struct mds_body *body;
2409         int rc;
2410
2411         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2412                              OBD_MD_FID, 0, &request);
2413         if (rc < 0) {
2414                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2415                        filename, rc);
2416                 return rc;
2417         }
2418
2419         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2420         LASSERT(body != NULL);
2421         LASSERT_REPSWABBED(request, 0);
2422
2423         *ret = body->id1;
2424         ptlrpc_req_finished(request);
2425
2426         return rc;
2427 }
2428 int ll_flush_cred(struct inode *inode)
2429 {
2430         struct ll_sb_info *sbi = ll_i2sbi(inode);
2431         int rc = 0;
2432
2433         /* XXX to avoid adding api, we simply use set_info() interface
2434          * to notify underlying obds. set_info() is more like a ioctl() now...
2435          */
2436         if (sbi->ll_md_exp) {
2437                 rc = obd_set_info(sbi->ll_md_exp,
2438                                   strlen("flush_cred"), "flush_cred",
2439                                   0, NULL);
2440                 if (rc)
2441                         return rc;
2442         }
2443
2444         if (sbi->ll_dt_exp) {
2445                 rc = obd_set_info(sbi->ll_dt_exp,
2446                                   strlen("flush_cred"), "flush_cred",
2447                                   0, NULL);
2448                 if (rc)
2449                         return rc;
2450         }
2451
2452         return rc;
2453 }