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         if (mds_security == NULL)
 157                 mds_security = "null";
 158
 159         err = obd_set_info(md_obd->obd_self_export, strlen("sec"), "sec",
 160                            strlen(mds_security), mds_security);
 161
 162         if (err) {
 163                 CERROR("LMV %s: failed to set security %s, err %d\n",
 164                         lmv, mds_security, err);
 165                 RETURN(err);
 166         }
 167
 168         if (pag) {
 169                 sec_flags = PTLRPC_SEC_FL_PAG;
 170                 err = obd_set_info(md_obd->obd_self_export,
 171                                    strlen("sec_flags"), "sec_flags",
 172                                    sizeof(sec_flags), &sec_flags);
 173                 if (err) {
 174                         OBD_FREE(data, sizeof(*data));
 175                         RETURN(err);
 176                 }
 177         }
 178
 179         err = obd_connect(&md_conn, md_obd, &sbi->ll_sb_uuid, data,
 180                           OBD_OPT_REAL_CLIENT);
 181         if (err == -EBUSY) {
 182                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 183                        " client is not a part.  Please wait for recovery to "
 184                        "complete, abort, or time out.\n", lmv);
 185                 GOTO(out, err);
 186         } else if (err) {
 187                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 188                 GOTO(out, err);
 189         }
 190
 191         sbi->ll_md_exp = class_conn2export(&md_conn);
 192
 193         err = obd_statfs(md_obd, &osfs, jiffies - HZ);
 194         if (err)
 195                 GOTO(out_disconnect, err);
 196
 197         if (!osfs.os_bsize) {
 198                 CERROR("Invalid block size is detected.");
 199                 GOTO(out_disconnect, err);
 200         }
 201
 202         sb->s_magic = LL_SUPER_MAGIC;
 203         sb->s_blocksize = osfs.os_bsize;
 204         sb->s_blocksize_bits = log2(osfs.os_bsize);
 205         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 206
 207         /* in 2.6.x FS is not allowed to form s_dev */
 208 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 209         {
 210                 kdev_t devno;
 211
 212                 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 213                                      strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 214
 215                 sb->s_dev = devno;
 216         }
 217 #endif
 218
 219         /* after statfs, we are supposed to have connected to MDSs,
 220          * so it's ok to check remote flag returned.
 221          */
 222         valsize = sizeof(&sbi->ll_remote);
 223         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 224                            &valsize, &sbi->ll_remote);
 225         if (err) {
 226                 CERROR("fail to obtain remote flag\n");
 227                 GOTO(out_disconnect, err);
 228         }
 229
 230 out_disconnect:
 231         if (err)
 232                 obd_disconnect(sbi->ll_md_exp, 0);
 233 out:
 234         RETURN(err);
 235 }
 236
 237 static int lustre_connect_ost(struct super_block *sb, char *lov,
 238                               struct obd_connect_data *data,
 239                               char *oss_security, int async, int pag)
 240 {
 241         struct ll_sb_info *sbi = ll_s2sbi(sb);
 242         struct lustre_handle dt_conn = {0, };
 243         struct obd_device *obd = NULL;
 244         unsigned long sec_flags;
 245         int err, mdsize;
 246
 247         obd = class_name2obd(lov);
 248         if (!obd) {
 249                 CERROR("OSC %s: not setup or attached\n", lov);
 250                 GOTO(out, err = -EINVAL);
 251         }
 252         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 253                      sizeof(async), &async);
 254
 255        if (oss_security == NULL)
 256                 oss_security = "null";
 257
 258         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 259                            strlen(oss_security), oss_security);
 260         if (err) {
 261                 CERROR("LOV %s: failed to set security %s, err %d\n",
 262                         lov, oss_security, err);
 263                 RETURN(err);
 264         }
 265
 266         if (pag) {
 267                 sec_flags = PTLRPC_SEC_FL_PAG;
 268                 err = obd_set_info(obd->obd_self_export,
 269                                    strlen("sec_flags"), "sec_flags",
 270                                    sizeof(sec_flags), &sec_flags);
 271                 if (err) {
 272                         OBD_FREE(data, sizeof(*data));
 273                         RETURN(err);
 274                 }
 275         }
 276
 277         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
 278         if (err == -EBUSY) {
 279                 CERROR("An OST (lov %s) is performing recovery, of which this"
 280                        " client is not a part.  Please wait for recovery to "
 281                        "complete, abort, or time out.\n", lov);
 282                 GOTO(out, err);
 283         } else if (err) {
 284                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 285                 GOTO(out, err);
 286         }
 287         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 288
 289         err = lustre_init_dt_desc(sbi);
 290
 291         if (err) {
 292                 CWARN("init dt_desc error %d \n", err);
 293                 GOTO(out, err = 0);
 294         }
 295         mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 296         obd_init_ea_size(sbi->ll_md_exp, mdsize, sbi->ll_dt_desc.ld_tgt_count *
 297                          sizeof(struct llog_cookie));
 298 out:
 299         RETURN(err);
 300 }
 301
 302 extern struct dentry_operations ll_d_ops;
 303
 304 static int lustre_init_root_inode(struct super_block *sb)
 305 {
 306         struct ll_sb_info *sbi = ll_s2sbi(sb);
 307         struct ptlrpc_request *request = NULL;
 308         struct inode *root = NULL;
 309         struct lustre_md md;
 310         int err = 0;
 311         ENTRY;
 312
 313         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 314         if (err) {
 315                 CERROR("cannot mds_connect: rc = %d\n", err);
 316                 GOTO(out, err);
 317         }
 318         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 319
 320         sb->s_op = &lustre_super_operations;
 321
 322         /* make root inode */
 323         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 324                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
 325                          NULL, NULL, 0, 0, NULL, &request);
 326         if (err) {
 327                 CERROR("md_getattr failed for root: rc = %d\n", err);
 328                 GOTO(out, err);
 329         }
 330
 331         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 332                                 sbi->ll_dt_exp, &md);
 333         if (err) {
 334                 CERROR("failed to understand root inode md: rc = %d\n", err);
 335                 ptlrpc_req_finished(request);
 336                 GOTO(out, err);
 337         }
 338
 339         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 340         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 341
 342         ptlrpc_req_finished(request);
 343
 344         if (root == NULL || is_bad_inode(root)) {
 345                 if (md.lsm != NULL)
 346                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 347                 if (md.mea != NULL)
 348                     obd_free_memmd(sbi->ll_md_exp,
 349                                    (struct lov_stripe_md**)&md.mea);
 350                 CERROR("lustre_lite: bad iget4 for root\n");
 351                 GOTO(out_root, err = -EBADF);
 352         }
 353         sb->s_root = d_alloc_root(root);
 354         sb->s_root->d_op = &ll_d_ops;
 355 out_root:
 356         if (err)
 357                 iput(root);
 358 out:
 359         RETURN(err);
 360 }
 361
 362 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 363                              char *gkc, int async,  char *mds_security,
 364                              char *oss_security, __u32 *nllu, int pag,
 365                              __u64 *remote)
 366 {
 367         struct ll_sb_info *sbi = ll_s2sbi(sb);
 368         struct obd_connect_data *data;
 369         int err;
 370         ENTRY;
 371
 372         /*process the connect flags*/
 373         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 374                        (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 375                 CERROR("wrong remote flag "LPX64"\n", *remote);
 376                 RETURN(-EINVAL);
 377         }
 378
 379         OBD_ALLOC(data, sizeof(*data));
 380         if (!data)
 381                 RETURN(-ENOMEM);
 382
 383         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 384                                               OBD_CONNECT_REMOTE);
 385         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 386
 387         if (proc_lustre_fs_root) {
 388                 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
 389                                                   sb, lov, lmv);
 390                 if (err < 0)
 391                         CERROR("could not register mount in /proc/lustre");
 392         }
 393
 394         /*connect mds */
 395         err = lustre_connect_mds(sb, lmv, data, mds_security, async, pag);
 396         if (err)
 397                 GOTO(out, err);
 398
 399         /*connect OST*/
 400         err = lustre_connect_ost(sb, lov, data, oss_security, async, pag);
 401         if (err)
 402                 GOTO(out_lmv, err);
 403
 404         err = lustre_init_crypto(sb, gkc, data, async);
 405         if (err) {
 406                 CERROR("Could not connect to GSS err %d\n", err);
 407                 err = 0;
 408         }
 409         /*connect GSS*/
 410         err = lustre_init_root_inode(sb);
 411         if (err)
 412                 GOTO(out_gks, err);
 413
 414         err = ll_close_thread_start(&sbi->ll_lcq);
 415         if (err) {
 416                 CERROR("cannot start close thread: rc %d\n", err);
 417                 GOTO(out_root, err);
 418         }
 419
 420         ll_gns_add_timer(sbi);
 421
 422         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 423            backing dev info assigned to inode mapping is used for
 424            determining maximal readahead. */
 425 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 426     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 427         /* bug 2805 - set VM readahead to zero */
 428         vm_max_readahead = vm_min_readahead = 0;
 429 #endif
 430         sb->s_flags |= MS_POSIXACL;
 431 #ifdef S_PDIROPS
 432         CWARN("Enabling PDIROPS\n");
 433         sb->s_flags |= S_PDIROPS;
 434 #endif
 435         if (data != NULL)
 436                 OBD_FREE(data, sizeof(*data));
 437         RETURN(err);
 438 out_root:
 439         if (sb->s_root)
 440                 dput(sb->s_root);
 441 out_gks:
 442         lustre_destroy_crypto(sb);
 443 out_lmv:
 444         obd_disconnect(sbi->ll_md_exp, 0);
 445 out:
 446         if (data != NULL)
 447                 OBD_FREE(data, sizeof(*data));
 448         lprocfs_unregister_mountpoint(sbi);
 449         RETURN(err);
 450 }
 451
 452 void lustre_common_put_super(struct super_block *sb)
 453 {
 454         struct ll_sb_info *sbi = ll_s2sbi(sb);
 455         struct hlist_node *tmp, *next;
 456         ENTRY;
 457
 458         ll_gns_del_timer(sbi);
 459         ll_close_thread_stop(sbi->ll_lcq);
 460
 461         lustre_destroy_crypto(sb);
 462
 463         list_del(&sbi->ll_conn_chain);
 464         obd_disconnect(sbi->ll_dt_exp, 0);
 465
 466         lprocfs_unregister_mountpoint(sbi);
 467         if (sbi->ll_proc_root) {
 468                 lprocfs_remove(sbi->ll_proc_root);
 469                 sbi->ll_proc_root = NULL;
 470         }
 471
 472         obd_disconnect(sbi->ll_md_exp, 0);
 473
 474         // We do this to get rid of orphaned dentries. That is not really trw.
 475         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 476                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 477                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 478                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 479                 shrink_dcache_parent(dentry);
 480         }
 481         EXIT;
 482 }
 483
 484 char *ll_read_opt(const char *opt, char *data)
 485 {
 486         char *value;
 487         char *retval;
 488         ENTRY;
 489
 490         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 491         if (strncmp(opt, data, strlen(opt)))
 492                 RETURN(NULL);
 493         if ((value = strchr(data, '=')) == NULL)
 494                 RETURN(NULL);
 495
 496         value++;
 497         OBD_ALLOC(retval, strlen(value) + 1);
 498         if (!retval) {
 499                 CERROR("out of memory!\n");
 500                 RETURN(NULL);
 501         }
 502
 503         memcpy(retval, value, strlen(value)+1);
 504         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 505         RETURN(retval);
 506 }
 507
 508 int ll_set_opt(const char *opt, char *data, int fl)
 509 {
 510         ENTRY;
 511
 512         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 513         if (strncmp(opt, data, strlen(opt)))
 514                 RETURN(0);
 515         else
 516                 RETURN(fl);
 517 }
 518
 519 void ll_options(char *options, char **lov, char **lmv, char **gkc,
 520                 char **mds_sec, char **oss_sec, int *async, int *flags)
 521 {
 522         char *this_char;
 523 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 524         char *opt_ptr = options;
 525 #endif
 526         ENTRY;
 527
 528         if (!options) {
 529                 EXIT;
 530                 return;
 531         }
 532
 533         *async = 0;
 534 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 535         for (this_char = strtok (options, ",");
 536              this_char != NULL;
 537              this_char = strtok (NULL, ",")) {
 538 #else
 539         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 540 #endif
 541                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 542                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 543                         continue;
 544                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 545                         continue;
 546                 if (!*gkc && (*gkc = ll_read_opt("gkc", this_char)))
 547                         continue;
 548                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 549                         *async = 1;
 550                         continue;
 551                 }
 552                 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
 553                         continue;
 554                 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
 555                         continue;
 556                 if (!(*flags & LL_SBI_NOLCK) &&
 557                     ((*flags) = (*flags) |
 558                                 ll_set_opt("nolock", this_char,
 559                                            LL_SBI_NOLCK)))
 560                         continue;
 561         }
 562
 563         EXIT;
 564 }
 565
 566 void ll_lli_init(struct ll_inode_info *lli)
 567 {
 568         sema_init(&lli->lli_open_sem, 1);
 569         sema_init(&lli->lli_size_sem, 1);
 570         lli->lli_flags = 0;
 571         lli->lli_size_pid = 0;
 572         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 573         spin_lock_init(&lli->lli_lock);
 574         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 575         INIT_LIST_HEAD(&lli->lli_close_item);
 576         lli->lli_inode_magic = LLI_INODE_MAGIC;
 577         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 578         sema_init(&lli->lli_och_sem, 1);
 579         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 580         lli->lli_mds_exec_och = NULL;
 581         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 582         lli->lli_open_fd_exec_count = 0;
 583         lli->lli_audit_mask = AUDIT_OFF;
 584         lli->lli_key_info = NULL;
 585         init_waitqueue_head(&lli->lli_dirty_wait);
 586         lli->lli_io_epoch = 0;
 587         INIT_LIST_HEAD(&lli->lli_capas);
 588 }
 589
 590 int ll_fill_super(struct super_block *sb, void *data, int silent)
 591 {
 592         struct ll_sb_info *sbi;
 593         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 594         char *mds_sec = NULL;
 595         char *oss_sec = NULL;
 596         int async, err;
 597         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 598         __u64 remote_flag = 0;
 599         ENTRY;
 600
 601         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 602
 603         sbi = lustre_init_sbi(sb);
 604         if (!sbi)
 605                 RETURN(-ENOMEM);
 606
 607         sbi->ll_flags |= LL_SBI_READAHEAD;
 608         ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
 609                    &async, &sbi->ll_flags);
 610
 611         if (!lov || !lmv) {
 612                 CERROR("no osc %p or no mdc %p\n", lov, lmv);
 613                 GOTO(out, err = -EINVAL);
 614         }
 615
 616         err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
 617                                        oss_sec, nllu, 0, &remote_flag);
 618         EXIT;
 619 out:
 620         if (err)
 621                 lustre_free_sbi(sb);
 622
 623         if (lmv)
 624                 OBD_FREE(lmv, strlen(lmv) + 1);
 625         if (lov)
 626                 OBD_FREE(lov, strlen(lov) + 1);
 627         if (mds_sec)
 628                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 629         if (oss_sec)
 630                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 631         if (gkc)
 632                 OBD_FREE(gkc, strlen(gkc) + 1);
 633
 634         return err;
 635 } /* ll_read_super */
 636
 637 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 638                               struct config_llog_instance *cfg, int allow_recov)
 639 {
 640         struct lustre_cfg *lcfg = NULL;
 641         struct lustre_cfg_bufs bufs;
 642         struct portals_cfg pcfg;
 643         char *peer = "MDS_PEER_UUID";
 644         struct obd_device *obd;
 645         struct lustre_handle md_conn = {0, };
 646         struct obd_export *exp;
 647         char *name = "mdc_dev";
 648         class_uuid_t uuid;
 649         struct obd_uuid lmv_uuid;
 650         struct llog_ctxt *ctxt;
 651         int rc = 0, err = 0;
 652         ENTRY;
 653
 654         if (lmd_bad_magic(lmd))
 655                 RETURN(-EINVAL);
 656
 657         generate_random_uuid(uuid);
 658         class_uuid_unparse(uuid, &lmv_uuid);
 659
 660         if (lmd->lmd_local_nid) {
 661                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 662                 pcfg.pcfg_nal = lmd->lmd_nal;
 663                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 664                 rc = libcfs_nal_cmd(&pcfg);
 665                 if (rc < 0)
 666                         GOTO(out, rc);
 667         }
 668
 669         if (lmd->lmd_nal == SOCKNAL ||
 670             lmd->lmd_nal == OPENIBNAL ||
 671             lmd->lmd_nal == IIBNAL ||
 672             lmd->lmd_nal == VIBNAL ||
 673             lmd->lmd_nal == RANAL) {
 674                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 675                 pcfg.pcfg_nal     = lmd->lmd_nal;
 676                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 677                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 678                 pcfg.pcfg_misc    = lmd->lmd_port;
 679                 rc = libcfs_nal_cmd(&pcfg);
 680                 if (rc < 0)
 681                         GOTO(out, rc);
 682         }
 683         lustre_cfg_bufs_reset(&bufs, name);
 684         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 685
 686         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 687         lcfg->lcfg_nal = lmd->lmd_nal;
 688         lcfg->lcfg_nid = lmd->lmd_server_nid;
 689         LASSERT(lcfg->lcfg_nal);
 690         LASSERT(lcfg->lcfg_nid);
 691         err = class_process_config(lcfg);
 692         lustre_cfg_free(lcfg);
 693         if (err < 0)
 694                 GOTO(out_del_conn, err);
 695
 696         lustre_cfg_bufs_reset(&bufs, name);
 697         lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
 698         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 699
 700         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 701         err = class_process_config(lcfg);
 702         lustre_cfg_free(lcfg);
 703         if (err < 0)
 704                 GOTO(out_del_uuid, err);
 705
 706         lustre_cfg_bufs_reset(&bufs, name);
 707         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 708         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 709
 710         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 711         err = class_process_config(lcfg);
 712         lustre_cfg_free(lcfg);
 713         if (err < 0)
 714                 GOTO(out_detach, err);
 715
 716         obd = class_name2obd(name);
 717         if (obd == NULL)
 718                 GOTO(out_cleanup, rc = -EINVAL);
 719
 720         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 721                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 722         if (rc)
 723                 GOTO(out_cleanup, rc);
 724
 725         if (lmd->lmd_pag) {
 726                 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
 727                 rc = obd_set_info(obd->obd_self_export,
 728                                   strlen("sec_flags"), "sec_flags",
 729                                   sizeof(sec_flags), &sec_flags);
 730                 if (rc)
 731                         GOTO(out_cleanup, rc);
 732         }
 733
 734         /* Disable initial recovery on this import */
 735         rc = obd_set_info(obd->obd_self_export,
 736                           strlen("initial_recov"), "initial_recov",
 737                           sizeof(allow_recov), &allow_recov);
 738         if (rc)
 739                 GOTO(out_cleanup, rc);
 740
 741         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, OBD_OPT_REAL_CLIENT);
 742         if (rc) {
 743                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 744                 GOTO(out_cleanup, rc);
 745         }
 746
 747         exp = class_conn2export(&md_conn);
 748
 749         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 750         rc = class_config_process_llog(ctxt, profile, cfg);
 751         if (rc)
 752                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 753
 754         err = obd_disconnect(exp, 0);
 755
 756         EXIT;
 757 out_cleanup:
 758         lustre_cfg_bufs_reset(&bufs, name);
 759         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 760         err = class_process_config(lcfg);
 761         lustre_cfg_free(lcfg);
 762         if (err < 0)
 763                 GOTO(out, err);
 764 out_detach:
 765         lustre_cfg_bufs_reset(&bufs, name);
 766         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 767         err = class_process_config(lcfg);
 768         lustre_cfg_free(lcfg);
 769         if (err < 0)
 770                 GOTO(out, err);
 771
 772 out_del_uuid:
 773         lustre_cfg_bufs_reset(&bufs, name);
 774         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 775         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 776         err = class_process_config(lcfg);
 777         lustre_cfg_free(lcfg);
 778
 779 out_del_conn:
 780         if (lmd->lmd_nal == SOCKNAL ||
 781             lmd->lmd_nal == OPENIBNAL ||
 782             lmd->lmd_nal == IIBNAL ||
 783             lmd->lmd_nal == VIBNAL ||
 784             lmd->lmd_nal == RANAL) {
 785                 int err2;
 786
 787                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 788                 pcfg.pcfg_nal     = lmd->lmd_nal;
 789                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 790                 pcfg.pcfg_flags   = 1;          /* single_share */
 791                 err2 = libcfs_nal_cmd(&pcfg);
 792                 if (err2 && !err)
 793                         err = err2;
 794                 if (err < 0)
 795                         GOTO(out, err);
 796         }
 797 out:
 798         if (rc == 0)
 799                 rc = err;
 800
 801         return rc;
 802 }
 803
 804 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 805 {
 806         struct lustre_cfg *lcfg;
 807         struct lustre_cfg_bufs bufs;
 808         struct obd_device *obd;
 809         int next = 0;
 810
 811         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
 812                 int err;
 813
 814                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 815                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 816                 err = class_process_config(lcfg);
 817                 if (err) {
 818                         CERROR("cleanup failed: %s\n", obd->obd_name);
 819                         //continue;
 820                 }
 821
 822                 lcfg->lcfg_command = LCFG_DETACH;
 823                 err = class_process_config(lcfg);
 824                 lustre_cfg_free(lcfg);
 825                 if (err) {
 826                         CERROR("detach failed: %s\n", obd->obd_name);
 827                         //continue;
 828                 }
 829         }
 830
 831         if (sbi->ll_lmd != NULL)
 832                 class_del_profile(sbi->ll_lmd->lmd_profile);
 833 }
 834
 835 static int lustre_process_profile(struct super_block *sb,
 836                                   struct lustre_mount_data *lmd,
 837                                   char **lov, char **lmv, char **gkc)
 838 {
 839         struct ll_sb_info *sbi = ll_s2sbi(sb);
 840         struct config_llog_instance cfg;
 841         struct lustre_profile *lprof;
 842         int len, err = 0;
 843         ENTRY;
 844
 845         if (!lmd->lmd_profile)
 846                 RETURN(0);
 847
 848         if (lmd->lmd_mds[0] == '\0') {
 849                 CERROR("no mds name\n");
 850                 GOTO(out, err = -EINVAL);
 851         }
 852         lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
 853         lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
 854
 855         OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 856         if (sbi->ll_lmd == NULL)
 857                 GOTO(out, err = -ENOMEM);
 858         memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 859
 860         /* generate a string unique to this super, let's try the address of the
 861          * super itself. */
 862         len = (sizeof(sb) * 2) + 1;
 863         OBD_ALLOC(sbi->ll_instance, len);
 864         if (sbi->ll_instance == NULL)
 865                 GOTO(out, err = -ENOMEM);
 866         sprintf(sbi->ll_instance, "%p", sb);
 867
 868         cfg.cfg_instance = sbi->ll_instance;
 869         cfg.cfg_uuid = sbi->ll_sb_uuid;
 870         cfg.cfg_local_nid = lmd->lmd_local_nid;
 871         err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 872         if (err < 0) {
 873                 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 874                 GOTO(out, err);
 875         }
 876
 877         lprof = class_get_profile(lmd->lmd_profile);
 878         if (lprof == NULL) {
 879                 CERROR("No profile found: %s\n", lmd->lmd_profile);
 880                 GOTO(out, err = -EINVAL);
 881         }
 882
 883         OBD_ALLOC(*lov, strlen(lprof->lp_lov) +
 884                   strlen(sbi->ll_instance) + 2);
 885         if (*lov == NULL)
 886                 GOTO(out, err = -ENOMEM);
 887
 888         sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 889
 890         OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
 891                   strlen(sbi->ll_instance) + 2);
 892         if (*lmv == NULL)
 893                 GOTO(out_free_lov, err = -ENOMEM);
 894
 895         sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 896
 897         if (lprof->lp_gkc) {
 898                OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
 899                          strlen(sbi->ll_instance) + 2);
 900                if (*gkc == NULL)
 901                        GOTO(out_free_lmv, err = -ENOMEM);
 902
 903                sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
 904         }
 905
 906         RETURN(err);
 907 out_free_lmv:
 908         OBD_FREE(*lmv, strlen(lprof->lp_lmv) +
 909                  strlen(sbi->ll_instance) + 2);
 910 out_free_lov:
 911         OBD_FREE(*lov, strlen(lprof->lp_lov) +
 912                  strlen(sbi->ll_instance) + 2);
 913 out:
 914         return err;
 915 }
 916
 917 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
 918 {
 919         struct lustre_mount_data *lmd = sbi->ll_lmd;
 920         struct config_llog_instance cfg;
 921         char *cl_prof;
 922         int len, err = 0;
 923         ENTRY;
 924
 925         if (!lmd)
 926                 RETURN(err);
 927
 928         len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
 929
 930         if (force_umount) {
 931                 CERROR("force umount, doing manual cleanup\n");
 932                 lustre_manual_cleanup(sbi);
 933                 GOTO(free_lmd, 0);
 934
 935         }
 936         if (sbi->ll_instance != NULL) {
 937                 cfg.cfg_instance = sbi->ll_instance;
 938                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 939
 940                 OBD_ALLOC(cl_prof, len);
 941                 if (!cl_prof) {
 942                         CERROR("can't allocate memory, "
 943                                "skipping processing cleanup profile.\n");
 944                         GOTO(free_lmd, err = -ENOMEM);
 945                 }
 946
 947                 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 948                 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 949                 if (err < 0) {
 950                         CERROR("Unable to process log: %s\n", cl_prof);
 951                         lustre_manual_cleanup(sbi);
 952                 }
 953                 OBD_FREE(cl_prof, len);
 954         }
 955         EXIT;
 956 free_lmd:
 957         if (sbi->ll_instance)
 958                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 959         OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 960         return err;
 961 }
 962
 963 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 964 {
 965         struct lustre_mount_data * lmd = data;
 966         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 967         struct ll_sb_info *sbi;
 968         int err;
 969         ENTRY;
 970
 971         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 972         if (lmd_bad_magic(lmd))
 973                 RETURN(-EINVAL);
 974
 975         sbi = lustre_init_sbi(sb);
 976         if (!sbi)
 977                 RETURN(-ENOMEM);
 978
 979         sbi->ll_flags |= LL_SBI_READAHEAD;
 980
 981         err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
 982         if (err) {
 983                 CERROR("Can not process the profile err %d \n", err);
 984                 GOTO(out_free, err);
 985         }
 986         if (!lov || !lmv) {
 987                 CERROR("no osc %p or no mdc %p \n", lov, lmv);
 988                 GOTO(out_free, err = -EINVAL);
 989         }
 990
 991         err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
 992                                        lmd->lmd_mds_security,
 993                                        lmd->lmd_oss_security,
 994                                        &lmd->lmd_nllu, lmd->lmd_pag,
 995                                        &lmd->lmd_remote_flag);
 996
 997         if (err)
 998                 GOTO(out_free, err);
 999
1000         EXIT;
1001 out_dev:
1002         if (lmv)
1003                 OBD_FREE(lmv, strlen(lmv) + 1);
1004         if (lov)
1005                 OBD_FREE(lov, strlen(lov) + 1);
1006         if (gkc)
1007                 OBD_FREE(gkc, strlen(gkc) + 1);
1008
1009         return err;
1010 out_free:
1011         lustre_clean_profile(sbi, 0);
1012         lustre_free_sbi(sb);
1013         goto out_dev;
1014
1015 } /* lustre_fill_super */
1016
1017 void lustre_put_super(struct super_block *sb)
1018 {
1019         struct obd_device *obd;
1020         struct ll_sb_info *sbi = ll_s2sbi(sb);
1021         int force_umount = 0;
1022         ENTRY;
1023
1024         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1025         obd = class_exp2obd(sbi->ll_md_exp);
1026         if (obd)
1027                 force_umount = obd->obd_no_recov;
1028         obd = NULL;
1029
1030         lustre_common_put_super(sb);
1031         lustre_clean_profile(sbi, force_umount);
1032         lustre_free_sbi(sb);
1033
1034         EXIT;
1035 } /* lustre_put_super */
1036
1037 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1038 {
1039         struct lustre_mount_data *lmd = sbi->ll_lmd;
1040         char *profile = lmd->lmd_profile, *name = NULL;
1041         struct config_llog_instance cfg;
1042         int rc, namelen =  0, version;
1043         struct llog_ctxt *ctxt;
1044         ENTRY;
1045
1046         if (profile == NULL)
1047                 RETURN(0);
1048         if (lmd == NULL) {
1049                 CERROR("Client not mounted with zero-conf; cannot "
1050                        "process update log.\n");
1051                 RETURN(0);
1052         }
1053
1054         cfg.cfg_instance = sbi->ll_instance;
1055         cfg.cfg_uuid = sbi->ll_sb_uuid;
1056         cfg.cfg_local_nid = lmd->lmd_local_nid;
1057
1058         namelen = strlen(profile) + 20; /* -clean-######### */
1059         OBD_ALLOC(name, namelen);
1060         if (name == NULL)
1061                 RETURN(-ENOMEM);
1062
1063         if (clean) {
1064                 version = sbi->ll_config_version - 1;
1065                 sprintf(name, "%s-clean-%d", profile, version);
1066         } else {
1067                 version = sbi->ll_config_version + 1;
1068                 sprintf(name, "%s-%d", profile, version);
1069         }
1070
1071         CWARN("Applying configuration log %s\n", name);
1072
1073         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1074                                 LLOG_CONFIG_REPL_CTXT);
1075         rc = class_config_process_llog(ctxt, name, &cfg);
1076         if (rc == 0)
1077                 sbi->ll_config_version = version;
1078         CWARN("Finished applying configuration log %s: %d\n", name, rc);
1079
1080         if (rc == 0 && clean == 0) {
1081                 struct lov_desc desc;
1082                 __u32 valsize;
1083                 int rc = 0;
1084
1085                 valsize = sizeof(desc);
1086                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1087                                   "lovdesc", &valsize, &desc);
1088
1089                 rc = obd_init_ea_size(sbi->ll_md_exp,
1090                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
1091                                       (desc.ld_tgt_count *
1092                                        sizeof(struct llog_cookie)));
1093         }
1094         OBD_FREE(name, namelen);
1095         RETURN(rc);
1096 }
1097
1098 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1099 {
1100         struct inode *inode = NULL;
1101
1102         /* NOTE: we depend on atomic igrab() -bzzz */
1103         lock_res_and_lock(lock);
1104         if (lock->l_ast_data) {
1105                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1106                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1107                         inode = igrab(lock->l_ast_data);
1108                 } else {
1109                         struct timeval now;
1110                         do_gettimeofday(&now);
1111                         inode = lock->l_ast_data;
1112                         LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1113                                    lock->l_enqueued_time.tv_sec,
1114                                    lock->l_enqueued_time.tv_usec,
1115                                    now.tv_sec, now.tv_usec);
1116                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1117                                "l_ast_data %p is bogus: magic %0x8\n",
1118                                lock->l_ast_data, lli->lli_inode_magic);
1119                         CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1120                                inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1121                         inode = NULL;
1122                         unlock_res_and_lock(lock);
1123                         LBUG();
1124                 }
1125         }
1126         unlock_res_and_lock(lock);
1127         return inode;
1128 }
1129
1130 int null_if_equal(struct ldlm_lock *lock, void *data)
1131 {
1132         if (data == lock->l_ast_data) {
1133                 lock->l_ast_data = NULL;
1134
1135                 if (lock->l_req_mode != lock->l_granted_mode)
1136                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1137         }
1138
1139         return LDLM_ITER_CONTINUE;
1140 }
1141
1142 static void remote_acl_free(struct remote_acl *racl);
1143
1144 void ll_clear_inode(struct inode *inode)
1145 {
1146         struct lustre_id id;
1147         struct ll_inode_info *lli = ll_i2info(inode);
1148         struct ll_sb_info *sbi = ll_i2sbi(inode);
1149         struct obd_capa *ocapa, *tmp;
1150         ENTRY;
1151
1152         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1153                inode->i_generation, inode);
1154
1155         LASSERT(ll_is_inode_dirty(inode) == 0);
1156         ll_inode2id(&id, inode);
1157
1158         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1159         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1160
1161         LASSERT(!lli->lli_open_fd_write_count);
1162         LASSERT(!lli->lli_open_fd_read_count);
1163         LASSERT(!lli->lli_open_fd_exec_count);
1164         if (lli->lli_mds_write_och)
1165                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1166         if (lli->lli_mds_exec_och)
1167                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1168         if (lli->lli_mds_read_och)
1169                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1170         if (lli->lli_smd)
1171                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1172                                   null_if_equal, inode);
1173
1174         if (lli->lli_smd) {
1175                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1176                 lli->lli_smd = NULL;
1177         }
1178
1179         if (lli->lli_mea) {
1180                 obd_free_memmd(sbi->ll_md_exp,
1181                                (struct lov_stripe_md **) &lli->lli_mea);
1182                 lli->lli_mea = NULL;
1183         }
1184         ll_crypto_destroy_inode_key(inode);
1185         if (lli->lli_symlink_name) {
1186                 OBD_FREE(lli->lli_symlink_name,
1187                          strlen(lli->lli_symlink_name) + 1);
1188                 lli->lli_symlink_name = NULL;
1189         }
1190
1191         if (lli->lli_posix_acl) {
1192                 LASSERT(lli->lli_remote_acl == NULL);
1193                 posix_acl_release(lli->lli_posix_acl);
1194                 lli->lli_posix_acl = NULL;
1195         }
1196
1197         if (lli->lli_remote_acl) {
1198                 LASSERT(lli->lli_posix_acl == NULL);
1199                 remote_acl_free(lli->lli_remote_acl);
1200                 lli->lli_remote_acl = NULL;
1201         }
1202
1203         list_for_each_entry_safe(ocapa, tmp, &lli->lli_capas, u.client.lli_list)
1204                 capa_put(ocapa);
1205
1206         lli->lli_inode_magic = LLI_INODE_DEAD;
1207         EXIT;
1208 }
1209
1210 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1211  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1212  * keep these values until such a time that objects are allocated for it.
1213  * We do the MDS operations first, as it is checking permissions for us.
1214  * We don't to the MDS RPC if there is nothing that we want to store there,
1215  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1216  * going to do an RPC anyways.
1217  *
1218  * If we are doing a truncate, we will send the mtime and ctime updates
1219  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1220  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1221  * at the same time.
1222  */
1223 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1224 {
1225         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1226         struct ll_inode_info *lli = ll_i2info(inode);
1227         struct ll_sb_info *sbi = ll_i2sbi(inode);
1228         struct ptlrpc_request *request = NULL;
1229         struct mdc_op_data *op_data;
1230         int ia_valid = attr->ia_valid;
1231         int err, rc = 0;
1232         ENTRY;
1233
1234         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1235         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1236
1237         if (ia_valid & ATTR_SIZE) {
1238                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1239                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1240                                attr->ia_size, ll_file_maxbytes(inode));
1241                         RETURN(-EFBIG);
1242                 }
1243
1244                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1245         }
1246
1247         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1248         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1249                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1250                         RETURN(-EPERM);
1251         }
1252
1253         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1254         if (attr->ia_valid & ATTR_CTIME) {
1255                 attr->ia_ctime = CURRENT_TIME;
1256                 attr->ia_valid |= ATTR_CTIME_SET;
1257         }
1258         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1259                 attr->ia_atime = CURRENT_TIME;
1260                 attr->ia_valid |= ATTR_ATIME_SET;
1261         }
1262         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1263                 attr->ia_mtime = CURRENT_TIME;
1264                 attr->ia_valid |= ATTR_MTIME_SET;
1265         }
1266
1267         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1268                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1269                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1270                        LTIME_S(CURRENT_TIME));
1271
1272         /* If only OST attributes being set on objects, don't do MDS RPC.
1273          * In that case, we need to check permissions and update the local
1274          * inode ourselves so we can call obdo_from_inode() always. */
1275         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1276                 struct lustre_md md;
1277                 void *key = NULL;
1278                 int  key_size = 0;
1279
1280                 OBD_ALLOC(op_data, sizeof(*op_data));
1281                 if (op_data == NULL)
1282                         RETURN(-ENOMEM);
1283                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
1284
1285                 if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) {
1286                         rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1287                                                &key_size);
1288                         if (rc) {
1289                                 CERROR("can not get right mac, rc=%d\n", rc);
1290                                 if (key && key_size)
1291                                         OBD_FREE(key, key_size);
1292                                 RETURN(rc);
1293                         }
1294                 }
1295                 rc = md_setattr(sbi->ll_md_exp, op_data,
1296                                 attr, key, key_size, NULL, 0, NULL,
1297                                 0, &request);
1298                 OBD_FREE(op_data, sizeof(*op_data));
1299
1300                 if (key && key_size)
1301                         OBD_FREE(key, key_size);
1302                 if (rc) {
1303                         ptlrpc_req_finished(request);
1304                         if (rc != -EPERM && rc != -EACCES)
1305                                 CERROR("md_setattr fails: rc = %d\n", rc);
1306                         RETURN(rc);
1307                 }
1308                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1309                                        sbi->ll_dt_exp, &md);
1310                 if (rc) {
1311                         ptlrpc_req_finished(request);
1312                         RETURN(rc);
1313                 }
1314
1315                 if (attr->ia_valid & ATTR_SIZE) {
1316                         rc = ll_set_trunc_capa(request, 0, inode);
1317                         if (rc) {
1318                                 ptlrpc_req_finished(request);
1319                                 RETURN(rc);
1320                         }
1321                 }
1322
1323                 /* We call inode_setattr to adjust timestamps, but we first
1324                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1325                  *
1326                  * NB: ATTR_SIZE will only be set at this point if the size
1327                  * resides on the MDS, ie, this file has no objects. */
1328                 attr->ia_valid &= ~ATTR_SIZE;
1329
1330                 /*
1331                  * assigning inode_setattr() to @err to disable warning that
1332                  * function's result should be checked by by caller. error is
1333                  * impossible here, as vmtruncate() control path is disabled.
1334                  */
1335                 err = inode_setattr(inode, attr);
1336                 ll_update_inode(inode, &md);
1337                 ptlrpc_req_finished(request);
1338
1339                 if (!lsm || !S_ISREG(inode->i_mode)) {
1340                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1341                         RETURN(0);
1342                 }
1343         } else {
1344                 /* The OST doesn't check permissions, but the alternative is
1345                  * a gratuitous RPC to the MDS.  We already rely on the client
1346                  * to do read/write/truncate permission checks, so is mtime OK?
1347                  */
1348                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1349                         /* from sys_utime() */
1350                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1351                                 if (current->fsuid != inode->i_uid &&
1352                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1353                                         RETURN(rc);
1354                         } else {
1355                                 /* from inode_change_ok() */
1356                                 if (current->fsuid != inode->i_uid &&
1357                                     !capable(CAP_FOWNER))
1358                                         RETURN(-EPERM);
1359                         }
1360                 }
1361
1362                 if (lsm)
1363                         attr->ia_valid &= ~ATTR_SIZE;
1364
1365                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1366                 err = inode_setattr(inode, attr);
1367                 /*
1368                  * assigning inode_setattr() to @err to disable warning that
1369                  * function's result should be checked by by caller. error is
1370                  * impossible here, as vmtruncate() control path is disabled.
1371                  */
1372         }
1373
1374         /* We really need to get our PW lock before we change inode->i_size.
1375          * If we don't we can race with other i_size updaters on our node, like
1376          * ll_file_read.  We can also race with i_size propogation to other
1377          * nodes through dirtying and writeback of final cached pages.  This
1378          * last one is especially bad for racing o_append users on other
1379          * nodes. */
1380         if (ia_valid & ATTR_SIZE) {
1381                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1382                                                            OBD_OBJECT_EOF } };
1383                 struct lustre_handle lockh = { 0 };
1384                 int err, ast_flags = 0;
1385                 /* XXX when we fix the AST intents to pass the discard-range
1386                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1387                  * XXX here. */
1388                 if (attr->ia_size == 0)
1389                         ast_flags = LDLM_AST_DISCARD_DATA;
1390
1391                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1392                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1393
1394                 if (rc != 0)
1395                         RETURN(rc);
1396
1397                 down(&lli->lli_size_sem);
1398                 lli->lli_size_pid = current->pid;
1399                 rc = vmtruncate(inode, attr->ia_size);
1400                 if (rc != 0) {
1401                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1402                         lli->lli_size_pid = 0;
1403                         up(&lli->lli_size_sem);
1404                 }
1405
1406                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1407                 if (err) {
1408                         CERROR("ll_extent_unlock failed: %d\n", err);
1409                         if (!rc)
1410                                 rc = err;
1411                 }
1412         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1413                 struct obdo *oa = NULL;
1414
1415                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1416                        inode->i_ino, LTIME_S(attr->ia_mtime));
1417
1418                 oa = obdo_alloc();
1419                 if (oa == NULL)
1420                         RETURN(-ENOMEM);
1421
1422                 oa->o_id = lsm->lsm_object_id;
1423                 oa->o_gr = lsm->lsm_object_gr;
1424                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1425
1426                 /* adding uid and gid, needed for quota */
1427                 if (ia_valid & ATTR_UID) {
1428                         oa->o_uid = inode->i_uid;
1429                         oa->o_valid |= OBD_MD_FLUID;
1430                 }
1431
1432                 if (ia_valid & ATTR_GID) {
1433                         oa->o_gid = inode->i_gid;
1434                         oa->o_valid |= OBD_MD_FLGID;
1435                 }
1436
1437                 *(obdo_id(oa)) = lli->lli_id;
1438                 oa->o_valid |= OBD_MD_FLIFID;
1439
1440                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1441                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1442                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL);
1443                 obdo_free(oa);
1444                 if (rc)
1445                         CERROR("obd_setattr fails: rc = %d\n", rc);
1446         }
1447
1448         RETURN(rc);
1449 }
1450
1451 int ll_setattr(struct dentry *de, struct iattr *attr)
1452 {
1453         LASSERT(de->d_inode);
1454         return ll_setattr_raw(de->d_inode, attr);
1455 }
1456
1457 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1458                        unsigned long max_age)
1459 {
1460         struct ll_sb_info *sbi = ll_s2sbi(sb);
1461         struct obd_statfs obd_osfs;
1462         int rc;
1463         ENTRY;
1464
1465         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1466         if (rc) {
1467                 CERROR("obd_statfs fails: rc = %d\n", rc);
1468                 RETURN(rc);
1469         }
1470
1471         osfs->os_type = sb->s_magic;
1472
1473         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1474                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1475
1476         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1477         if (rc) {
1478                 CERROR("obd_statfs fails: rc = %d\n", rc);
1479                 RETURN(rc);
1480         }
1481
1482         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1483                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1484                obd_osfs.os_files);
1485
1486         osfs->os_blocks = obd_osfs.os_blocks;
1487         osfs->os_bfree = obd_osfs.os_bfree;
1488         osfs->os_bavail = obd_osfs.os_bavail;
1489
1490         /* If we don't have as many objects free on the OST as inodes
1491          * on the MDS, we reduce the total number of inodes to
1492          * compensate, so that the "inodes in use" number is correct.
1493          */
1494         if (obd_osfs.os_ffree < osfs->os_ffree) {
1495                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1496                         obd_osfs.os_ffree;
1497                 osfs->os_ffree = obd_osfs.os_ffree;
1498         }
1499
1500         RETURN(rc);
1501 }
1502
1503 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1504 {
1505         struct obd_statfs osfs;
1506         int rc;
1507
1508         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1509         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1510
1511         /* For now we will always get up-to-date statfs values, but in the
1512          * future we may allow some amount of caching on the client (e.g.
1513          * from QOS or lprocfs updates). */
1514         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1515         if (rc)
1516                 return rc;
1517
1518         statfs_unpack(sfs, &osfs);
1519
1520         if (sizeof(sfs->f_blocks) == 4) {
1521                 while (osfs.os_blocks > ~0UL) {
1522                         sfs->f_bsize <<= 1;
1523
1524                         osfs.os_blocks >>= 1;
1525                         osfs.os_bfree >>= 1;
1526                         osfs.os_bavail >>= 1;
1527                 }
1528         }
1529
1530         sfs->f_blocks = osfs.os_blocks;
1531         sfs->f_bfree = osfs.os_bfree;
1532         sfs->f_bavail = osfs.os_bavail;
1533
1534         return 0;
1535 }
1536
1537
1538 /********************************
1539  * remote acl                   *
1540  ********************************/
1541
1542 static struct remote_acl *remote_acl_alloc(void)
1543 {
1544         struct remote_acl *racl;
1545         int i;
1546
1547         OBD_ALLOC(racl, sizeof(*racl));
1548         if (!racl)
1549                 return NULL;
1550
1551         spin_lock_init(&racl->ra_lock);
1552         init_MUTEX(&racl->ra_update_sem);
1553
1554         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1555                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1556
1557         return racl;
1558 }
1559
1560 /*
1561  * caller should guarantee no race here.
1562  */
1563 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1564 {
1565         struct remote_perm_setxid *xperm;
1566
1567         while (!list_empty(&perm->lrp_setxid_perms)) {
1568                 xperm = list_entry(perm->lrp_setxid_perms.next,
1569                                    struct remote_perm_setxid,
1570                                    list);
1571                 list_del(&xperm->list);
1572                 OBD_FREE(xperm, sizeof(*xperm));
1573         }
1574 }
1575
1576 /*
1577  * caller should guarantee no race here.
1578  */
1579 static void remote_acl_flush(struct remote_acl *racl)
1580 {
1581         struct list_head *head;
1582         struct lustre_remote_perm *perm, *tmp;
1583         int i;
1584
1585         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1586                 head = &racl->ra_perm_cache[i];
1587
1588                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1589                         remote_perm_flush_xperms(perm);
1590                         list_del(&perm->lrp_list);
1591                         OBD_FREE(perm, sizeof(*perm));
1592                 }
1593         }
1594 }
1595
1596 static void remote_acl_free(struct remote_acl *racl)
1597 {
1598         if (!racl)
1599                 return;
1600
1601         down(&racl->ra_update_sem);
1602         spin_lock(&racl->ra_lock);
1603         remote_acl_flush(racl);
1604         spin_unlock(&racl->ra_lock);
1605         up(&racl->ra_update_sem);
1606
1607         OBD_FREE(racl, sizeof(*racl));
1608 }
1609
1610 static inline int remote_acl_hashfunc(__u32 id)
1611 {
1612         return (id & (REMOTE_ACL_HASHSIZE - 1));
1613 }
1614
1615 static
1616 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1617 {
1618         struct list_head *head;
1619         struct lustre_remote_perm *lperm;
1620         struct remote_perm_setxid *xperm;
1621         int found = 0, rc = -ENOENT;
1622
1623         LASSERT(racl);
1624         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1625         spin_lock(&racl->ra_lock);
1626
1627         list_for_each_entry(lperm, head, lrp_list) {
1628                 if (lperm->lrp_auth_uid == current->uid) {
1629                         found = 1;
1630                         break;
1631                 }
1632         }
1633
1634         if (!found)
1635                 goto out;
1636
1637         if (lperm->lrp_auth_uid == current->fsuid &&
1638             lperm->lrp_auth_gid == current->fsgid) {
1639                 if (lperm->lrp_valid) {
1640                         *perm = lperm->lrp_perm;
1641                         rc = 0;
1642                 }
1643                 goto out;
1644         } else if ((!lperm->lrp_setuid &&
1645                     lperm->lrp_auth_uid != current->fsuid) ||
1646                    (!lperm->lrp_setgid &&
1647                     lperm->lrp_auth_gid != current->fsgid))  {
1648                 *perm = 0;
1649                 rc = 0;
1650                 goto out;
1651         }
1652
1653         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1654                 if (xperm->uid == current->fsuid &&
1655                     xperm->gid == current->fsgid) {
1656                         *perm = xperm->perm;
1657                         rc = 0;
1658                         goto out;
1659                 }
1660         }
1661
1662 out:
1663         spin_unlock(&racl->ra_lock);
1664         return rc;
1665 }
1666
1667 static
1668 int __remote_acl_update(struct remote_acl *racl,
1669                         struct mds_remote_perm *mperm,
1670                         struct lustre_remote_perm *lperm,
1671                         struct remote_perm_setxid *xperm)
1672 {
1673         struct list_head *head;
1674         struct lustre_remote_perm *lp;
1675         struct remote_perm_setxid *xp;
1676         int found = 0, setuid = 0, setgid = 0;
1677
1678         LASSERT(racl);
1679         LASSERT(mperm);
1680         LASSERT(lperm);
1681         LASSERT(current->uid == mperm->mrp_auth_uid);
1682
1683         if (current->fsuid != mperm->mrp_auth_uid)
1684                 setuid = 1;
1685         if (current->fsgid != mperm->mrp_auth_gid)
1686                 setgid = 1;
1687
1688         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1689         spin_lock(&racl->ra_lock);
1690
1691         list_for_each_entry(lp, head, lrp_list) {
1692                 if (lp->lrp_auth_uid == current->uid) {
1693                         found = 1;
1694                         break;
1695                 }
1696         }
1697
1698         if (found) {
1699                 OBD_FREE(lperm, sizeof(*lperm));
1700
1701                 if (!lp->lrp_valid && !setuid && !setgid) {
1702                         lp->lrp_perm = mperm->mrp_perm;
1703                         lp->lrp_valid = 1;
1704                 }
1705
1706                 /* sanity check for changes of setxid rules */
1707                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1708                         CWARN("setuid changes: %d => %d\n",
1709                               (lp->lrp_setuid != 0),
1710                               (mperm->mrp_allow_setuid != 0));
1711                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1712                 }
1713
1714                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1715                         CWARN("setgid changes: %d => %d\n",
1716                               (lp->lrp_setgid != 0),
1717                               (mperm->mrp_allow_setgid != 0));
1718                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1719                 }
1720
1721                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1722                     !list_empty(&lp->lrp_setxid_perms)) {
1723                         remote_perm_flush_xperms(lp);
1724                 }
1725         } else {
1726                 /* initialize lperm and linked into hashtable
1727                  */
1728                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1729                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1730                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1731                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1732                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1733                 list_add(&lperm->lrp_list, head);
1734
1735                 if (!setuid && !setgid) {
1736                         /* in this case, i'm the authenticated user,
1737                          * and mrp_perm is for me.
1738                          */
1739                         lperm->lrp_perm = mperm->mrp_perm;
1740                         lperm->lrp_valid = 1;
1741                         spin_unlock(&racl->ra_lock);
1742
1743                         if (xperm)
1744                                 OBD_FREE(xperm, sizeof(*xperm));
1745                         return 0;
1746                 }
1747
1748                 lp = lperm;
1749                 /* fall through */
1750         }
1751
1752         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1753                 list_empty(&lp->lrp_setxid_perms));
1754
1755         /* if no xperm supplied, we are all done here */
1756         if (!xperm) {
1757                 spin_unlock(&racl->ra_lock);
1758                 return 0;
1759         }
1760
1761         /* whether we allow setuid/setgid */
1762         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1763                 OBD_FREE(xperm, sizeof(*xperm));
1764                 spin_unlock(&racl->ra_lock);
1765                 return 0;
1766         }
1767
1768         /* traverse xperm list */
1769         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1770                 if (xp->uid == current->fsuid &&
1771                     xp->gid == current->fsgid) {
1772                         if (xp->perm != mperm->mrp_perm) {
1773                                 /* actually this should not happen */
1774                                 CWARN("perm changed: %o => %o\n",
1775                                       xp->perm, mperm->mrp_perm);
1776                                 xp->perm = mperm->mrp_perm;
1777                         }
1778                         OBD_FREE(xperm, sizeof(*xperm));
1779                         spin_unlock(&racl->ra_lock);
1780                         return 0;
1781                 }
1782         }
1783
1784         /* finally insert this xperm */
1785         xperm->uid = current->fsuid;
1786         xperm->gid = current->fsgid;
1787         xperm->perm = mperm->mrp_perm;
1788         list_add(&xperm->list, &lp->lrp_setxid_perms);
1789
1790         spin_unlock(&racl->ra_lock);
1791         return 0;
1792 }
1793
1794 /*
1795  * remote_acl semaphore must be held by caller
1796  */
1797 static
1798 int remote_acl_update_locked(struct remote_acl *racl,
1799                              struct mds_remote_perm *mperm)
1800 {
1801         struct lustre_remote_perm *lperm;
1802         struct remote_perm_setxid *xperm;
1803         int setuid = 0, setgid = 0;
1804
1805         might_sleep();
1806
1807         if (current->uid != mperm->mrp_auth_uid) {
1808                 CERROR("current uid %u while authenticated as %u\n",
1809                        current->uid, mperm->mrp_auth_uid);
1810                 return -EINVAL;
1811         }
1812
1813         if (current->fsuid != mperm->mrp_auth_uid)
1814                 setuid = 1;
1815         if (current->fsgid == mperm->mrp_auth_gid)
1816                 setgid = 1;
1817
1818         OBD_ALLOC(lperm, sizeof(*lperm));
1819         if (!lperm)
1820                 return -ENOMEM;
1821
1822         if ((setuid || setgid) &&
1823             !(setuid && !mperm->mrp_allow_setuid) &&
1824             !(setgid && !mperm->mrp_allow_setgid)) {
1825                 OBD_ALLOC(xperm, sizeof(*xperm));
1826                 if (!xperm) {
1827                         OBD_FREE(lperm, sizeof(*lperm));
1828                         return -ENOMEM;
1829                 }
1830         } else
1831                 xperm = NULL;
1832
1833         return __remote_acl_update(racl, mperm, lperm, xperm);
1834 }
1835
1836 /*
1837  * return -EACCES at any error cases
1838  */
1839 int ll_remote_acl_permission(struct inode *inode, int mode)
1840 {
1841         struct ll_sb_info *sbi = ll_i2sbi(inode);
1842         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1843         struct ptlrpc_request *req = NULL;
1844         struct lustre_id id;
1845         struct mds_remote_perm *mperm;
1846         int rc = -EACCES, perm;
1847
1848         if (!racl)
1849                 return -EACCES;
1850
1851         if (__remote_acl_check(racl, &perm) == 0) {
1852                 return ((perm & mode) == mode ? 0 : -EACCES);
1853         }
1854
1855         might_sleep();
1856
1857         /* doing update
1858          */
1859         down(&racl->ra_update_sem);
1860
1861         /* we might lose the race when obtain semaphore,
1862          * so check again.
1863          */
1864         if (__remote_acl_check(racl, &perm) == 0) {
1865                 if ((perm & mode) == mode)
1866                         rc = 0;
1867                 goto out;
1868         }
1869
1870         /* really fetch from mds
1871          */
1872         ll_inode2id(&id, inode);
1873         if (md_access_check(sbi->ll_md_exp, &id, &req))
1874                 goto out;
1875
1876         /* status non-zero indicate there's more apparent error
1877          * detected by mds, e.g. didn't allow this user at all.
1878          * we simply ignore and didn't cache it.
1879          */
1880         if (req->rq_repmsg->status)
1881                 goto out;
1882
1883         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1884                                    lustre_swab_remote_perm);
1885         LASSERT(mperm);
1886         LASSERT_REPSWABBED(req, 1);
1887
1888         if ((mperm->mrp_perm & mode) == mode)
1889                 rc = 0;
1890
1891         remote_acl_update_locked(racl, mperm);
1892 out:
1893         if (req)
1894                 ptlrpc_req_finished(req);
1895
1896         up(&racl->ra_update_sem);
1897         return rc;
1898 }
1899
1900 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1901 {
1902         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1903         int rc;
1904
1905         LASSERT(perm);
1906
1907         if (!racl)
1908                 return -EACCES;
1909
1910         down(&racl->ra_update_sem);
1911         rc = remote_acl_update_locked(racl, perm);
1912         up(&racl->ra_update_sem);
1913
1914         return rc;
1915 }
1916
1917 void ll_inode_invalidate_acl(struct inode *inode)
1918 {
1919         struct ll_sb_info *sbi = ll_i2sbi(inode);
1920         struct ll_inode_info *lli = ll_i2info(inode);
1921
1922         if (sbi->ll_remote) {
1923                 struct remote_acl *racl = lli->lli_remote_acl;
1924
1925                 LASSERT(!lli->lli_posix_acl);
1926                 if (racl) {
1927                         down(&racl->ra_update_sem);
1928                         spin_lock(&racl->ra_lock);
1929                         remote_acl_flush(lli->lli_remote_acl);
1930                         spin_unlock(&racl->ra_lock);
1931                         up(&racl->ra_update_sem);
1932                 }
1933         } else {
1934                 /* we can't invalide acl here: suppose we touch a new file
1935                  * under a dir, blocking ast on dir will lead to open failure
1936                  * on client, although succeed on mds. it's kind of weird,
1937                  * the real fix i think is improve client-vfs interaction.
1938                  *
1939                  * currently we just do nothing here.
1940                  */
1941                 return;
1942
1943                 LASSERT(!lli->lli_remote_acl);
1944                 spin_lock(&lli->lli_lock);
1945                 posix_acl_release(lli->lli_posix_acl);
1946                 lli->lli_posix_acl = NULL;
1947                 spin_unlock(&lli->lli_lock);
1948         }
1949 }
1950
1951 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1952 {
1953         struct ll_inode_info *lli = ll_i2info(inode);
1954         struct lov_stripe_md *lsm = md->lsm;
1955         struct mds_body *body = md->body;
1956         struct mea *mea = md->mea;
1957         struct posix_acl *posix_acl = md->posix_acl;
1958         struct ll_sb_info *sbi = ll_i2sbi(inode);
1959         struct lustre_key *mkey = md->key;
1960         ENTRY;
1961
1962         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1963
1964         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1965                 /* check for default striping info for dir. */
1966                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1967         }
1968
1969         if (lsm != NULL) {
1970                 LASSERT(lsm->lsm_object_gr > 0);
1971                 if (lli->lli_smd == NULL) {
1972                         lli->lli_smd = lsm;
1973                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1974                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1975                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1976                 } else {
1977                         int i;
1978                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1979                                 CERROR("lsm mismatch for inode %ld\n",
1980                                        inode->i_ino);
1981                                 CERROR("lli_smd:\n");
1982                                 dump_lsm(D_ERROR, lli->lli_smd);
1983                                 CERROR("lsm:\n");
1984                                 dump_lsm(D_ERROR, lsm);
1985                                 LBUG();
1986                         }
1987                         /* XXX FIXME -- We should decide on a safer (atomic) and
1988                          * more elegant way to update the lsm */
1989                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1990                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1991                                         lsm->lsm_oinfo[i].loi_id;
1992                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1993                                         lsm->lsm_oinfo[i].loi_gr;
1994                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1995                                         lsm->lsm_oinfo[i].loi_ost_idx;
1996                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1997                                         lsm->lsm_oinfo[i].loi_ost_gen;
1998                         }
1999                 }
2000                 /* bug 2844 - limit i_blksize for broken user-space apps */
2001                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
2002                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
2003                 if (lli->lli_smd != lsm)
2004                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
2005         }
2006
2007         if (mea != NULL) {
2008                 if (lli->lli_mea == NULL) {
2009                         lli->lli_mea = mea;
2010                 } else {
2011                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
2012                                 CERROR("mea mismatch for inode %lu\n",
2013                                         inode->i_ino);
2014                                 LBUG();
2015                         }
2016                 }
2017                 if (lli->lli_mea != mea)
2018                         obd_free_memmd(ll_i2mdexp(inode),
2019                                        (struct lov_stripe_md **) &mea);
2020         }
2021
2022         if (body->valid & OBD_MD_FID)
2023                 id_assign_fid(&lli->lli_id, &body->id1);
2024
2025         if (body->valid & OBD_MD_FLID)
2026                 id_ino(&lli->lli_id) = id_ino(&body->id1);
2027
2028         if (body->valid & OBD_MD_FLGENER)
2029                 id_gen(&lli->lli_id) = id_gen(&body->id1);
2030
2031         /* local/remote ACL */
2032         if (sbi->ll_remote) {
2033                 LASSERT(md->posix_acl == NULL);
2034                 if (md->remote_perm) {
2035                         ll_remote_acl_update(inode, md->remote_perm);
2036                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2037                         md->remote_perm = NULL;
2038                 }
2039         } else {
2040                 LASSERT(md->remote_perm == NULL);
2041                 spin_lock(&lli->lli_lock);
2042                 if (posix_acl != NULL) {
2043                         if (lli->lli_posix_acl != NULL)
2044                                 posix_acl_release(lli->lli_posix_acl);
2045                         lli->lli_posix_acl = posix_acl;
2046                 }
2047                 spin_unlock(&lli->lli_lock);
2048         }
2049
2050         if (body->valid & OBD_MD_FLID)
2051                 inode->i_ino = id_ino(&body->id1);
2052         if (body->valid & OBD_MD_FLGENER)
2053                 inode->i_generation = id_gen(&body->id1);
2054         if (body->valid & OBD_MD_FLATIME)
2055                 LTIME_S(inode->i_atime) = body->atime;
2056         if (body->valid & OBD_MD_FLMTIME &&
2057             body->mtime > LTIME_S(inode->i_mtime)) {
2058                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2059                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2060                 LTIME_S(inode->i_mtime) = body->mtime;
2061         }
2062         if (body->valid & OBD_MD_FLCTIME &&
2063             body->ctime > LTIME_S(inode->i_ctime))
2064                 LTIME_S(inode->i_ctime) = body->ctime;
2065         if (body->valid & OBD_MD_FLMODE) {
2066                 inode->i_mode = (inode->i_mode & S_IFMT) |
2067                         (body->mode & ~S_IFMT);
2068         }
2069         if (body->valid & OBD_MD_FLTYPE) {
2070                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2071                         (body->mode & S_IFMT);
2072         }
2073         if (body->valid & OBD_MD_FLUID)
2074                 inode->i_uid = body->uid;
2075         if (body->valid & OBD_MD_FLGID)
2076                 inode->i_gid = body->gid;
2077         if (body->valid & OBD_MD_FLFLAGS)
2078                 inode->i_flags = body->flags;
2079         if (body->valid & OBD_MD_FLNLINK)
2080                 inode->i_nlink = body->nlink;
2081         if (body->valid & OBD_MD_FLRDEV)
2082 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2083                 inode->i_rdev = body->rdev;
2084 #else
2085                 inode->i_rdev = old_decode_dev(body->rdev);
2086 #endif
2087         if (body->valid & OBD_MD_FLSIZE)
2088                 inode->i_size = body->size;
2089         if (body->valid & OBD_MD_FLBLOCKS)
2090                 inode->i_blocks = body->blocks;
2091
2092         if (body->valid & OBD_MD_FLSIZE)
2093                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2094
2095         if (body->valid & OBD_MD_FLAUDIT) {
2096                 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2097                 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2098                         sbi->ll_audit_mask = body->audit;
2099                 else
2100                         lli->lli_audit_mask = body->audit;
2101         }
2102
2103         if (mkey != NULL) {
2104                 LASSERT(body->valid & OBD_MD_FLKEY);
2105                 ll_crypto_init_inode_key(inode, mkey);
2106         }
2107
2108 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2109         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2110 #endif
2111         LASSERT(id_fid(&lli->lli_id) != 0);
2112 }
2113
2114 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2115 static struct backing_dev_info ll_backing_dev_info = {
2116         .ra_pages       = 0,    /* No readahead */
2117         .memory_backed  = 0,    /* Does contribute to dirty memory */
2118 };
2119 #endif
2120
2121 void ll_read_inode2(struct inode *inode, void *opaque)
2122 {
2123         struct lustre_md *md = opaque;
2124         struct ll_inode_info *lli = ll_i2info(inode);
2125         ENTRY;
2126
2127         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2128                inode->i_generation, inode);
2129
2130         ll_lli_init(lli);
2131
2132         LASSERT(!lli->lli_smd);
2133
2134         if (ll_i2sbi(inode)->ll_remote) {
2135                 lli->lli_remote_acl = remote_acl_alloc();
2136                 /* if failed alloc, nobody will be able to access this inode */
2137         }
2138
2139         /* Core attributes from the MDS first.  This is a new inode, and
2140          * the VFS doesn't zero times in the core inode so we have to do
2141          * it ourselves.  They will be overwritten by either MDS or OST
2142          * attributes - we just need to make sure they aren't newer. */
2143         LTIME_S(inode->i_mtime) = 0;
2144         LTIME_S(inode->i_atime) = 0;
2145         LTIME_S(inode->i_ctime) = 0;
2146
2147         inode->i_rdev = 0;
2148         ll_update_inode(inode, md);
2149
2150         /* OIDEBUG(inode); */
2151
2152         if (S_ISREG(inode->i_mode)) {
2153                 inode->i_op = &ll_file_inode_operations;
2154                 inode->i_fop = &ll_file_operations;
2155                 inode->i_mapping->a_ops = &ll_aops;
2156                 EXIT;
2157         } else if (S_ISDIR(inode->i_mode)) {
2158                 inode->i_op = &ll_dir_inode_operations;
2159                 inode->i_fop = &ll_dir_operations;
2160                 inode->i_mapping->a_ops = &ll_dir_aops;
2161                 EXIT;
2162         } else if (S_ISLNK(inode->i_mode)) {
2163                 inode->i_op = &ll_fast_symlink_inode_operations;
2164                 EXIT;
2165         } else {
2166                 inode->i_op = &ll_special_inode_operations;
2167
2168 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2169                 init_special_inode(inode, inode->i_mode,
2170                                    kdev_t_to_nr(inode->i_rdev));
2171
2172                 /* initializing backing dev info. */
2173                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2174 #else
2175                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2176 #endif
2177                 lli->ll_save_ifop = inode->i_fop;
2178
2179                 if (S_ISCHR(inode->i_mode))
2180                         inode->i_fop = &ll_special_chr_inode_fops;
2181                 else if (S_ISBLK(inode->i_mode))
2182                         inode->i_fop = &ll_special_blk_inode_fops;
2183                 else if (S_ISFIFO(inode->i_mode))
2184                         inode->i_fop = &ll_special_fifo_inode_fops;
2185                 else if (S_ISSOCK(inode->i_mode))
2186                         inode->i_fop = &ll_special_sock_inode_fops;
2187
2188                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2189                       inode->i_fop);
2190
2191                 if (lli->ll_save_ifop->owner) {
2192                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2193                               lli->ll_save_ifop->owner);
2194                 }
2195                 EXIT;
2196         }
2197 }
2198
2199 void ll_delete_inode(struct inode *inode)
2200 {
2201         struct ll_sb_info *sbi = ll_i2sbi(inode);
2202         struct lustre_id id;
2203         int rc;
2204         ENTRY;
2205
2206         ll_inode2id(&id, inode);
2207
2208         rc = md_delete_inode(sbi->ll_md_exp, &id);
2209         if (rc) {
2210                 CERROR("md_delete_inode() failed, error %d\n",
2211                        rc);
2212         }
2213
2214         clear_inode(inode);
2215         EXIT;
2216 }
2217
2218 int ll_iocontrol(struct inode *inode, struct file *file,
2219                  unsigned int cmd, unsigned long arg)
2220 {
2221         struct ll_sb_info *sbi = ll_i2sbi(inode);
2222         struct ptlrpc_request *req = NULL;
2223         int rc, flags = 0;
2224         ENTRY;
2225
2226         switch(cmd) {
2227         case EXT3_IOC_GETFLAGS: {
2228                 struct lustre_id id;
2229                 __u64 valid = OBD_MD_FLFLAGS;
2230                 struct mds_body *body;
2231
2232                 ll_inode2id(&id, inode);
2233                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2234                                 0, 0, NULL, &req);
2235                 if (rc) {
2236                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2237                         RETURN(-abs(rc));
2238                 }
2239
2240                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2241
2242                 if (body->flags & S_APPEND)
2243                         flags |= EXT3_APPEND_FL;
2244                 if (body->flags & S_IMMUTABLE)
2245                         flags |= EXT3_IMMUTABLE_FL;
2246                 if (body->flags & S_NOATIME)
2247                         flags |= EXT3_NOATIME_FL;
2248
2249                 ptlrpc_req_finished (req);
2250
2251                 RETURN(put_user(flags, (int *)arg));
2252         }
2253         case EXT3_IOC_SETFLAGS: {
2254                 struct mdc_op_data *op_data;
2255                 struct iattr attr;
2256                 struct obdo *oa;
2257                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2258
2259                 if (get_user(flags, (int *)arg))
2260                         RETURN(-EFAULT);
2261
2262                 oa = obdo_alloc();
2263                 if (!oa)
2264                         RETURN(-ENOMEM);
2265
2266                 OBD_ALLOC(op_data, sizeof(*op_data));
2267                 if (op_data == NULL) {
2268                         obdo_free(oa);
2269                         RETURN(-ENOMEM);
2270                 }
2271                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
2272
2273                 memset(&attr, 0x0, sizeof(attr));
2274                 attr.ia_attr_flags = flags;
2275                 attr.ia_valid |= ATTR_ATTR_FLAG;
2276
2277                 rc = md_setattr(sbi->ll_md_exp, op_data,
2278                                 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2279                 OBD_FREE(op_data, sizeof(*op_data));
2280                 if (rc) {
2281                         ptlrpc_req_finished(req);
2282                         if (rc != -EPERM && rc != -EACCES)
2283                                 CERROR("md_setattr fails: rc = %d\n", rc);
2284                         obdo_free(oa);
2285                         RETURN(rc);
2286                 }
2287                 ptlrpc_req_finished(req);
2288
2289                 oa->o_id = lsm->lsm_object_id;
2290                 oa->o_gr = lsm->lsm_object_gr;
2291                 oa->o_flags = flags;
2292                 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2293                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2294                               | OBD_MD_FLIFID;
2295
2296                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL);
2297                 obdo_free(oa);
2298                 if (rc) {
2299                         if (rc != -EPERM && rc != -EACCES)
2300                                 CERROR("md_setattr fails: rc = %d\n", rc);
2301                         RETURN(rc);
2302                 }
2303
2304                 if (flags & EXT3_APPEND_FL)
2305                         inode->i_flags |= S_APPEND;
2306                 else
2307                         inode->i_flags &= ~S_APPEND;
2308                 if (flags & EXT3_IMMUTABLE_FL)
2309                         inode->i_flags |= S_IMMUTABLE;
2310                 else
2311                         inode->i_flags &= ~S_IMMUTABLE;
2312                 if (flags & EXT3_NOATIME_FL)
2313                         inode->i_flags |= S_NOATIME;
2314                 else
2315                         inode->i_flags &= ~S_NOATIME;
2316
2317                 RETURN(0);
2318         }
2319         default:
2320                 RETURN(-ENOSYS);
2321         }
2322
2323         RETURN(0);
2324 }
2325
2326 /* this is only called in the case of forced umount. */
2327 void ll_umount_begin(struct super_block *sb)
2328 {
2329         struct ll_sb_info *sbi = ll_s2sbi(sb);
2330         struct obd_ioctl_data ioc_data = { 0 };
2331         struct obd_device *obd;
2332         ENTRY;
2333
2334         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2335                sb->s_count, atomic_read(&sb->s_active));
2336
2337         obd = class_exp2obd(sbi->ll_md_exp);
2338         if (obd == NULL) {
2339                 CERROR("Invalid MDC connection handle "LPX64"\n",
2340                        sbi->ll_md_exp->exp_handle.h_cookie);
2341                 EXIT;
2342                 return;
2343         }
2344         obd->obd_no_recov = 1;
2345         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2346                       sizeof(ioc_data), &ioc_data, NULL);
2347
2348         obd = class_exp2obd(sbi->ll_dt_exp);
2349         if (obd == NULL) {
2350                 CERROR("Invalid LOV connection handle "LPX64"\n",
2351                        sbi->ll_dt_exp->exp_handle.h_cookie);
2352                 EXIT;
2353                 return;
2354         }
2355
2356         obd->obd_no_recov = 1;
2357         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2358                       sizeof(ioc_data), &ioc_data, NULL);
2359
2360         /*
2361          * really, we'd like to wait until there are no requests outstanding,
2362          * and then continue.  For now, we just invalidate the requests,
2363          * schedule, and hope.
2364          */
2365         schedule();
2366
2367         EXIT;
2368 }
2369
2370 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2371                   struct inode **inode, struct ptlrpc_request *req,
2372                   int offset, struct super_block *sb)
2373 {
2374         struct lustre_md md;
2375         int rc = 0;
2376
2377         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2378         if (rc)
2379                 RETURN(rc);
2380
2381         if (*inode) {
2382                 ll_update_inode(*inode, &md);
2383         } else {
2384                 LASSERT(sb);
2385                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2386                 if (*inode == NULL || is_bad_inode(*inode)) {
2387                         /* free the lsm if we allocated one above */
2388                         if (md.lsm != NULL)
2389                                 obd_free_memmd(dt_exp, &md.lsm);
2390                         if (md.mea != NULL)
2391                                 obd_free_memmd(md_exp,
2392                                                (struct lov_stripe_md**)&md.mea);
2393                         rc = -ENOMEM;
2394                         CERROR("new_inode -fatal: rc %d\n", rc);
2395                 }
2396         }
2397
2398         RETURN(rc);
2399 }
2400
2401 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2402 {
2403         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2404         struct lustre_mount_data *lmd = sbi->ll_lmd;
2405
2406         if (lmd) {
2407                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2408                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2409         }
2410         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2411         if (sbi->ll_remote && lmd)
2412                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2413
2414         if (lmd && lmd->lmd_pag)
2415                 seq_printf(m, ",pag");
2416
2417         return 0;
2418 }
2419
2420 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2421                char *filename, struct lustre_id *ret)
2422 {
2423         struct ptlrpc_request *request = NULL;
2424         struct mds_body *body;
2425         int rc;
2426
2427         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2428                              OBD_MD_FID, 0, &request);
2429         if (rc < 0) {
2430                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2431                        filename, rc);
2432                 return rc;
2433         }
2434
2435         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2436         LASSERT(body != NULL);
2437         LASSERT_REPSWABBED(request, 0);
2438
2439         *ret = body->id1;
2440         ptlrpc_req_finished(request);
2441
2442         return rc;
2443 }
2444 int ll_flush_cred(struct inode *inode)
2445 {
2446         struct ll_sb_info *sbi = ll_i2sbi(inode);
2447         int rc = 0;
2448
2449         /* XXX to avoid adding api, we simply use set_info() interface
2450          * to notify underlying obds. set_info() is more like a ioctl() now...
2451          */
2452         if (sbi->ll_md_exp) {
2453                 rc = obd_set_info(sbi->ll_md_exp,
2454                                   strlen("flush_cred"), "flush_cred",
2455                                   0, NULL);
2456                 if (rc)
2457                         return rc;
2458         }
2459
2460         if (sbi->ll_dt_exp) {
2461                 rc = obd_set_info(sbi->ll_dt_exp,
2462                                   strlen("flush_cred"), "flush_cred",
2463                                   0, NULL);
2464                 if (rc)
2465                         return rc;
2466         }
2467
2468         return rc;
2469 }