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 }
 587
 588 int ll_fill_super(struct super_block *sb, void *data, int silent)
 589 {
 590         struct ll_sb_info *sbi;
 591         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 592         char *mds_sec = NULL;
 593         char *oss_sec = NULL;
 594         int async, err;
 595         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 596         __u64 remote_flag = 0;
 597         ENTRY;
 598
 599         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 600
 601         sbi = lustre_init_sbi(sb);
 602         if (!sbi)
 603                 RETURN(-ENOMEM);
 604
 605         sbi->ll_flags |= LL_SBI_READAHEAD;
 606         ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
 607                    &async, &sbi->ll_flags);
 608
 609         if (!lov || !lmv) {
 610                 CERROR("no osc %p or no mdc %p\n", lov, lmv);
 611                 GOTO(out, err = -EINVAL);
 612         }
 613
 614         err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
 615                                        oss_sec, nllu, 0, &remote_flag);
 616         EXIT;
 617 out:
 618         if (err)
 619                 lustre_free_sbi(sb);
 620
 621         if (lmv)
 622                 OBD_FREE(lmv, strlen(lmv) + 1);
 623         if (lov)
 624                 OBD_FREE(lov, strlen(lov) + 1);
 625         if (mds_sec)
 626                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 627         if (oss_sec)
 628                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 629         if (gkc)
 630                 OBD_FREE(gkc, strlen(gkc) + 1);
 631
 632         return err;
 633 } /* ll_read_super */
 634
 635 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 636                               struct config_llog_instance *cfg, int allow_recov)
 637 {
 638         struct lustre_cfg *lcfg = NULL;
 639         struct lustre_cfg_bufs bufs;
 640         struct portals_cfg pcfg;
 641         char *peer = "MDS_PEER_UUID";
 642         struct obd_device *obd;
 643         struct lustre_handle md_conn = {0, };
 644         struct obd_export *exp;
 645         char *name = "mdc_dev";
 646         class_uuid_t uuid;
 647         struct obd_uuid lmv_uuid;
 648         struct llog_ctxt *ctxt;
 649         int rc = 0, err = 0;
 650         ENTRY;
 651
 652         if (lmd_bad_magic(lmd))
 653                 RETURN(-EINVAL);
 654
 655         generate_random_uuid(uuid);
 656         class_uuid_unparse(uuid, &lmv_uuid);
 657
 658         if (lmd->lmd_local_nid) {
 659                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 660                 pcfg.pcfg_nal = lmd->lmd_nal;
 661                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 662                 rc = libcfs_nal_cmd(&pcfg);
 663                 if (rc < 0)
 664                         GOTO(out, rc);
 665         }
 666
 667         if (lmd->lmd_nal == SOCKNAL ||
 668             lmd->lmd_nal == OPENIBNAL ||
 669             lmd->lmd_nal == IIBNAL ||
 670             lmd->lmd_nal == VIBNAL ||
 671             lmd->lmd_nal == RANAL) {
 672                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 673                 pcfg.pcfg_nal     = lmd->lmd_nal;
 674                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 675                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 676                 pcfg.pcfg_misc    = lmd->lmd_port;
 677                 rc = libcfs_nal_cmd(&pcfg);
 678                 if (rc < 0)
 679                         GOTO(out, rc);
 680         }
 681         lustre_cfg_bufs_reset(&bufs, name);
 682         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 683
 684         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 685         lcfg->lcfg_nal = lmd->lmd_nal;
 686         lcfg->lcfg_nid = lmd->lmd_server_nid;
 687         LASSERT(lcfg->lcfg_nal);
 688         LASSERT(lcfg->lcfg_nid);
 689         err = class_process_config(lcfg);
 690         lustre_cfg_free(lcfg);
 691         if (err < 0)
 692                 GOTO(out_del_conn, err);
 693
 694         lustre_cfg_bufs_reset(&bufs, name);
 695         lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
 696         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 697
 698         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 699         err = class_process_config(lcfg);
 700         lustre_cfg_free(lcfg);
 701         if (err < 0)
 702                 GOTO(out_del_uuid, err);
 703
 704         lustre_cfg_bufs_reset(&bufs, name);
 705         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 706         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 707
 708         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 709         err = class_process_config(lcfg);
 710         lustre_cfg_free(lcfg);
 711         if (err < 0)
 712                 GOTO(out_detach, err);
 713
 714         obd = class_name2obd(name);
 715         if (obd == NULL)
 716                 GOTO(out_cleanup, rc = -EINVAL);
 717
 718         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 719                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 720         if (rc)
 721                 GOTO(out_cleanup, rc);
 722
 723         if (lmd->lmd_pag) {
 724                 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
 725                 rc = obd_set_info(obd->obd_self_export,
 726                                   strlen("sec_flags"), "sec_flags",
 727                                   sizeof(sec_flags), &sec_flags);
 728                 if (rc)
 729                         GOTO(out_cleanup, rc);
 730         }
 731
 732         /* Disable initial recovery on this import */
 733         rc = obd_set_info(obd->obd_self_export,
 734                           strlen("initial_recov"), "initial_recov",
 735                           sizeof(allow_recov), &allow_recov);
 736         if (rc)
 737                 GOTO(out_cleanup, rc);
 738
 739         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, OBD_OPT_REAL_CLIENT);
 740         if (rc) {
 741                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 742                 GOTO(out_cleanup, rc);
 743         }
 744
 745         exp = class_conn2export(&md_conn);
 746
 747         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 748         rc = class_config_process_llog(ctxt, profile, cfg);
 749         if (rc)
 750                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 751
 752         err = obd_disconnect(exp, 0);
 753
 754         EXIT;
 755 out_cleanup:
 756         lustre_cfg_bufs_reset(&bufs, name);
 757         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 758         err = class_process_config(lcfg);
 759         lustre_cfg_free(lcfg);
 760         if (err < 0)
 761                 GOTO(out, err);
 762 out_detach:
 763         lustre_cfg_bufs_reset(&bufs, name);
 764         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 765         err = class_process_config(lcfg);
 766         lustre_cfg_free(lcfg);
 767         if (err < 0)
 768                 GOTO(out, err);
 769
 770 out_del_uuid:
 771         lustre_cfg_bufs_reset(&bufs, name);
 772         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 773         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 774         err = class_process_config(lcfg);
 775         lustre_cfg_free(lcfg);
 776
 777 out_del_conn:
 778         if (lmd->lmd_nal == SOCKNAL ||
 779             lmd->lmd_nal == OPENIBNAL ||
 780             lmd->lmd_nal == IIBNAL ||
 781             lmd->lmd_nal == VIBNAL ||
 782             lmd->lmd_nal == RANAL) {
 783                 int err2;
 784
 785                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 786                 pcfg.pcfg_nal     = lmd->lmd_nal;
 787                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 788                 pcfg.pcfg_flags   = 1;          /* single_share */
 789                 err2 = libcfs_nal_cmd(&pcfg);
 790                 if (err2 && !err)
 791                         err = err2;
 792                 if (err < 0)
 793                         GOTO(out, err);
 794         }
 795 out:
 796         if (rc == 0)
 797                 rc = err;
 798
 799         return rc;
 800 }
 801
 802 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 803 {
 804         struct lustre_cfg *lcfg;
 805         struct lustre_cfg_bufs bufs;
 806         struct obd_device *obd;
 807         int next = 0;
 808
 809         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
 810                 int err;
 811
 812                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 813                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 814                 err = class_process_config(lcfg);
 815                 if (err) {
 816                         CERROR("cleanup failed: %s\n", obd->obd_name);
 817                         //continue;
 818                 }
 819
 820                 lcfg->lcfg_command = LCFG_DETACH;
 821                 err = class_process_config(lcfg);
 822                 lustre_cfg_free(lcfg);
 823                 if (err) {
 824                         CERROR("detach failed: %s\n", obd->obd_name);
 825                         //continue;
 826                 }
 827         }
 828
 829         if (sbi->ll_lmd != NULL)
 830                 class_del_profile(sbi->ll_lmd->lmd_profile);
 831 }
 832
 833 static int lustre_process_profile(struct super_block *sb,
 834                                   struct lustre_mount_data *lmd,
 835                                   char **lov, char **lmv, char **gkc)
 836 {
 837         struct ll_sb_info *sbi = ll_s2sbi(sb);
 838         struct config_llog_instance cfg;
 839         struct lustre_profile *lprof;
 840         int len, err = 0;
 841         ENTRY;
 842
 843         if (!lmd->lmd_profile)
 844                 RETURN(0);
 845
 846         if (lmd->lmd_mds[0] == '\0') {
 847                 CERROR("no mds name\n");
 848                 GOTO(out, err = -EINVAL);
 849         }
 850         lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
 851         lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
 852
 853         OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 854         if (sbi->ll_lmd == NULL)
 855                 GOTO(out, err = -ENOMEM);
 856         memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 857
 858         /* generate a string unique to this super, let's try the address of the
 859          * super itself. */
 860         len = (sizeof(sb) * 2) + 1;
 861         OBD_ALLOC(sbi->ll_instance, len);
 862         if (sbi->ll_instance == NULL)
 863                 GOTO(out, err = -ENOMEM);
 864         sprintf(sbi->ll_instance, "%p", sb);
 865
 866         cfg.cfg_instance = sbi->ll_instance;
 867         cfg.cfg_uuid = sbi->ll_sb_uuid;
 868         cfg.cfg_local_nid = lmd->lmd_local_nid;
 869         err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 870         if (err < 0) {
 871                 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 872                 GOTO(out, err);
 873         }
 874
 875         lprof = class_get_profile(lmd->lmd_profile);
 876         if (lprof == NULL) {
 877                 CERROR("No profile found: %s\n", lmd->lmd_profile);
 878                 GOTO(out, err = -EINVAL);
 879         }
 880
 881         OBD_ALLOC(*lov, strlen(lprof->lp_lov) +
 882                   strlen(sbi->ll_instance) + 2);
 883         if (*lov == NULL)
 884                 GOTO(out, err = -ENOMEM);
 885
 886         sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 887
 888         OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
 889                   strlen(sbi->ll_instance) + 2);
 890         if (*lmv == NULL)
 891                 GOTO(out_free_lov, err = -ENOMEM);
 892
 893         sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 894
 895         if (lprof->lp_gkc) {
 896                OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
 897                          strlen(sbi->ll_instance) + 2);
 898                if (*gkc == NULL)
 899                        GOTO(out_free_lmv, err = -ENOMEM);
 900
 901                sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
 902         }
 903
 904         RETURN(err);
 905 out_free_lmv:
 906         OBD_FREE(*lmv, strlen(lprof->lp_lmv) +
 907                  strlen(sbi->ll_instance) + 2);
 908 out_free_lov:
 909         OBD_FREE(*lov, strlen(lprof->lp_lov) +
 910                  strlen(sbi->ll_instance) + 2);
 911 out:
 912         return err;
 913 }
 914
 915 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
 916 {
 917         struct lustre_mount_data *lmd = sbi->ll_lmd;
 918         struct config_llog_instance cfg;
 919         char *cl_prof;
 920         int len, err = 0;
 921         ENTRY;
 922
 923         if (!lmd)
 924                 RETURN(err);
 925
 926         len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
 927
 928         if (force_umount) {
 929                 CERROR("force umount, doing manual cleanup\n");
 930                 lustre_manual_cleanup(sbi);
 931                 GOTO(free_lmd, 0);
 932
 933         }
 934         if (sbi->ll_instance != NULL) {
 935                 cfg.cfg_instance = sbi->ll_instance;
 936                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 937
 938                 OBD_ALLOC(cl_prof, len);
 939                 if (!cl_prof) {
 940                         CERROR("can't allocate memory, "
 941                                "skipping processing cleanup profile.\n");
 942                         GOTO(free_lmd, err = -ENOMEM);
 943                 }
 944
 945                 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 946                 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 947                 if (err < 0) {
 948                         CERROR("Unable to process log: %s\n", cl_prof);
 949                         lustre_manual_cleanup(sbi);
 950                 }
 951                 OBD_FREE(cl_prof, len);
 952         }
 953         EXIT;
 954 free_lmd:
 955         if (sbi->ll_instance)
 956                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 957         OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 958         return err;
 959 }
 960
 961 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 962 {
 963         struct lustre_mount_data * lmd = data;
 964         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 965         struct ll_sb_info *sbi;
 966         int err;
 967         ENTRY;
 968
 969         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 970         if (lmd_bad_magic(lmd))
 971                 RETURN(-EINVAL);
 972
 973         sbi = lustre_init_sbi(sb);
 974         if (!sbi)
 975                 RETURN(-ENOMEM);
 976
 977         sbi->ll_flags |= LL_SBI_READAHEAD;
 978
 979         err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
 980         if (err) {
 981                 CERROR("Can not process the profile err %d \n", err);
 982                 GOTO(out_free, err);
 983         }
 984         if (!lov || !lmv) {
 985                 CERROR("no osc %p or no mdc %p \n", lov, lmv);
 986                 GOTO(out_free, err = -EINVAL);
 987         }
 988
 989         err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
 990                                        lmd->lmd_mds_security,
 991                                        lmd->lmd_oss_security,
 992                                        &lmd->lmd_nllu, lmd->lmd_pag,
 993                                        &lmd->lmd_remote_flag);
 994
 995         if (err)
 996                 GOTO(out_free, err);
 997
 998         EXIT;
 999 out_dev:
1000         if (lmv)
1001                 OBD_FREE(lmv, strlen(lmv) + 1);
1002         if (lov)
1003                 OBD_FREE(lov, strlen(lov) + 1);
1004         if (gkc)
1005                 OBD_FREE(gkc, strlen(gkc) + 1);
1006
1007         return err;
1008 out_free:
1009         lustre_clean_profile(sbi, 0);
1010         lustre_free_sbi(sb);
1011         goto out_dev;
1012
1013 } /* lustre_fill_super */
1014
1015 void lustre_put_super(struct super_block *sb)
1016 {
1017         struct obd_device *obd;
1018         struct ll_sb_info *sbi = ll_s2sbi(sb);
1019         int force_umount = 0;
1020         ENTRY;
1021
1022         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1023         obd = class_exp2obd(sbi->ll_md_exp);
1024         if (obd)
1025                 force_umount = obd->obd_no_recov;
1026         obd = NULL;
1027
1028         lustre_common_put_super(sb);
1029         lustre_clean_profile(sbi, force_umount);
1030         lustre_free_sbi(sb);
1031
1032         EXIT;
1033 } /* lustre_put_super */
1034
1035 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1036 {
1037         struct lustre_mount_data *lmd = sbi->ll_lmd;
1038         char *profile = lmd->lmd_profile, *name = NULL;
1039         struct config_llog_instance cfg;
1040         int rc, namelen =  0, version;
1041         struct llog_ctxt *ctxt;
1042         ENTRY;
1043
1044         if (profile == NULL)
1045                 RETURN(0);
1046         if (lmd == NULL) {
1047                 CERROR("Client not mounted with zero-conf; cannot "
1048                        "process update log.\n");
1049                 RETURN(0);
1050         }
1051
1052         cfg.cfg_instance = sbi->ll_instance;
1053         cfg.cfg_uuid = sbi->ll_sb_uuid;
1054         cfg.cfg_local_nid = lmd->lmd_local_nid;
1055
1056         namelen = strlen(profile) + 20; /* -clean-######### */
1057         OBD_ALLOC(name, namelen);
1058         if (name == NULL)
1059                 RETURN(-ENOMEM);
1060
1061         if (clean) {
1062                 version = sbi->ll_config_version - 1;
1063                 sprintf(name, "%s-clean-%d", profile, version);
1064         } else {
1065                 version = sbi->ll_config_version + 1;
1066                 sprintf(name, "%s-%d", profile, version);
1067         }
1068
1069         CWARN("Applying configuration log %s\n", name);
1070
1071         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1072                                 LLOG_CONFIG_REPL_CTXT);
1073         rc = class_config_process_llog(ctxt, name, &cfg);
1074         if (rc == 0)
1075                 sbi->ll_config_version = version;
1076         CWARN("Finished applying configuration log %s: %d\n", name, rc);
1077
1078         if (rc == 0 && clean == 0) {
1079                 struct lov_desc desc;
1080                 __u32 valsize;
1081                 int rc = 0;
1082
1083                 valsize = sizeof(desc);
1084                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1085                                   "lovdesc", &valsize, &desc);
1086
1087                 rc = obd_init_ea_size(sbi->ll_md_exp,
1088                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
1089                                       (desc.ld_tgt_count *
1090                                        sizeof(struct llog_cookie)));
1091         }
1092         OBD_FREE(name, namelen);
1093         RETURN(rc);
1094 }
1095
1096 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1097 {
1098         struct inode *inode = NULL;
1099
1100         /* NOTE: we depend on atomic igrab() -bzzz */
1101         lock_res_and_lock(lock);
1102         if (lock->l_ast_data) {
1103                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1104                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1105                         inode = igrab(lock->l_ast_data);
1106                 } else {
1107                         struct timeval now;
1108                         do_gettimeofday(&now);
1109                         inode = lock->l_ast_data;
1110                         LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1111                                    lock->l_enqueued_time.tv_sec,
1112                                    lock->l_enqueued_time.tv_usec,
1113                                    now.tv_sec, now.tv_usec);
1114                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1115                                "l_ast_data %p is bogus: magic %0x8\n",
1116                                lock->l_ast_data, lli->lli_inode_magic);
1117                         CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1118                                inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1119                         inode = NULL;
1120                         unlock_res_and_lock(lock);
1121                         LBUG();
1122                 }
1123         }
1124         unlock_res_and_lock(lock);
1125         return inode;
1126 }
1127
1128 int null_if_equal(struct ldlm_lock *lock, void *data)
1129 {
1130         if (data == lock->l_ast_data) {
1131                 lock->l_ast_data = NULL;
1132
1133                 if (lock->l_req_mode != lock->l_granted_mode)
1134                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1135         }
1136
1137         return LDLM_ITER_CONTINUE;
1138 }
1139
1140 static void remote_acl_free(struct remote_acl *racl);
1141
1142 void ll_clear_inode(struct inode *inode)
1143 {
1144         struct lustre_id id;
1145         struct ll_inode_info *lli = ll_i2info(inode);
1146         struct ll_sb_info *sbi = ll_i2sbi(inode);
1147         ENTRY;
1148
1149         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1150                inode->i_generation, inode);
1151
1152         LASSERT(ll_is_inode_dirty(inode) == 0);
1153         ll_inode2id(&id, inode);
1154
1155         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1156         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1157
1158         LASSERT(!lli->lli_open_fd_write_count);
1159         LASSERT(!lli->lli_open_fd_read_count);
1160         LASSERT(!lli->lli_open_fd_exec_count);
1161         if (lli->lli_mds_write_och)
1162                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1163         if (lli->lli_mds_exec_och)
1164                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1165         if (lli->lli_mds_read_och)
1166                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1167         if (lli->lli_smd)
1168                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1169                                   null_if_equal, inode);
1170
1171         if (lli->lli_smd) {
1172                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1173                 lli->lli_smd = NULL;
1174         }
1175
1176         if (lli->lli_mea) {
1177                 obd_free_memmd(sbi->ll_md_exp,
1178                                (struct lov_stripe_md **) &lli->lli_mea);
1179                 lli->lli_mea = NULL;
1180         }
1181         ll_crypto_destroy_inode_key(inode);
1182         if (lli->lli_symlink_name) {
1183                 OBD_FREE(lli->lli_symlink_name,
1184                          strlen(lli->lli_symlink_name) + 1);
1185                 lli->lli_symlink_name = NULL;
1186         }
1187
1188         if (lli->lli_posix_acl) {
1189                 LASSERT(lli->lli_remote_acl == NULL);
1190                 posix_acl_release(lli->lli_posix_acl);
1191                 lli->lli_posix_acl = NULL;
1192         }
1193
1194         if (lli->lli_remote_acl) {
1195                 LASSERT(lli->lli_posix_acl == NULL);
1196                 remote_acl_free(lli->lli_remote_acl);
1197                 lli->lli_remote_acl = NULL;
1198         }
1199
1200         if (lli->lli_trunc_capa) {
1201                 OBD_FREE(lli->lli_trunc_capa, sizeof(struct lustre_capa));
1202                 lli->lli_trunc_capa = NULL;
1203         }
1204
1205         lli->lli_inode_magic = LLI_INODE_DEAD;
1206         EXIT;
1207 }
1208
1209 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1210  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1211  * keep these values until such a time that objects are allocated for it.
1212  * We do the MDS operations first, as it is checking permissions for us.
1213  * We don't to the MDS RPC if there is nothing that we want to store there,
1214  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1215  * going to do an RPC anyways.
1216  *
1217  * If we are doing a truncate, we will send the mtime and ctime updates
1218  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1219  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1220  * at the same time.
1221  */
1222 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1223 {
1224         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1225         struct ll_inode_info *lli = ll_i2info(inode);
1226         struct ll_sb_info *sbi = ll_i2sbi(inode);
1227         struct ptlrpc_request *request = NULL;
1228         struct mdc_op_data *op_data;
1229         struct lustre_capa *trunc_capa = NULL;
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                         /* XXX: hack for truncate capa */
1317                         rc = mdc_req2lustre_capa(request, 0, &trunc_capa);
1318                         if (rc) {
1319                                 ptlrpc_req_finished(request);
1320                                 RETURN(rc);
1321                         }
1322
1323                         spin_lock(&lli->lli_lock);
1324                         if (trunc_capa) {
1325                                 if (lli->lli_trunc_capa)
1326                                         OBD_FREE(lli->lli_trunc_capa,
1327                                                  sizeof(*trunc_capa));
1328                                 lli->lli_trunc_capa = trunc_capa;
1329                         }
1330                         spin_unlock(&lli->lli_lock);
1331                 }
1332
1333                 /* We call inode_setattr to adjust timestamps, but we first
1334                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1335                  *
1336                  * NB: ATTR_SIZE will only be set at this point if the size
1337                  * resides on the MDS, ie, this file has no objects. */
1338                 attr->ia_valid &= ~ATTR_SIZE;
1339
1340                 /*
1341                  * assigning inode_setattr() to @err to disable warning that
1342                  * function's result should be checked by by caller. error is
1343                  * impossible here, as vmtruncate() control path is disabled.
1344                  */
1345                 err = inode_setattr(inode, attr);
1346                 ll_update_inode(inode, &md);
1347                 ptlrpc_req_finished(request);
1348
1349                 if (!lsm || !S_ISREG(inode->i_mode)) {
1350                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1351                         RETURN(0);
1352                 }
1353         } else {
1354                 /* The OST doesn't check permissions, but the alternative is
1355                  * a gratuitous RPC to the MDS.  We already rely on the client
1356                  * to do read/write/truncate permission checks, so is mtime OK?
1357                  */
1358                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1359                         /* from sys_utime() */
1360                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1361                                 if (current->fsuid != inode->i_uid &&
1362                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1363                                         RETURN(rc);
1364                         } else {
1365                                 /* from inode_change_ok() */
1366                                 if (current->fsuid != inode->i_uid &&
1367                                     !capable(CAP_FOWNER))
1368                                         RETURN(-EPERM);
1369                         }
1370                 }
1371
1372                 if (lsm)
1373                         attr->ia_valid &= ~ATTR_SIZE;
1374
1375                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1376                 err = inode_setattr(inode, attr);
1377                 /*
1378                  * assigning inode_setattr() to @err to disable warning that
1379                  * function's result should be checked by by caller. error is
1380                  * impossible here, as vmtruncate() control path is disabled.
1381                  */
1382         }
1383
1384         /* We really need to get our PW lock before we change inode->i_size.
1385          * If we don't we can race with other i_size updaters on our node, like
1386          * ll_file_read.  We can also race with i_size propogation to other
1387          * nodes through dirtying and writeback of final cached pages.  This
1388          * last one is especially bad for racing o_append users on other
1389          * nodes. */
1390         if (ia_valid & ATTR_SIZE) {
1391                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1392                                                            OBD_OBJECT_EOF } };
1393                 struct lustre_handle lockh = { 0 };
1394                 int err, ast_flags = 0;
1395                 /* XXX when we fix the AST intents to pass the discard-range
1396                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1397                  * XXX here. */
1398                 if (attr->ia_size == 0)
1399                         ast_flags = LDLM_AST_DISCARD_DATA;
1400
1401                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1402                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1403
1404                 if (rc != 0)
1405                         RETURN(rc);
1406
1407                 down(&lli->lli_size_sem);
1408                 lli->lli_size_pid = current->pid;
1409                 rc = vmtruncate(inode, attr->ia_size);
1410                 if (rc != 0) {
1411                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1412                         lli->lli_size_pid = 0;
1413                         up(&lli->lli_size_sem);
1414                 }
1415
1416                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1417                 if (err) {
1418                         CERROR("ll_extent_unlock failed: %d\n", err);
1419                         if (!rc)
1420                                 rc = err;
1421                 }
1422         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1423                 struct obdo *oa = NULL;
1424
1425                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1426                        inode->i_ino, LTIME_S(attr->ia_mtime));
1427
1428                 oa = obdo_alloc();
1429                 if (oa == NULL)
1430                         RETURN(-ENOMEM);
1431
1432                 oa->o_id = lsm->lsm_object_id;
1433                 oa->o_gr = lsm->lsm_object_gr;
1434                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1435
1436                 /* adding uid and gid, needed for quota */
1437                 if (ia_valid & ATTR_UID) {
1438                         oa->o_uid = inode->i_uid;
1439                         oa->o_valid |= OBD_MD_FLUID;
1440                 }
1441
1442                 if (ia_valid & ATTR_GID) {
1443                         oa->o_gid = inode->i_gid;
1444                         oa->o_valid |= OBD_MD_FLGID;
1445                 }
1446
1447                 *(obdo_id(oa)) = lli->lli_id;
1448                 oa->o_valid |= OBD_MD_FLIFID;
1449
1450                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1451                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1452                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1453                 obdo_free(oa);
1454                 if (rc)
1455                         CERROR("obd_setattr fails: rc = %d\n", rc);
1456         }
1457
1458         RETURN(rc);
1459 }
1460
1461 int ll_setattr(struct dentry *de, struct iattr *attr)
1462 {
1463         LASSERT(de->d_inode);
1464         return ll_setattr_raw(de->d_inode, attr);
1465 }
1466
1467 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1468                        unsigned long max_age)
1469 {
1470         struct ll_sb_info *sbi = ll_s2sbi(sb);
1471         struct obd_statfs obd_osfs;
1472         int rc;
1473         ENTRY;
1474
1475         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1476         if (rc) {
1477                 CERROR("obd_statfs fails: rc = %d\n", rc);
1478                 RETURN(rc);
1479         }
1480
1481         osfs->os_type = sb->s_magic;
1482
1483         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1484                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1485
1486         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1487         if (rc) {
1488                 CERROR("obd_statfs fails: rc = %d\n", rc);
1489                 RETURN(rc);
1490         }
1491
1492         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1493                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1494                obd_osfs.os_files);
1495
1496         osfs->os_blocks = obd_osfs.os_blocks;
1497         osfs->os_bfree = obd_osfs.os_bfree;
1498         osfs->os_bavail = obd_osfs.os_bavail;
1499
1500         /* If we don't have as many objects free on the OST as inodes
1501          * on the MDS, we reduce the total number of inodes to
1502          * compensate, so that the "inodes in use" number is correct.
1503          */
1504         if (obd_osfs.os_ffree < osfs->os_ffree) {
1505                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1506                         obd_osfs.os_ffree;
1507                 osfs->os_ffree = obd_osfs.os_ffree;
1508         }
1509
1510         RETURN(rc);
1511 }
1512
1513 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1514 {
1515         struct obd_statfs osfs;
1516         int rc;
1517
1518         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1519         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1520
1521         /* For now we will always get up-to-date statfs values, but in the
1522          * future we may allow some amount of caching on the client (e.g.
1523          * from QOS or lprocfs updates). */
1524         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1525         if (rc)
1526                 return rc;
1527
1528         statfs_unpack(sfs, &osfs);
1529
1530         if (sizeof(sfs->f_blocks) == 4) {
1531                 while (osfs.os_blocks > ~0UL) {
1532                         sfs->f_bsize <<= 1;
1533
1534                         osfs.os_blocks >>= 1;
1535                         osfs.os_bfree >>= 1;
1536                         osfs.os_bavail >>= 1;
1537                 }
1538         }
1539
1540         sfs->f_blocks = osfs.os_blocks;
1541         sfs->f_bfree = osfs.os_bfree;
1542         sfs->f_bavail = osfs.os_bavail;
1543
1544         return 0;
1545 }
1546
1547
1548 /********************************
1549  * remote acl                   *
1550  ********************************/
1551
1552 static struct remote_acl *remote_acl_alloc(void)
1553 {
1554         struct remote_acl *racl;
1555         int i;
1556
1557         OBD_ALLOC(racl, sizeof(*racl));
1558         if (!racl)
1559                 return NULL;
1560
1561         spin_lock_init(&racl->ra_lock);
1562         init_MUTEX(&racl->ra_update_sem);
1563
1564         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1565                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1566
1567         return racl;
1568 }
1569
1570 /*
1571  * caller should guarantee no race here.
1572  */
1573 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1574 {
1575         struct remote_perm_setxid *xperm;
1576
1577         while (!list_empty(&perm->lrp_setxid_perms)) {
1578                 xperm = list_entry(perm->lrp_setxid_perms.next,
1579                                    struct remote_perm_setxid,
1580                                    list);
1581                 list_del(&xperm->list);
1582                 OBD_FREE(xperm, sizeof(*xperm));
1583         }
1584 }
1585
1586 /*
1587  * caller should guarantee no race here.
1588  */
1589 static void remote_acl_flush(struct remote_acl *racl)
1590 {
1591         struct list_head *head;
1592         struct lustre_remote_perm *perm, *tmp;
1593         int i;
1594
1595         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1596                 head = &racl->ra_perm_cache[i];
1597
1598                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1599                         remote_perm_flush_xperms(perm);
1600                         list_del(&perm->lrp_list);
1601                         OBD_FREE(perm, sizeof(*perm));
1602                 }
1603         }
1604 }
1605
1606 static void remote_acl_free(struct remote_acl *racl)
1607 {
1608         if (!racl)
1609                 return;
1610
1611         down(&racl->ra_update_sem);
1612         spin_lock(&racl->ra_lock);
1613         remote_acl_flush(racl);
1614         spin_unlock(&racl->ra_lock);
1615         up(&racl->ra_update_sem);
1616
1617         OBD_FREE(racl, sizeof(*racl));
1618 }
1619
1620 static inline int remote_acl_hashfunc(__u32 id)
1621 {
1622         return (id & (REMOTE_ACL_HASHSIZE - 1));
1623 }
1624
1625 static
1626 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1627 {
1628         struct list_head *head;
1629         struct lustre_remote_perm *lperm;
1630         struct remote_perm_setxid *xperm;
1631         int found = 0, rc = -ENOENT;
1632
1633         LASSERT(racl);
1634         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1635         spin_lock(&racl->ra_lock);
1636
1637         list_for_each_entry(lperm, head, lrp_list) {
1638                 if (lperm->lrp_auth_uid == current->uid) {
1639                         found = 1;
1640                         break;
1641                 }
1642         }
1643
1644         if (!found)
1645                 goto out;
1646
1647         if (lperm->lrp_auth_uid == current->fsuid &&
1648             lperm->lrp_auth_gid == current->fsgid) {
1649                 if (lperm->lrp_valid) {
1650                         *perm = lperm->lrp_perm;
1651                         rc = 0;
1652                 }
1653                 goto out;
1654         } else if ((!lperm->lrp_setuid &&
1655                     lperm->lrp_auth_uid != current->fsuid) ||
1656                    (!lperm->lrp_setgid &&
1657                     lperm->lrp_auth_gid != current->fsgid))  {
1658                 *perm = 0;
1659                 rc = 0;
1660                 goto out;
1661         }
1662
1663         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1664                 if (xperm->uid == current->fsuid &&
1665                     xperm->gid == current->fsgid) {
1666                         *perm = xperm->perm;
1667                         rc = 0;
1668                         goto out;
1669                 }
1670         }
1671
1672 out:
1673         spin_unlock(&racl->ra_lock);
1674         return rc;
1675 }
1676
1677 static
1678 int __remote_acl_update(struct remote_acl *racl,
1679                         struct mds_remote_perm *mperm,
1680                         struct lustre_remote_perm *lperm,
1681                         struct remote_perm_setxid *xperm)
1682 {
1683         struct list_head *head;
1684         struct lustre_remote_perm *lp;
1685         struct remote_perm_setxid *xp;
1686         int found = 0, setuid = 0, setgid = 0;
1687
1688         LASSERT(racl);
1689         LASSERT(mperm);
1690         LASSERT(lperm);
1691         LASSERT(current->uid == mperm->mrp_auth_uid);
1692
1693         if (current->fsuid != mperm->mrp_auth_uid)
1694                 setuid = 1;
1695         if (current->fsgid != mperm->mrp_auth_gid)
1696                 setgid = 1;
1697
1698         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1699         spin_lock(&racl->ra_lock);
1700
1701         list_for_each_entry(lp, head, lrp_list) {
1702                 if (lp->lrp_auth_uid == current->uid) {
1703                         found = 1;
1704                         break;
1705                 }
1706         }
1707
1708         if (found) {
1709                 OBD_FREE(lperm, sizeof(*lperm));
1710
1711                 if (!lp->lrp_valid && !setuid && !setgid) {
1712                         lp->lrp_perm = mperm->mrp_perm;
1713                         lp->lrp_valid = 1;
1714                 }
1715
1716                 /* sanity check for changes of setxid rules */
1717                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1718                         CWARN("setuid changes: %d => %d\n",
1719                               (lp->lrp_setuid != 0),
1720                               (mperm->mrp_allow_setuid != 0));
1721                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1722                 }
1723
1724                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1725                         CWARN("setgid changes: %d => %d\n",
1726                               (lp->lrp_setgid != 0),
1727                               (mperm->mrp_allow_setgid != 0));
1728                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1729                 }
1730
1731                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1732                     !list_empty(&lp->lrp_setxid_perms)) {
1733                         remote_perm_flush_xperms(lp);
1734                 }
1735         } else {
1736                 /* initialize lperm and linked into hashtable
1737                  */
1738                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1739                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1740                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1741                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1742                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1743                 list_add(&lperm->lrp_list, head);
1744
1745                 if (!setuid && !setgid) {
1746                         /* in this case, i'm the authenticated user,
1747                          * and mrp_perm is for me.
1748                          */
1749                         lperm->lrp_perm = mperm->mrp_perm;
1750                         lperm->lrp_valid = 1;
1751                         spin_unlock(&racl->ra_lock);
1752
1753                         if (xperm)
1754                                 OBD_FREE(xperm, sizeof(*xperm));
1755                         return 0;
1756                 }
1757
1758                 lp = lperm;
1759                 /* fall through */
1760         }
1761
1762         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1763                 list_empty(&lp->lrp_setxid_perms));
1764
1765         /* if no xperm supplied, we are all done here */
1766         if (!xperm) {
1767                 spin_unlock(&racl->ra_lock);
1768                 return 0;
1769         }
1770
1771         /* whether we allow setuid/setgid */
1772         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1773                 OBD_FREE(xperm, sizeof(*xperm));
1774                 spin_unlock(&racl->ra_lock);
1775                 return 0;
1776         }
1777
1778         /* traverse xperm list */
1779         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1780                 if (xp->uid == current->fsuid &&
1781                     xp->gid == current->fsgid) {
1782                         if (xp->perm != mperm->mrp_perm) {
1783                                 /* actually this should not happen */
1784                                 CWARN("perm changed: %o => %o\n",
1785                                       xp->perm, mperm->mrp_perm);
1786                                 xp->perm = mperm->mrp_perm;
1787                         }
1788                         OBD_FREE(xperm, sizeof(*xperm));
1789                         spin_unlock(&racl->ra_lock);
1790                         return 0;
1791                 }
1792         }
1793
1794         /* finally insert this xperm */
1795         xperm->uid = current->fsuid;
1796         xperm->gid = current->fsgid;
1797         xperm->perm = mperm->mrp_perm;
1798         list_add(&xperm->list, &lp->lrp_setxid_perms);
1799
1800         spin_unlock(&racl->ra_lock);
1801         return 0;
1802 }
1803
1804 /*
1805  * remote_acl semaphore must be held by caller
1806  */
1807 static
1808 int remote_acl_update_locked(struct remote_acl *racl,
1809                              struct mds_remote_perm *mperm)
1810 {
1811         struct lustre_remote_perm *lperm;
1812         struct remote_perm_setxid *xperm;
1813         int setuid = 0, setgid = 0;
1814
1815         might_sleep();
1816
1817         if (current->uid != mperm->mrp_auth_uid) {
1818                 CERROR("current uid %u while authenticated as %u\n",
1819                        current->uid, mperm->mrp_auth_uid);
1820                 return -EINVAL;
1821         }
1822
1823         if (current->fsuid != mperm->mrp_auth_uid)
1824                 setuid = 1;
1825         if (current->fsgid == mperm->mrp_auth_gid)
1826                 setgid = 1;
1827
1828         OBD_ALLOC(lperm, sizeof(*lperm));
1829         if (!lperm)
1830                 return -ENOMEM;
1831
1832         if ((setuid || setgid) &&
1833             !(setuid && !mperm->mrp_allow_setuid) &&
1834             !(setgid && !mperm->mrp_allow_setgid)) {
1835                 OBD_ALLOC(xperm, sizeof(*xperm));
1836                 if (!xperm) {
1837                         OBD_FREE(lperm, sizeof(*lperm));
1838                         return -ENOMEM;
1839                 }
1840         } else
1841                 xperm = NULL;
1842
1843         return __remote_acl_update(racl, mperm, lperm, xperm);
1844 }
1845
1846 /*
1847  * return -EACCES at any error cases
1848  */
1849 int ll_remote_acl_permission(struct inode *inode, int mode)
1850 {
1851         struct ll_sb_info *sbi = ll_i2sbi(inode);
1852         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1853         struct ptlrpc_request *req = NULL;
1854         struct lustre_id id;
1855         struct mds_remote_perm *mperm;
1856         int rc = -EACCES, perm;
1857
1858         if (!racl)
1859                 return -EACCES;
1860
1861         if (__remote_acl_check(racl, &perm) == 0) {
1862                 return ((perm & mode) == mode ? 0 : -EACCES);
1863         }
1864
1865         might_sleep();
1866
1867         /* doing update
1868          */
1869         down(&racl->ra_update_sem);
1870
1871         /* we might lose the race when obtain semaphore,
1872          * so check again.
1873          */
1874         if (__remote_acl_check(racl, &perm) == 0) {
1875                 if ((perm & mode) == mode)
1876                         rc = 0;
1877                 goto out;
1878         }
1879
1880         /* really fetch from mds
1881          */
1882         ll_inode2id(&id, inode);
1883         if (md_access_check(sbi->ll_md_exp, &id, &req))
1884                 goto out;
1885
1886         /* status non-zero indicate there's more apparent error
1887          * detected by mds, e.g. didn't allow this user at all.
1888          * we simply ignore and didn't cache it.
1889          */
1890         if (req->rq_repmsg->status)
1891                 goto out;
1892
1893         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1894                                    lustre_swab_remote_perm);
1895         LASSERT(mperm);
1896         LASSERT_REPSWABBED(req, 1);
1897
1898         if ((mperm->mrp_perm & mode) == mode)
1899                 rc = 0;
1900
1901         remote_acl_update_locked(racl, mperm);
1902 out:
1903         if (req)
1904                 ptlrpc_req_finished(req);
1905
1906         up(&racl->ra_update_sem);
1907         return rc;
1908 }
1909
1910 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1911 {
1912         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1913         int rc;
1914
1915         LASSERT(perm);
1916
1917         if (!racl)
1918                 return -EACCES;
1919
1920         down(&racl->ra_update_sem);
1921         rc = remote_acl_update_locked(racl, perm);
1922         up(&racl->ra_update_sem);
1923
1924         return rc;
1925 }
1926
1927 void ll_inode_invalidate_acl(struct inode *inode)
1928 {
1929         struct ll_sb_info *sbi = ll_i2sbi(inode);
1930         struct ll_inode_info *lli = ll_i2info(inode);
1931
1932         if (sbi->ll_remote) {
1933                 struct remote_acl *racl = lli->lli_remote_acl;
1934
1935                 LASSERT(!lli->lli_posix_acl);
1936                 if (racl) {
1937                         down(&racl->ra_update_sem);
1938                         spin_lock(&racl->ra_lock);
1939                         remote_acl_flush(lli->lli_remote_acl);
1940                         spin_unlock(&racl->ra_lock);
1941                         up(&racl->ra_update_sem);
1942                 }
1943         } else {
1944                 /* we can't invalide acl here: suppose we touch a new file
1945                  * under a dir, blocking ast on dir will lead to open failure
1946                  * on client, although succeed on mds. it's kind of weird,
1947                  * the real fix i think is improve client-vfs interaction.
1948                  *
1949                  * currently we just do nothing here.
1950                  */
1951                 return;
1952
1953                 LASSERT(!lli->lli_remote_acl);
1954                 spin_lock(&lli->lli_lock);
1955                 posix_acl_release(lli->lli_posix_acl);
1956                 lli->lli_posix_acl = NULL;
1957                 spin_unlock(&lli->lli_lock);
1958         }
1959 }
1960
1961 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1962 {
1963         struct ll_inode_info *lli = ll_i2info(inode);
1964         struct lov_stripe_md *lsm = md->lsm;
1965         struct mds_body *body = md->body;
1966         struct mea *mea = md->mea;
1967         struct posix_acl *posix_acl = md->posix_acl;
1968         struct ll_sb_info *sbi = ll_i2sbi(inode);
1969         struct lustre_key *mkey = md->key;
1970         ENTRY;
1971
1972         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1973
1974         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1975                 /* check for default striping info for dir. */
1976                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1977         }
1978
1979         if (lsm != NULL) {
1980                 LASSERT(lsm->lsm_object_gr > 0);
1981                 if (lli->lli_smd == NULL) {
1982                         lli->lli_smd = lsm;
1983                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1984                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1985                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1986                 } else {
1987                         int i;
1988                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1989                                 CERROR("lsm mismatch for inode %ld\n",
1990                                        inode->i_ino);
1991                                 CERROR("lli_smd:\n");
1992                                 dump_lsm(D_ERROR, lli->lli_smd);
1993                                 CERROR("lsm:\n");
1994                                 dump_lsm(D_ERROR, lsm);
1995                                 LBUG();
1996                         }
1997                         /* XXX FIXME -- We should decide on a safer (atomic) and
1998                          * more elegant way to update the lsm */
1999                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2000                                 lli->lli_smd->lsm_oinfo[i].loi_id =
2001                                         lsm->lsm_oinfo[i].loi_id;
2002                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
2003                                         lsm->lsm_oinfo[i].loi_gr;
2004                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
2005                                         lsm->lsm_oinfo[i].loi_ost_idx;
2006                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
2007                                         lsm->lsm_oinfo[i].loi_ost_gen;
2008                         }
2009                 }
2010                 /* bug 2844 - limit i_blksize for broken user-space apps */
2011                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
2012                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
2013                 if (lli->lli_smd != lsm)
2014                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
2015         }
2016
2017         if (mea != NULL) {
2018                 if (lli->lli_mea == NULL) {
2019                         lli->lli_mea = mea;
2020                 } else {
2021                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
2022                                 CERROR("mea mismatch for inode %lu\n",
2023                                         inode->i_ino);
2024                                 LBUG();
2025                         }
2026                 }
2027                 if (lli->lli_mea != mea)
2028                         obd_free_memmd(ll_i2mdexp(inode),
2029                                        (struct lov_stripe_md **) &mea);
2030         }
2031
2032         if (body->valid & OBD_MD_FID)
2033                 id_assign_fid(&lli->lli_id, &body->id1);
2034
2035         if (body->valid & OBD_MD_FLID)
2036                 id_ino(&lli->lli_id) = id_ino(&body->id1);
2037
2038         if (body->valid & OBD_MD_FLGENER)
2039                 id_gen(&lli->lli_id) = id_gen(&body->id1);
2040
2041         /* local/remote ACL */
2042         if (sbi->ll_remote) {
2043                 LASSERT(md->posix_acl == NULL);
2044                 if (md->remote_perm) {
2045                         ll_remote_acl_update(inode, md->remote_perm);
2046                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2047                         md->remote_perm = NULL;
2048                 }
2049         } else {
2050                 LASSERT(md->remote_perm == NULL);
2051                 spin_lock(&lli->lli_lock);
2052                 if (posix_acl != NULL) {
2053                         if (lli->lli_posix_acl != NULL)
2054                                 posix_acl_release(lli->lli_posix_acl);
2055                         lli->lli_posix_acl = posix_acl;
2056                 }
2057                 spin_unlock(&lli->lli_lock);
2058         }
2059
2060         if (body->valid & OBD_MD_FLID)
2061                 inode->i_ino = id_ino(&body->id1);
2062         if (body->valid & OBD_MD_FLGENER)
2063                 inode->i_generation = id_gen(&body->id1);
2064         if (body->valid & OBD_MD_FLATIME)
2065                 LTIME_S(inode->i_atime) = body->atime;
2066         if (body->valid & OBD_MD_FLMTIME &&
2067             body->mtime > LTIME_S(inode->i_mtime)) {
2068                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2069                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2070                 LTIME_S(inode->i_mtime) = body->mtime;
2071         }
2072         if (body->valid & OBD_MD_FLCTIME &&
2073             body->ctime > LTIME_S(inode->i_ctime))
2074                 LTIME_S(inode->i_ctime) = body->ctime;
2075         if (body->valid & OBD_MD_FLMODE) {
2076                 inode->i_mode = (inode->i_mode & S_IFMT) |
2077                         (body->mode & ~S_IFMT);
2078         }
2079         if (body->valid & OBD_MD_FLTYPE) {
2080                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2081                         (body->mode & S_IFMT);
2082         }
2083         if (body->valid & OBD_MD_FLUID)
2084                 inode->i_uid = body->uid;
2085         if (body->valid & OBD_MD_FLGID)
2086                 inode->i_gid = body->gid;
2087         if (body->valid & OBD_MD_FLFLAGS)
2088                 inode->i_flags = body->flags;
2089         if (body->valid & OBD_MD_FLNLINK)
2090                 inode->i_nlink = body->nlink;
2091         if (body->valid & OBD_MD_FLRDEV)
2092 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2093                 inode->i_rdev = body->rdev;
2094 #else
2095                 inode->i_rdev = old_decode_dev(body->rdev);
2096 #endif
2097         if (body->valid & OBD_MD_FLSIZE)
2098                 inode->i_size = body->size;
2099         if (body->valid & OBD_MD_FLBLOCKS)
2100                 inode->i_blocks = body->blocks;
2101
2102         if (body->valid & OBD_MD_FLSIZE)
2103                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2104
2105         if (body->valid & OBD_MD_FLAUDIT) {
2106                 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2107                 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2108                         sbi->ll_audit_mask = body->audit;
2109                 else
2110                         lli->lli_audit_mask = body->audit;
2111         }
2112
2113         if (mkey != NULL) {
2114                 LASSERT(body->valid & OBD_MD_FLKEY);
2115                 ll_crypto_init_inode_key(inode, mkey);
2116         }
2117
2118 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2119         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2120 #endif
2121         LASSERT(id_fid(&lli->lli_id) != 0);
2122 }
2123
2124 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2125 static struct backing_dev_info ll_backing_dev_info = {
2126         .ra_pages       = 0,    /* No readahead */
2127         .memory_backed  = 0,    /* Does contribute to dirty memory */
2128 };
2129 #endif
2130
2131 void ll_read_inode2(struct inode *inode, void *opaque)
2132 {
2133         struct lustre_md *md = opaque;
2134         struct ll_inode_info *lli = ll_i2info(inode);
2135         ENTRY;
2136
2137         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2138                inode->i_generation, inode);
2139
2140         ll_lli_init(lli);
2141
2142         LASSERT(!lli->lli_smd);
2143
2144         if (ll_i2sbi(inode)->ll_remote) {
2145                 lli->lli_remote_acl = remote_acl_alloc();
2146                 /* if failed alloc, nobody will be able to access this inode */
2147         }
2148
2149         /* Core attributes from the MDS first.  This is a new inode, and
2150          * the VFS doesn't zero times in the core inode so we have to do
2151          * it ourselves.  They will be overwritten by either MDS or OST
2152          * attributes - we just need to make sure they aren't newer. */
2153         LTIME_S(inode->i_mtime) = 0;
2154         LTIME_S(inode->i_atime) = 0;
2155         LTIME_S(inode->i_ctime) = 0;
2156
2157         inode->i_rdev = 0;
2158         ll_update_inode(inode, md);
2159
2160         /* OIDEBUG(inode); */
2161
2162         if (S_ISREG(inode->i_mode)) {
2163                 inode->i_op = &ll_file_inode_operations;
2164                 inode->i_fop = &ll_file_operations;
2165                 inode->i_mapping->a_ops = &ll_aops;
2166                 EXIT;
2167         } else if (S_ISDIR(inode->i_mode)) {
2168                 inode->i_op = &ll_dir_inode_operations;
2169                 inode->i_fop = &ll_dir_operations;
2170                 inode->i_mapping->a_ops = &ll_dir_aops;
2171                 EXIT;
2172         } else if (S_ISLNK(inode->i_mode)) {
2173                 inode->i_op = &ll_fast_symlink_inode_operations;
2174                 EXIT;
2175         } else {
2176                 inode->i_op = &ll_special_inode_operations;
2177
2178 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2179                 init_special_inode(inode, inode->i_mode,
2180                                    kdev_t_to_nr(inode->i_rdev));
2181
2182                 /* initializing backing dev info. */
2183                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2184 #else
2185                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2186 #endif
2187                 lli->ll_save_ifop = inode->i_fop;
2188
2189                 if (S_ISCHR(inode->i_mode))
2190                         inode->i_fop = &ll_special_chr_inode_fops;
2191                 else if (S_ISBLK(inode->i_mode))
2192                         inode->i_fop = &ll_special_blk_inode_fops;
2193                 else if (S_ISFIFO(inode->i_mode))
2194                         inode->i_fop = &ll_special_fifo_inode_fops;
2195                 else if (S_ISSOCK(inode->i_mode))
2196                         inode->i_fop = &ll_special_sock_inode_fops;
2197
2198                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2199                       inode->i_fop);
2200
2201                 if (lli->ll_save_ifop->owner) {
2202                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2203                               lli->ll_save_ifop->owner);
2204                 }
2205                 EXIT;
2206         }
2207 }
2208
2209 void ll_delete_inode(struct inode *inode)
2210 {
2211         struct ll_sb_info *sbi = ll_i2sbi(inode);
2212         struct lustre_id id;
2213         int rc;
2214         ENTRY;
2215
2216         ll_inode2id(&id, inode);
2217
2218         rc = md_delete_inode(sbi->ll_md_exp, &id);
2219         if (rc) {
2220                 CERROR("md_delete_inode() failed, error %d\n",
2221                        rc);
2222         }
2223
2224         clear_inode(inode);
2225         EXIT;
2226 }
2227
2228 int ll_iocontrol(struct inode *inode, struct file *file,
2229                  unsigned int cmd, unsigned long arg)
2230 {
2231         struct ll_sb_info *sbi = ll_i2sbi(inode);
2232         struct ptlrpc_request *req = NULL;
2233         int rc, flags = 0;
2234         ENTRY;
2235
2236         switch(cmd) {
2237         case EXT3_IOC_GETFLAGS: {
2238                 struct lustre_id id;
2239                 __u64 valid = OBD_MD_FLFLAGS;
2240                 struct mds_body *body;
2241
2242                 ll_inode2id(&id, inode);
2243                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2244                                 0, 0, NULL, &req);
2245                 if (rc) {
2246                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2247                         RETURN(-abs(rc));
2248                 }
2249
2250                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2251
2252                 if (body->flags & S_APPEND)
2253                         flags |= EXT3_APPEND_FL;
2254                 if (body->flags & S_IMMUTABLE)
2255                         flags |= EXT3_IMMUTABLE_FL;
2256                 if (body->flags & S_NOATIME)
2257                         flags |= EXT3_NOATIME_FL;
2258
2259                 ptlrpc_req_finished (req);
2260
2261                 RETURN(put_user(flags, (int *)arg));
2262         }
2263         case EXT3_IOC_SETFLAGS: {
2264                 struct mdc_op_data *op_data;
2265                 struct iattr attr;
2266                 struct obdo *oa;
2267                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2268
2269                 if (get_user(flags, (int *)arg))
2270                         RETURN(-EFAULT);
2271
2272                 oa = obdo_alloc();
2273                 if (!oa)
2274                         RETURN(-ENOMEM);
2275
2276                 OBD_ALLOC(op_data, sizeof(*op_data));
2277                 if (op_data == NULL) {
2278                         obdo_free(oa);
2279                         RETURN(-ENOMEM);
2280                 }
2281                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
2282
2283                 memset(&attr, 0x0, sizeof(attr));
2284                 attr.ia_attr_flags = flags;
2285                 attr.ia_valid |= ATTR_ATTR_FLAG;
2286
2287                 rc = md_setattr(sbi->ll_md_exp, op_data,
2288                                 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2289                 OBD_FREE(op_data, sizeof(*op_data));
2290                 if (rc) {
2291                         ptlrpc_req_finished(req);
2292                         if (rc != -EPERM && rc != -EACCES)
2293                                 CERROR("md_setattr fails: rc = %d\n", rc);
2294                         obdo_free(oa);
2295                         RETURN(rc);
2296                 }
2297                 ptlrpc_req_finished(req);
2298
2299                 oa->o_id = lsm->lsm_object_id;
2300                 oa->o_gr = lsm->lsm_object_gr;
2301                 oa->o_flags = flags;
2302                 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2303                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2304                               | OBD_MD_FLIFID;
2305
2306                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
2307                 obdo_free(oa);
2308                 if (rc) {
2309                         if (rc != -EPERM && rc != -EACCES)
2310                                 CERROR("md_setattr fails: rc = %d\n", rc);
2311                         RETURN(rc);
2312                 }
2313
2314                 if (flags & EXT3_APPEND_FL)
2315                         inode->i_flags |= S_APPEND;
2316                 else
2317                         inode->i_flags &= ~S_APPEND;
2318                 if (flags & EXT3_IMMUTABLE_FL)
2319                         inode->i_flags |= S_IMMUTABLE;
2320                 else
2321                         inode->i_flags &= ~S_IMMUTABLE;
2322                 if (flags & EXT3_NOATIME_FL)
2323                         inode->i_flags |= S_NOATIME;
2324                 else
2325                         inode->i_flags &= ~S_NOATIME;
2326
2327                 RETURN(0);
2328         }
2329         default:
2330                 RETURN(-ENOSYS);
2331         }
2332
2333         RETURN(0);
2334 }
2335
2336 /* this is only called in the case of forced umount. */
2337 void ll_umount_begin(struct super_block *sb)
2338 {
2339         struct ll_sb_info *sbi = ll_s2sbi(sb);
2340         struct obd_ioctl_data ioc_data = { 0 };
2341         struct obd_device *obd;
2342         ENTRY;
2343
2344         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2345                sb->s_count, atomic_read(&sb->s_active));
2346
2347         obd = class_exp2obd(sbi->ll_md_exp);
2348         if (obd == NULL) {
2349                 CERROR("Invalid MDC connection handle "LPX64"\n",
2350                        sbi->ll_md_exp->exp_handle.h_cookie);
2351                 EXIT;
2352                 return;
2353         }
2354         obd->obd_no_recov = 1;
2355         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2356                       sizeof(ioc_data), &ioc_data, NULL);
2357
2358         obd = class_exp2obd(sbi->ll_dt_exp);
2359         if (obd == NULL) {
2360                 CERROR("Invalid LOV connection handle "LPX64"\n",
2361                        sbi->ll_dt_exp->exp_handle.h_cookie);
2362                 EXIT;
2363                 return;
2364         }
2365
2366         obd->obd_no_recov = 1;
2367         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2368                       sizeof(ioc_data), &ioc_data, NULL);
2369
2370         /*
2371          * really, we'd like to wait until there are no requests outstanding,
2372          * and then continue.  For now, we just invalidate the requests,
2373          * schedule, and hope.
2374          */
2375         schedule();
2376
2377         EXIT;
2378 }
2379
2380 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2381                   struct inode **inode, struct ptlrpc_request *req,
2382                   int offset, struct super_block *sb)
2383 {
2384         struct lustre_md md;
2385         int rc = 0;
2386
2387         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2388         if (rc)
2389                 RETURN(rc);
2390
2391         if (*inode) {
2392                 ll_update_inode(*inode, &md);
2393         } else {
2394                 LASSERT(sb);
2395                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2396                 if (*inode == NULL || is_bad_inode(*inode)) {
2397                         /* free the lsm if we allocated one above */
2398                         if (md.lsm != NULL)
2399                                 obd_free_memmd(dt_exp, &md.lsm);
2400                         if (md.mea != NULL)
2401                                 obd_free_memmd(md_exp,
2402                                                (struct lov_stripe_md**)&md.mea);
2403                         rc = -ENOMEM;
2404                         CERROR("new_inode -fatal: rc %d\n", rc);
2405                 }
2406         }
2407
2408         RETURN(rc);
2409 }
2410
2411 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2412 {
2413         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2414         struct lustre_mount_data *lmd = sbi->ll_lmd;
2415
2416         if (lmd) {
2417                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2418                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2419         }
2420         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2421         if (sbi->ll_remote && lmd)
2422                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2423
2424         if (lmd && lmd->lmd_pag)
2425                 seq_printf(m, ",pag");
2426
2427         return 0;
2428 }
2429
2430 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2431                char *filename, struct lustre_id *ret)
2432 {
2433         struct ptlrpc_request *request = NULL;
2434         struct mds_body *body;
2435         int rc;
2436
2437         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2438                              OBD_MD_FID, 0, &request);
2439         if (rc < 0) {
2440                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2441                        filename, rc);
2442                 return rc;
2443         }
2444
2445         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2446         LASSERT(body != NULL);
2447         LASSERT_REPSWABBED(request, 0);
2448
2449         *ret = body->id1;
2450         ptlrpc_req_finished(request);
2451
2452         return rc;
2453 }
2454 int ll_flush_cred(struct inode *inode)
2455 {
2456         struct ll_sb_info *sbi = ll_i2sbi(inode);
2457         int rc = 0;
2458
2459         /* XXX to avoid adding api, we simply use set_info() interface
2460          * to notify underlying obds. set_info() is more like a ioctl() now...
2461          */
2462         if (sbi->ll_md_exp) {
2463                 rc = obd_set_info(sbi->ll_md_exp,
2464                                   strlen("flush_cred"), "flush_cred",
2465                                   0, NULL);
2466                 if (rc)
2467                         return rc;
2468         }
2469
2470         if (sbi->ll_dt_exp) {
2471                 rc = obd_set_info(sbi->ll_dt_exp,
2472                                   strlen("flush_cred"), "flush_cred",
2473                                   0, NULL);
2474                 if (rc)
2475                         return rc;
2476         }
2477
2478         return rc;
2479 }