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         rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
1053                                LDLM_FL_CONFIG_CHANGE, NULL);
1054         if (rc != 0)
1055                 CWARN("obd_cancel_unused(mdc): %d\n", rc);
1056
1057         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
1058                                LDLM_FL_CONFIG_CHANGE, NULL);
1059         if (rc != 0)
1060                 CWARN("obd_cancel_unused(lov): %d\n", rc);
1061
1062         cfg.cfg_instance = sbi->ll_instance;
1063         cfg.cfg_uuid = sbi->ll_sb_uuid;
1064         cfg.cfg_local_nid = lmd->lmd_local_nid;
1065
1066         namelen = strlen(profile) + 20; /* -clean-######### */
1067         OBD_ALLOC(name, namelen);
1068         if (name == NULL)
1069                 RETURN(-ENOMEM);
1070
1071         if (clean) {
1072                 version = sbi->ll_config_version - 1;
1073                 sprintf(name, "%s-clean-%d", profile, version);
1074         } else {
1075                 version = sbi->ll_config_version + 1;
1076                 sprintf(name, "%s-%d", profile, version);
1077         }
1078
1079         CWARN("Applying configuration log %s\n", name);
1080
1081         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1082                                 LLOG_CONFIG_REPL_CTXT);
1083         rc = class_config_process_llog(ctxt, name, &cfg);
1084         if (rc == 0)
1085                 sbi->ll_config_version = version;
1086         CWARN("Finished applying configuration log %s: %d\n", name, rc);
1087
1088         if (rc == 0 && clean == 0) {
1089                 struct lov_desc desc;
1090                 __u32 valsize;
1091                 int rc = 0;
1092
1093                 valsize = sizeof(desc);
1094                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1095                                   "lovdesc", &valsize, &desc);
1096
1097                 rc = obd_init_ea_size(sbi->ll_md_exp,
1098                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
1099                                       (desc.ld_tgt_count *
1100                                        sizeof(struct llog_cookie)));
1101         }
1102         OBD_FREE(name, namelen);
1103         RETURN(rc);
1104 }
1105
1106 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1107 {
1108         struct inode *inode = NULL;
1109
1110         /* NOTE: we depend on atomic igrab() -bzzz */
1111         lock_res_and_lock(lock);
1112         if (lock->l_ast_data) {
1113                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1114                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1115                         inode = igrab(lock->l_ast_data);
1116                 } else {
1117                         struct timeval now;
1118                         do_gettimeofday(&now);
1119                         inode = lock->l_ast_data;
1120                         LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1121                                    lock->l_enqueued_time.tv_sec,
1122                                    lock->l_enqueued_time.tv_usec,
1123                                    now.tv_sec, now.tv_usec);
1124                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1125                                "l_ast_data %p is bogus: magic %0x8\n",
1126                                lock->l_ast_data, lli->lli_inode_magic);
1127                         CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1128                                inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1129                         inode = NULL;
1130                         unlock_res_and_lock(lock);
1131                         LBUG();
1132                 }
1133         }
1134         unlock_res_and_lock(lock);
1135         return inode;
1136 }
1137
1138 int null_if_equal(struct ldlm_lock *lock, void *data)
1139 {
1140         if (data == lock->l_ast_data) {
1141                 lock->l_ast_data = NULL;
1142
1143                 if (lock->l_req_mode != lock->l_granted_mode)
1144                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1145         }
1146
1147         return LDLM_ITER_CONTINUE;
1148 }
1149
1150 static void remote_acl_free(struct remote_acl *racl);
1151
1152 void ll_clear_inode(struct inode *inode)
1153 {
1154         struct lustre_id id;
1155         struct ll_inode_info *lli = ll_i2info(inode);
1156         struct ll_sb_info *sbi = ll_i2sbi(inode);
1157         ENTRY;
1158
1159         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1160                inode->i_generation, inode);
1161
1162         LASSERT(ll_is_inode_dirty(inode) == 0);
1163         ll_inode2id(&id, inode);
1164
1165         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1166         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1167
1168         LASSERT(!lli->lli_open_fd_write_count);
1169         LASSERT(!lli->lli_open_fd_read_count);
1170         LASSERT(!lli->lli_open_fd_exec_count);
1171         if (lli->lli_mds_write_och)
1172                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1173         if (lli->lli_mds_exec_och)
1174                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1175         if (lli->lli_mds_read_och)
1176                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1177         if (lli->lli_smd)
1178                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1179                                   null_if_equal, inode);
1180
1181         if (lli->lli_smd) {
1182                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1183                 lli->lli_smd = NULL;
1184         }
1185
1186         if (lli->lli_mea) {
1187                 obd_free_memmd(sbi->ll_md_exp,
1188                                (struct lov_stripe_md **) &lli->lli_mea);
1189                 lli->lli_mea = NULL;
1190         }
1191         ll_crypto_destroy_inode_key(inode);
1192         if (lli->lli_symlink_name) {
1193                 OBD_FREE(lli->lli_symlink_name,
1194                          strlen(lli->lli_symlink_name) + 1);
1195                 lli->lli_symlink_name = NULL;
1196         }
1197
1198         if (lli->lli_posix_acl) {
1199                 LASSERT(lli->lli_remote_acl == NULL);
1200                 posix_acl_release(lli->lli_posix_acl);
1201                 lli->lli_posix_acl = NULL;
1202         }
1203
1204         if (lli->lli_remote_acl) {
1205                 LASSERT(lli->lli_posix_acl == NULL);
1206                 remote_acl_free(lli->lli_remote_acl);
1207                 lli->lli_remote_acl = NULL;
1208         }
1209
1210         if (lli->lli_trunc_capa) {
1211                 OBD_FREE(lli->lli_trunc_capa, sizeof(struct lustre_capa));
1212                 lli->lli_trunc_capa = NULL;
1213         }
1214
1215         lli->lli_inode_magic = LLI_INODE_DEAD;
1216         EXIT;
1217 }
1218
1219 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1220  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1221  * keep these values until such a time that objects are allocated for it.
1222  * We do the MDS operations first, as it is checking permissions for us.
1223  * We don't to the MDS RPC if there is nothing that we want to store there,
1224  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1225  * going to do an RPC anyways.
1226  *
1227  * If we are doing a truncate, we will send the mtime and ctime updates
1228  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1229  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1230  * at the same time.
1231  */
1232 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1233 {
1234         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1235         struct ll_inode_info *lli = ll_i2info(inode);
1236         struct ll_sb_info *sbi = ll_i2sbi(inode);
1237         struct ptlrpc_request *request = NULL;
1238         struct mdc_op_data *op_data;
1239         struct lustre_capa *trunc_capa = NULL;
1240         int ia_valid = attr->ia_valid;
1241         int err, rc = 0;
1242         ENTRY;
1243
1244         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1245         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1246
1247         if (ia_valid & ATTR_SIZE) {
1248                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1249                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1250                                attr->ia_size, ll_file_maxbytes(inode));
1251                         RETURN(-EFBIG);
1252                 }
1253
1254                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1255         }
1256
1257         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1258         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1259                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1260                         RETURN(-EPERM);
1261         }
1262
1263         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1264         if (attr->ia_valid & ATTR_CTIME) {
1265                 attr->ia_ctime = CURRENT_TIME;
1266                 attr->ia_valid |= ATTR_CTIME_SET;
1267         }
1268         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1269                 attr->ia_atime = CURRENT_TIME;
1270                 attr->ia_valid |= ATTR_ATIME_SET;
1271         }
1272         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1273                 attr->ia_mtime = CURRENT_TIME;
1274                 attr->ia_valid |= ATTR_MTIME_SET;
1275         }
1276
1277         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1278                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1279                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1280                        LTIME_S(CURRENT_TIME));
1281
1282         /* If only OST attributes being set on objects, don't do MDS RPC.
1283          * In that case, we need to check permissions and update the local
1284          * inode ourselves so we can call obdo_from_inode() always. */
1285         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1286                 struct lustre_md md;
1287                 void *key = NULL;
1288                 int  key_size = 0;
1289
1290                 OBD_ALLOC(op_data, sizeof(*op_data));
1291                 if (op_data == NULL)
1292                         RETURN(-ENOMEM);
1293                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
1294
1295                 if (ia_valid & (ATTR_UID | ATTR_GID)) {
1296                         rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1297                                                &key_size);
1298                 }
1299                 rc = md_setattr(sbi->ll_md_exp, op_data,
1300                                 attr, key, key_size, NULL, 0, NULL,
1301                                 0, &request);
1302                 OBD_FREE(op_data, sizeof(*op_data));
1303
1304                 if (key && key_size)
1305                         OBD_FREE(key, key_size);
1306                 if (rc) {
1307                         ptlrpc_req_finished(request);
1308                         if (rc != -EPERM && rc != -EACCES)
1309                                 CERROR("md_setattr fails: rc = %d\n", rc);
1310                         RETURN(rc);
1311                 }
1312                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1313                                        sbi->ll_dt_exp, &md);
1314                 if (rc) {
1315                         ptlrpc_req_finished(request);
1316                         RETURN(rc);
1317                 }
1318
1319                 if (attr->ia_valid & ATTR_SIZE) {
1320                         /* XXX: hack for truncate capa */
1321                         rc = mdc_req2lustre_capa(request, 0, &trunc_capa);
1322                         if (rc) {
1323                                 ptlrpc_req_finished(request);
1324                                 RETURN(rc);
1325                         }
1326
1327                         spin_lock(&lli->lli_lock);
1328                         if (trunc_capa) {
1329                                 if (lli->lli_trunc_capa)
1330                                         OBD_FREE(lli->lli_trunc_capa,
1331                                                  sizeof(*trunc_capa));
1332                                 lli->lli_trunc_capa = trunc_capa;
1333                         }
1334                         spin_unlock(&lli->lli_lock);
1335                 }
1336
1337                 /* We call inode_setattr to adjust timestamps, but we first
1338                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1339                  *
1340                  * NB: ATTR_SIZE will only be set at this point if the size
1341                  * resides on the MDS, ie, this file has no objects. */
1342                 attr->ia_valid &= ~ATTR_SIZE;
1343
1344                 /*
1345                  * assigning inode_setattr() to @err to disable warning that
1346                  * function's result should be checked by by caller. error is
1347                  * impossible here, as vmtruncate() control path is disabled.
1348                  */
1349                 err = inode_setattr(inode, attr);
1350                 ll_update_inode(inode, &md);
1351                 ptlrpc_req_finished(request);
1352
1353                 if (!lsm || !S_ISREG(inode->i_mode)) {
1354                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1355                         RETURN(0);
1356                 }
1357         } else {
1358                 /* The OST doesn't check permissions, but the alternative is
1359                  * a gratuitous RPC to the MDS.  We already rely on the client
1360                  * to do read/write/truncate permission checks, so is mtime OK?
1361                  */
1362                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1363                         /* from sys_utime() */
1364                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1365                                 if (current->fsuid != inode->i_uid &&
1366                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1367                                         RETURN(rc);
1368                         } else {
1369                                 /* from inode_change_ok() */
1370                                 if (current->fsuid != inode->i_uid &&
1371                                     !capable(CAP_FOWNER))
1372                                         RETURN(-EPERM);
1373                         }
1374                 }
1375
1376                 if (lsm)
1377                         attr->ia_valid &= ~ATTR_SIZE;
1378
1379                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1380                 err = inode_setattr(inode, attr);
1381                 /*
1382                  * assigning inode_setattr() to @err to disable warning that
1383                  * function's result should be checked by by caller. error is
1384                  * impossible here, as vmtruncate() control path is disabled.
1385                  */
1386         }
1387
1388         /* We really need to get our PW lock before we change inode->i_size.
1389          * If we don't we can race with other i_size updaters on our node, like
1390          * ll_file_read.  We can also race with i_size propogation to other
1391          * nodes through dirtying and writeback of final cached pages.  This
1392          * last one is especially bad for racing o_append users on other
1393          * nodes. */
1394         if (ia_valid & ATTR_SIZE) {
1395                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1396                                                            OBD_OBJECT_EOF } };
1397                 struct lustre_handle lockh = { 0 };
1398                 int err, ast_flags = 0;
1399                 /* XXX when we fix the AST intents to pass the discard-range
1400                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1401                  * XXX here. */
1402                 if (attr->ia_size == 0)
1403                         ast_flags = LDLM_AST_DISCARD_DATA;
1404
1405                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1406                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1407
1408                 if (rc != 0)
1409                         RETURN(rc);
1410
1411                 down(&lli->lli_size_sem);
1412                 lli->lli_size_pid = current->pid;
1413                 rc = vmtruncate(inode, attr->ia_size);
1414                 if (rc != 0) {
1415                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1416                         lli->lli_size_pid = 0;
1417                         up(&lli->lli_size_sem);
1418                 }
1419
1420                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1421                 if (err) {
1422                         CERROR("ll_extent_unlock failed: %d\n", err);
1423                         if (!rc)
1424                                 rc = err;
1425                 }
1426         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1427                 struct obdo *oa = NULL;
1428
1429                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1430                        inode->i_ino, LTIME_S(attr->ia_mtime));
1431
1432                 oa = obdo_alloc();
1433                 if (oa == NULL)
1434                         RETURN(-ENOMEM);
1435
1436                 oa->o_id = lsm->lsm_object_id;
1437                 oa->o_gr = lsm->lsm_object_gr;
1438                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1439
1440                 /* adding uid and gid, needed for quota */
1441                 if (ia_valid & ATTR_UID) {
1442                         oa->o_uid = inode->i_uid;
1443                         oa->o_valid |= OBD_MD_FLUID;
1444                 }
1445
1446                 if (ia_valid & ATTR_GID) {
1447                         oa->o_gid = inode->i_gid;
1448                         oa->o_valid |= OBD_MD_FLGID;
1449                 }
1450
1451                 *(obdo_id(oa)) = lli->lli_id;
1452                 oa->o_valid |= OBD_MD_FLIFID;
1453
1454                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1455                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1456                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1457                 obdo_free(oa);
1458                 if (rc)
1459                         CERROR("obd_setattr fails: rc = %d\n", rc);
1460         }
1461
1462         RETURN(rc);
1463 }
1464
1465 int ll_setattr(struct dentry *de, struct iattr *attr)
1466 {
1467         LASSERT(de->d_inode);
1468         return ll_setattr_raw(de->d_inode, attr);
1469 }
1470
1471 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1472                        unsigned long max_age)
1473 {
1474         struct ll_sb_info *sbi = ll_s2sbi(sb);
1475         struct obd_statfs obd_osfs;
1476         int rc;
1477         ENTRY;
1478
1479         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1480         if (rc) {
1481                 CERROR("obd_statfs fails: rc = %d\n", rc);
1482                 RETURN(rc);
1483         }
1484
1485         osfs->os_type = sb->s_magic;
1486
1487         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1488                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1489
1490         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1491         if (rc) {
1492                 CERROR("obd_statfs fails: rc = %d\n", rc);
1493                 RETURN(rc);
1494         }
1495
1496         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1497                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1498                obd_osfs.os_files);
1499
1500         osfs->os_blocks = obd_osfs.os_blocks;
1501         osfs->os_bfree = obd_osfs.os_bfree;
1502         osfs->os_bavail = obd_osfs.os_bavail;
1503
1504         /* If we don't have as many objects free on the OST as inodes
1505          * on the MDS, we reduce the total number of inodes to
1506          * compensate, so that the "inodes in use" number is correct.
1507          */
1508         if (obd_osfs.os_ffree < osfs->os_ffree) {
1509                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1510                         obd_osfs.os_ffree;
1511                 osfs->os_ffree = obd_osfs.os_ffree;
1512         }
1513
1514         RETURN(rc);
1515 }
1516
1517 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1518 {
1519         struct obd_statfs osfs;
1520         int rc;
1521
1522         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1523         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1524
1525         /* For now we will always get up-to-date statfs values, but in the
1526          * future we may allow some amount of caching on the client (e.g.
1527          * from QOS or lprocfs updates). */
1528         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1529         if (rc)
1530                 return rc;
1531
1532         statfs_unpack(sfs, &osfs);
1533
1534         if (sizeof(sfs->f_blocks) == 4) {
1535                 while (osfs.os_blocks > ~0UL) {
1536                         sfs->f_bsize <<= 1;
1537
1538                         osfs.os_blocks >>= 1;
1539                         osfs.os_bfree >>= 1;
1540                         osfs.os_bavail >>= 1;
1541                 }
1542         }
1543
1544         sfs->f_blocks = osfs.os_blocks;
1545         sfs->f_bfree = osfs.os_bfree;
1546         sfs->f_bavail = osfs.os_bavail;
1547
1548         return 0;
1549 }
1550
1551
1552 /********************************
1553  * remote acl                   *
1554  ********************************/
1555
1556 static struct remote_acl *remote_acl_alloc(void)
1557 {
1558         struct remote_acl *racl;
1559         int i;
1560
1561         OBD_ALLOC(racl, sizeof(*racl));
1562         if (!racl)
1563                 return NULL;
1564
1565         spin_lock_init(&racl->ra_lock);
1566         init_MUTEX(&racl->ra_update_sem);
1567
1568         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1569                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1570
1571         return racl;
1572 }
1573
1574 /*
1575  * caller should guarantee no race here.
1576  */
1577 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1578 {
1579         struct remote_perm_setxid *xperm;
1580
1581         while (!list_empty(&perm->lrp_setxid_perms)) {
1582                 xperm = list_entry(perm->lrp_setxid_perms.next,
1583                                    struct remote_perm_setxid,
1584                                    list);
1585                 list_del(&xperm->list);
1586                 OBD_FREE(xperm, sizeof(*xperm));
1587         }
1588 }
1589
1590 /*
1591  * caller should guarantee no race here.
1592  */
1593 static void remote_acl_flush(struct remote_acl *racl)
1594 {
1595         struct list_head *head;
1596         struct lustre_remote_perm *perm, *tmp;
1597         int i;
1598
1599         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1600                 head = &racl->ra_perm_cache[i];
1601
1602                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1603                         remote_perm_flush_xperms(perm);
1604                         list_del(&perm->lrp_list);
1605                         OBD_FREE(perm, sizeof(*perm));
1606                 }
1607         }
1608 }
1609
1610 static void remote_acl_free(struct remote_acl *racl)
1611 {
1612         if (!racl)
1613                 return;
1614
1615         down(&racl->ra_update_sem);
1616         spin_lock(&racl->ra_lock);
1617         remote_acl_flush(racl);
1618         spin_unlock(&racl->ra_lock);
1619         up(&racl->ra_update_sem);
1620
1621         OBD_FREE(racl, sizeof(*racl));
1622 }
1623
1624 static inline int remote_acl_hashfunc(__u32 id)
1625 {
1626         return (id & (REMOTE_ACL_HASHSIZE - 1));
1627 }
1628
1629 static
1630 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1631 {
1632         struct list_head *head;
1633         struct lustre_remote_perm *lperm;
1634         struct remote_perm_setxid *xperm;
1635         int found = 0, rc = -ENOENT;
1636
1637         LASSERT(racl);
1638         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1639         spin_lock(&racl->ra_lock);
1640
1641         list_for_each_entry(lperm, head, lrp_list) {
1642                 if (lperm->lrp_auth_uid == current->uid) {
1643                         found = 1;
1644                         break;
1645                 }
1646         }
1647
1648         if (!found)
1649                 goto out;
1650
1651         if (lperm->lrp_auth_uid == current->fsuid &&
1652             lperm->lrp_auth_gid == current->fsgid) {
1653                 if (lperm->lrp_valid) {
1654                         *perm = lperm->lrp_perm;
1655                         rc = 0;
1656                 }
1657                 goto out;
1658         } else if ((!lperm->lrp_setuid &&
1659                     lperm->lrp_auth_uid != current->fsuid) ||
1660                    (!lperm->lrp_setgid &&
1661                     lperm->lrp_auth_gid != current->fsgid))  {
1662                 *perm = 0;
1663                 rc = 0;
1664                 goto out;
1665         }
1666
1667         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1668                 if (xperm->uid == current->fsuid &&
1669                     xperm->gid == current->fsgid) {
1670                         *perm = xperm->perm;
1671                         rc = 0;
1672                         goto out;
1673                 }
1674         }
1675
1676 out:
1677         spin_unlock(&racl->ra_lock);
1678         return rc;
1679 }
1680
1681 static
1682 int __remote_acl_update(struct remote_acl *racl,
1683                         struct mds_remote_perm *mperm,
1684                         struct lustre_remote_perm *lperm,
1685                         struct remote_perm_setxid *xperm)
1686 {
1687         struct list_head *head;
1688         struct lustre_remote_perm *lp;
1689         struct remote_perm_setxid *xp;
1690         int found = 0, setuid = 0, setgid = 0;
1691
1692         LASSERT(racl);
1693         LASSERT(mperm);
1694         LASSERT(lperm);
1695         LASSERT(current->uid == mperm->mrp_auth_uid);
1696
1697         if (current->fsuid != mperm->mrp_auth_uid)
1698                 setuid = 1;
1699         if (current->fsgid != mperm->mrp_auth_gid)
1700                 setgid = 1;
1701
1702         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1703         spin_lock(&racl->ra_lock);
1704
1705         list_for_each_entry(lp, head, lrp_list) {
1706                 if (lp->lrp_auth_uid == current->uid) {
1707                         found = 1;
1708                         break;
1709                 }
1710         }
1711
1712         if (found) {
1713                 OBD_FREE(lperm, sizeof(*lperm));
1714
1715                 if (!lp->lrp_valid && !setuid && !setgid) {
1716                         lp->lrp_perm = mperm->mrp_perm;
1717                         lp->lrp_valid = 1;
1718                 }
1719
1720                 /* sanity check for changes of setxid rules */
1721                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1722                         CWARN("setuid changes: %d => %d\n",
1723                               (lp->lrp_setuid != 0),
1724                               (mperm->mrp_allow_setuid != 0));
1725                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1726                 }
1727
1728                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1729                         CWARN("setgid changes: %d => %d\n",
1730                               (lp->lrp_setgid != 0),
1731                               (mperm->mrp_allow_setgid != 0));
1732                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1733                 }
1734
1735                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1736                     !list_empty(&lp->lrp_setxid_perms)) {
1737                         remote_perm_flush_xperms(lp);
1738                 }
1739         } else {
1740                 /* initialize lperm and linked into hashtable
1741                  */
1742                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1743                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1744                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1745                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1746                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1747                 list_add(&lperm->lrp_list, head);
1748
1749                 if (!setuid && !setgid) {
1750                         /* in this case, i'm the authenticated user,
1751                          * and mrp_perm is for me.
1752                          */
1753                         lperm->lrp_perm = mperm->mrp_perm;
1754                         lperm->lrp_valid = 1;
1755                         spin_unlock(&racl->ra_lock);
1756
1757                         if (xperm)
1758                                 OBD_FREE(xperm, sizeof(*xperm));
1759                         return 0;
1760                 }
1761
1762                 lp = lperm;
1763                 /* fall through */
1764         }
1765
1766         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1767                 list_empty(&lp->lrp_setxid_perms));
1768
1769         /* if no xperm supplied, we are all done here */
1770         if (!xperm) {
1771                 spin_unlock(&racl->ra_lock);
1772                 return 0;
1773         }
1774
1775         /* whether we allow setuid/setgid */
1776         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1777                 OBD_FREE(xperm, sizeof(*xperm));
1778                 spin_unlock(&racl->ra_lock);
1779                 return 0;
1780         }
1781
1782         /* traverse xperm list */
1783         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1784                 if (xp->uid == current->fsuid &&
1785                     xp->gid == current->fsgid) {
1786                         if (xp->perm != mperm->mrp_perm) {
1787                                 /* actually this should not happen */
1788                                 CWARN("perm changed: %o => %o\n",
1789                                       xp->perm, mperm->mrp_perm);
1790                                 xp->perm = mperm->mrp_perm;
1791                         }
1792                         OBD_FREE(xperm, sizeof(*xperm));
1793                         spin_unlock(&racl->ra_lock);
1794                         return 0;
1795                 }
1796         }
1797
1798         /* finally insert this xperm */
1799         xperm->uid = current->fsuid;
1800         xperm->gid = current->fsgid;
1801         xperm->perm = mperm->mrp_perm;
1802         list_add(&xperm->list, &lp->lrp_setxid_perms);
1803
1804         spin_unlock(&racl->ra_lock);
1805         return 0;
1806 }
1807
1808 /*
1809  * remote_acl semaphore must be held by caller
1810  */
1811 static
1812 int remote_acl_update_locked(struct remote_acl *racl,
1813                              struct mds_remote_perm *mperm)
1814 {
1815         struct lustre_remote_perm *lperm;
1816         struct remote_perm_setxid *xperm;
1817         int setuid = 0, setgid = 0;
1818
1819         might_sleep();
1820
1821         if (current->uid != mperm->mrp_auth_uid) {
1822                 CERROR("current uid %u while authenticated as %u\n",
1823                        current->uid, mperm->mrp_auth_uid);
1824                 return -EINVAL;
1825         }
1826
1827         if (current->fsuid != mperm->mrp_auth_uid)
1828                 setuid = 1;
1829         if (current->fsgid == mperm->mrp_auth_gid)
1830                 setgid = 1;
1831
1832         OBD_ALLOC(lperm, sizeof(*lperm));
1833         if (!lperm)
1834                 return -ENOMEM;
1835
1836         if ((setuid || setgid) &&
1837             !(setuid && !mperm->mrp_allow_setuid) &&
1838             !(setgid && !mperm->mrp_allow_setgid)) {
1839                 OBD_ALLOC(xperm, sizeof(*xperm));
1840                 if (!xperm) {
1841                         OBD_FREE(lperm, sizeof(*lperm));
1842                         return -ENOMEM;
1843                 }
1844         } else
1845                 xperm = NULL;
1846
1847         return __remote_acl_update(racl, mperm, lperm, xperm);
1848 }
1849
1850 /*
1851  * return -EACCES at any error cases
1852  */
1853 int ll_remote_acl_permission(struct inode *inode, int mode)
1854 {
1855         struct ll_sb_info *sbi = ll_i2sbi(inode);
1856         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1857         struct ptlrpc_request *req = NULL;
1858         struct lustre_id id;
1859         struct mds_remote_perm *mperm;
1860         int rc = -EACCES, perm;
1861
1862         if (!racl)
1863                 return -EACCES;
1864
1865         if (__remote_acl_check(racl, &perm) == 0) {
1866                 return ((perm & mode) == mode ? 0 : -EACCES);
1867         }
1868
1869         might_sleep();
1870
1871         /* doing update
1872          */
1873         down(&racl->ra_update_sem);
1874
1875         /* we might lose the race when obtain semaphore,
1876          * so check again.
1877          */
1878         if (__remote_acl_check(racl, &perm) == 0) {
1879                 if ((perm & mode) == mode)
1880                         rc = 0;
1881                 goto out;
1882         }
1883
1884         /* really fetch from mds
1885          */
1886         ll_inode2id(&id, inode);
1887         if (md_access_check(sbi->ll_md_exp, &id, &req))
1888                 goto out;
1889
1890         /* status non-zero indicate there's more apparent error
1891          * detected by mds, e.g. didn't allow this user at all.
1892          * we simply ignore and didn't cache it.
1893          */
1894         if (req->rq_repmsg->status)
1895                 goto out;
1896
1897         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1898                                    lustre_swab_remote_perm);
1899         LASSERT(mperm);
1900         LASSERT_REPSWABBED(req, 1);
1901
1902         if ((mperm->mrp_perm & mode) == mode)
1903                 rc = 0;
1904
1905         remote_acl_update_locked(racl, mperm);
1906 out:
1907         if (req)
1908                 ptlrpc_req_finished(req);
1909
1910         up(&racl->ra_update_sem);
1911         return rc;
1912 }
1913
1914 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1915 {
1916         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1917         int rc;
1918
1919         LASSERT(perm);
1920
1921         if (!racl)
1922                 return -EACCES;
1923
1924         down(&racl->ra_update_sem);
1925         rc = remote_acl_update_locked(racl, perm);
1926         up(&racl->ra_update_sem);
1927
1928         return rc;
1929 }
1930
1931 void ll_inode_invalidate_acl(struct inode *inode)
1932 {
1933         struct ll_sb_info *sbi = ll_i2sbi(inode);
1934         struct ll_inode_info *lli = ll_i2info(inode);
1935
1936         if (sbi->ll_remote) {
1937                 struct remote_acl *racl = lli->lli_remote_acl;
1938
1939                 LASSERT(!lli->lli_posix_acl);
1940                 if (racl) {
1941                         down(&racl->ra_update_sem);
1942                         spin_lock(&racl->ra_lock);
1943                         remote_acl_flush(lli->lli_remote_acl);
1944                         spin_unlock(&racl->ra_lock);
1945                         up(&racl->ra_update_sem);
1946                 }
1947         } else {
1948                 /* we can't invalide acl here: suppose we touch a new file
1949                  * under a dir, blocking ast on dir will lead to open failure
1950                  * on client, although succeed on mds. it's kind of weird,
1951                  * the real fix i think is improve client-vfs interaction.
1952                  *
1953                  * currently we just do nothing here.
1954                  */
1955                 return;
1956
1957                 LASSERT(!lli->lli_remote_acl);
1958                 spin_lock(&lli->lli_lock);
1959                 posix_acl_release(lli->lli_posix_acl);
1960                 lli->lli_posix_acl = NULL;
1961                 spin_unlock(&lli->lli_lock);
1962         }
1963 }
1964
1965 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1966 {
1967         struct ll_inode_info *lli = ll_i2info(inode);
1968         struct lov_stripe_md *lsm = md->lsm;
1969         struct mds_body *body = md->body;
1970         struct mea *mea = md->mea;
1971         struct posix_acl *posix_acl = md->posix_acl;
1972         struct ll_sb_info *sbi = ll_i2sbi(inode);
1973         struct lustre_key *mkey = md->key;
1974         ENTRY;
1975
1976         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1977
1978         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1979                 /* check for default striping info for dir. */
1980                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1981         }
1982
1983         if (lsm != NULL) {
1984                 LASSERT(lsm->lsm_object_gr > 0);
1985                 if (lli->lli_smd == NULL) {
1986                         lli->lli_smd = lsm;
1987                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1988                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1989                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1990                 } else {
1991                         int i;
1992                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1993                                 CERROR("lsm mismatch for inode %ld\n",
1994                                        inode->i_ino);
1995                                 CERROR("lli_smd:\n");
1996                                 dump_lsm(D_ERROR, lli->lli_smd);
1997                                 CERROR("lsm:\n");
1998                                 dump_lsm(D_ERROR, lsm);
1999                                 LBUG();
2000                         }
2001                         /* XXX FIXME -- We should decide on a safer (atomic) and
2002                          * more elegant way to update the lsm */
2003                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2004                                 lli->lli_smd->lsm_oinfo[i].loi_id =
2005                                         lsm->lsm_oinfo[i].loi_id;
2006                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
2007                                         lsm->lsm_oinfo[i].loi_gr;
2008                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
2009                                         lsm->lsm_oinfo[i].loi_ost_idx;
2010                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
2011                                         lsm->lsm_oinfo[i].loi_ost_gen;
2012                         }
2013                 }
2014                 /* bug 2844 - limit i_blksize for broken user-space apps */
2015                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
2016                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
2017                 if (lli->lli_smd != lsm)
2018                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
2019         }
2020
2021         if (mea != NULL) {
2022                 if (lli->lli_mea == NULL) {
2023                         lli->lli_mea = mea;
2024                 } else {
2025                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
2026                                 CERROR("mea mismatch for inode %lu\n",
2027                                         inode->i_ino);
2028                                 LBUG();
2029                         }
2030                 }
2031                 if (lli->lli_mea != mea)
2032                         obd_free_memmd(ll_i2mdexp(inode),
2033                                        (struct lov_stripe_md **) &mea);
2034         }
2035
2036         if (body->valid & OBD_MD_FID)
2037                 id_assign_fid(&lli->lli_id, &body->id1);
2038
2039         if (body->valid & OBD_MD_FLID)
2040                 id_ino(&lli->lli_id) = id_ino(&body->id1);
2041
2042         if (body->valid & OBD_MD_FLGENER)
2043                 id_gen(&lli->lli_id) = id_gen(&body->id1);
2044
2045         /* local/remote ACL */
2046         if (sbi->ll_remote) {
2047                 LASSERT(md->posix_acl == NULL);
2048                 if (md->remote_perm) {
2049                         ll_remote_acl_update(inode, md->remote_perm);
2050                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2051                         md->remote_perm = NULL;
2052                 }
2053         } else {
2054                 LASSERT(md->remote_perm == NULL);
2055                 spin_lock(&lli->lli_lock);
2056                 if (posix_acl != NULL) {
2057                         if (lli->lli_posix_acl != NULL)
2058                                 posix_acl_release(lli->lli_posix_acl);
2059                         lli->lli_posix_acl = posix_acl;
2060                 }
2061                 spin_unlock(&lli->lli_lock);
2062         }
2063
2064         if (body->valid & OBD_MD_FLID)
2065                 inode->i_ino = id_ino(&body->id1);
2066         if (body->valid & OBD_MD_FLGENER)
2067                 inode->i_generation = id_gen(&body->id1);
2068         if (body->valid & OBD_MD_FLATIME)
2069                 LTIME_S(inode->i_atime) = body->atime;
2070         if (body->valid & OBD_MD_FLMTIME &&
2071             body->mtime > LTIME_S(inode->i_mtime)) {
2072                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2073                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2074                 LTIME_S(inode->i_mtime) = body->mtime;
2075         }
2076         if (body->valid & OBD_MD_FLCTIME &&
2077             body->ctime > LTIME_S(inode->i_ctime))
2078                 LTIME_S(inode->i_ctime) = body->ctime;
2079         if (body->valid & OBD_MD_FLMODE) {
2080                 inode->i_mode = (inode->i_mode & S_IFMT) |
2081                         (body->mode & ~S_IFMT);
2082         }
2083         if (body->valid & OBD_MD_FLTYPE) {
2084                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2085                         (body->mode & S_IFMT);
2086         }
2087         if (body->valid & OBD_MD_FLUID)
2088                 inode->i_uid = body->uid;
2089         if (body->valid & OBD_MD_FLGID)
2090                 inode->i_gid = body->gid;
2091         if (body->valid & OBD_MD_FLFLAGS)
2092                 inode->i_flags = body->flags;
2093         if (body->valid & OBD_MD_FLNLINK)
2094                 inode->i_nlink = body->nlink;
2095         if (body->valid & OBD_MD_FLRDEV)
2096 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2097                 inode->i_rdev = body->rdev;
2098 #else
2099                 inode->i_rdev = old_decode_dev(body->rdev);
2100 #endif
2101         if (body->valid & OBD_MD_FLSIZE)
2102                 inode->i_size = body->size;
2103         if (body->valid & OBD_MD_FLBLOCKS)
2104                 inode->i_blocks = body->blocks;
2105
2106         if (body->valid & OBD_MD_FLSIZE)
2107                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2108
2109         if (body->valid & OBD_MD_FLAUDIT) {
2110                 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2111                 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2112                         sbi->ll_audit_mask = body->audit;
2113                 else
2114                         lli->lli_audit_mask = body->audit;
2115         }
2116
2117         if (mkey != NULL) {
2118                 LASSERT(body->valid & OBD_MD_FLKEY);
2119                 ll_crypto_init_inode_key(inode, mkey);
2120         }
2121
2122 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2123         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2124 #endif
2125         LASSERT(id_fid(&lli->lli_id) != 0);
2126 }
2127
2128 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2129 static struct backing_dev_info ll_backing_dev_info = {
2130         .ra_pages       = 0,    /* No readahead */
2131         .memory_backed  = 0,    /* Does contribute to dirty memory */
2132 };
2133 #endif
2134
2135 void ll_read_inode2(struct inode *inode, void *opaque)
2136 {
2137         struct lustre_md *md = opaque;
2138         struct ll_inode_info *lli = ll_i2info(inode);
2139         ENTRY;
2140
2141         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2142                inode->i_generation, inode);
2143
2144         ll_lli_init(lli);
2145
2146         LASSERT(!lli->lli_smd);
2147
2148         if (ll_i2sbi(inode)->ll_remote) {
2149                 lli->lli_remote_acl = remote_acl_alloc();
2150                 /* if failed alloc, nobody will be able to access this inode */
2151         }
2152
2153         /* Core attributes from the MDS first.  This is a new inode, and
2154          * the VFS doesn't zero times in the core inode so we have to do
2155          * it ourselves.  They will be overwritten by either MDS or OST
2156          * attributes - we just need to make sure they aren't newer. */
2157         LTIME_S(inode->i_mtime) = 0;
2158         LTIME_S(inode->i_atime) = 0;
2159         LTIME_S(inode->i_ctime) = 0;
2160
2161         inode->i_rdev = 0;
2162         ll_update_inode(inode, md);
2163
2164         /* OIDEBUG(inode); */
2165
2166         if (S_ISREG(inode->i_mode)) {
2167                 inode->i_op = &ll_file_inode_operations;
2168                 inode->i_fop = &ll_file_operations;
2169                 inode->i_mapping->a_ops = &ll_aops;
2170                 EXIT;
2171         } else if (S_ISDIR(inode->i_mode)) {
2172                 inode->i_op = &ll_dir_inode_operations;
2173                 inode->i_fop = &ll_dir_operations;
2174                 inode->i_mapping->a_ops = &ll_dir_aops;
2175                 EXIT;
2176         } else if (S_ISLNK(inode->i_mode)) {
2177                 inode->i_op = &ll_fast_symlink_inode_operations;
2178                 EXIT;
2179         } else {
2180                 inode->i_op = &ll_special_inode_operations;
2181
2182 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2183                 init_special_inode(inode, inode->i_mode,
2184                                    kdev_t_to_nr(inode->i_rdev));
2185
2186                 /* initializing backing dev info. */
2187                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2188 #else
2189                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2190 #endif
2191                 lli->ll_save_ifop = inode->i_fop;
2192
2193                 if (S_ISCHR(inode->i_mode))
2194                         inode->i_fop = &ll_special_chr_inode_fops;
2195                 else if (S_ISBLK(inode->i_mode))
2196                         inode->i_fop = &ll_special_blk_inode_fops;
2197                 else if (S_ISFIFO(inode->i_mode))
2198                         inode->i_fop = &ll_special_fifo_inode_fops;
2199                 else if (S_ISSOCK(inode->i_mode))
2200                         inode->i_fop = &ll_special_sock_inode_fops;
2201
2202                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2203                       inode->i_fop);
2204
2205                 if (lli->ll_save_ifop->owner) {
2206                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2207                               lli->ll_save_ifop->owner);
2208                 }
2209                 EXIT;
2210         }
2211 }
2212
2213 void ll_delete_inode(struct inode *inode)
2214 {
2215         struct ll_sb_info *sbi = ll_i2sbi(inode);
2216         struct lustre_id id;
2217         int rc;
2218         ENTRY;
2219
2220         ll_inode2id(&id, inode);
2221
2222         rc = md_delete_inode(sbi->ll_md_exp, &id);
2223         if (rc) {
2224                 CERROR("md_delete_inode() failed, error %d\n",
2225                        rc);
2226         }
2227
2228         clear_inode(inode);
2229         EXIT;
2230 }
2231
2232 int ll_iocontrol(struct inode *inode, struct file *file,
2233                  unsigned int cmd, unsigned long arg)
2234 {
2235         struct ll_sb_info *sbi = ll_i2sbi(inode);
2236         struct ptlrpc_request *req = NULL;
2237         int rc, flags = 0;
2238         ENTRY;
2239
2240         switch(cmd) {
2241         case EXT3_IOC_GETFLAGS: {
2242                 struct lustre_id id;
2243                 __u64 valid = OBD_MD_FLFLAGS;
2244                 struct mds_body *body;
2245
2246                 ll_inode2id(&id, inode);
2247                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2248                                 0, 0, NULL, &req);
2249                 if (rc) {
2250                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2251                         RETURN(-abs(rc));
2252                 }
2253
2254                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2255
2256                 if (body->flags & S_APPEND)
2257                         flags |= EXT3_APPEND_FL;
2258                 if (body->flags & S_IMMUTABLE)
2259                         flags |= EXT3_IMMUTABLE_FL;
2260                 if (body->flags & S_NOATIME)
2261                         flags |= EXT3_NOATIME_FL;
2262
2263                 ptlrpc_req_finished (req);
2264
2265                 RETURN(put_user(flags, (int *)arg));
2266         }
2267         case EXT3_IOC_SETFLAGS: {
2268                 struct mdc_op_data *op_data;
2269                 struct iattr attr;
2270                 struct obdo *oa;
2271                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2272
2273                 if (get_user(flags, (int *)arg))
2274                         RETURN(-EFAULT);
2275
2276                 oa = obdo_alloc();
2277                 if (!oa)
2278                         RETURN(-ENOMEM);
2279
2280                 OBD_ALLOC(op_data, sizeof(*op_data));
2281                 if (op_data == NULL) {
2282                         obdo_free(oa);
2283                         RETURN(-ENOMEM);
2284                 }
2285                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
2286
2287                 memset(&attr, 0x0, sizeof(attr));
2288                 attr.ia_attr_flags = flags;
2289                 attr.ia_valid |= ATTR_ATTR_FLAG;
2290
2291                 rc = md_setattr(sbi->ll_md_exp, op_data,
2292                                 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2293                 OBD_FREE(op_data, sizeof(*op_data));
2294                 if (rc) {
2295                         ptlrpc_req_finished(req);
2296                         if (rc != -EPERM && rc != -EACCES)
2297                                 CERROR("md_setattr fails: rc = %d\n", rc);
2298                         obdo_free(oa);
2299                         RETURN(rc);
2300                 }
2301                 ptlrpc_req_finished(req);
2302
2303                 oa->o_id = lsm->lsm_object_id;
2304                 oa->o_gr = lsm->lsm_object_gr;
2305                 oa->o_flags = flags;
2306                 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2307                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2308                               | OBD_MD_FLIFID;
2309
2310                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
2311                 obdo_free(oa);
2312                 if (rc) {
2313                         if (rc != -EPERM && rc != -EACCES)
2314                                 CERROR("md_setattr fails: rc = %d\n", rc);
2315                         RETURN(rc);
2316                 }
2317
2318                 if (flags & EXT3_APPEND_FL)
2319                         inode->i_flags |= S_APPEND;
2320                 else
2321                         inode->i_flags &= ~S_APPEND;
2322                 if (flags & EXT3_IMMUTABLE_FL)
2323                         inode->i_flags |= S_IMMUTABLE;
2324                 else
2325                         inode->i_flags &= ~S_IMMUTABLE;
2326                 if (flags & EXT3_NOATIME_FL)
2327                         inode->i_flags |= S_NOATIME;
2328                 else
2329                         inode->i_flags &= ~S_NOATIME;
2330
2331                 RETURN(0);
2332         }
2333         default:
2334                 RETURN(-ENOSYS);
2335         }
2336
2337         RETURN(0);
2338 }
2339
2340 /* this is only called in the case of forced umount. */
2341 void ll_umount_begin(struct super_block *sb)
2342 {
2343         struct ll_sb_info *sbi = ll_s2sbi(sb);
2344         struct obd_ioctl_data ioc_data = { 0 };
2345         struct obd_device *obd;
2346         ENTRY;
2347
2348         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2349                sb->s_count, atomic_read(&sb->s_active));
2350
2351         obd = class_exp2obd(sbi->ll_md_exp);
2352         if (obd == NULL) {
2353                 CERROR("Invalid MDC connection handle "LPX64"\n",
2354                        sbi->ll_md_exp->exp_handle.h_cookie);
2355                 EXIT;
2356                 return;
2357         }
2358         obd->obd_no_recov = 1;
2359         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2360                       sizeof(ioc_data), &ioc_data, NULL);
2361
2362         obd = class_exp2obd(sbi->ll_dt_exp);
2363         if (obd == NULL) {
2364                 CERROR("Invalid LOV connection handle "LPX64"\n",
2365                        sbi->ll_dt_exp->exp_handle.h_cookie);
2366                 EXIT;
2367                 return;
2368         }
2369
2370         obd->obd_no_recov = 1;
2371         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2372                       sizeof(ioc_data), &ioc_data, NULL);
2373
2374         /*
2375          * really, we'd like to wait until there are no requests outstanding,
2376          * and then continue.  For now, we just invalidate the requests,
2377          * schedule, and hope.
2378          */
2379         schedule();
2380
2381         EXIT;
2382 }
2383
2384 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2385                   struct inode **inode, struct ptlrpc_request *req,
2386                   int offset, struct super_block *sb)
2387 {
2388         struct lustre_md md;
2389         int rc = 0;
2390
2391         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2392         if (rc)
2393                 RETURN(rc);
2394
2395         if (*inode) {
2396                 ll_update_inode(*inode, &md);
2397         } else {
2398                 LASSERT(sb);
2399                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2400                 if (*inode == NULL || is_bad_inode(*inode)) {
2401                         /* free the lsm if we allocated one above */
2402                         if (md.lsm != NULL)
2403                                 obd_free_memmd(dt_exp, &md.lsm);
2404                         if (md.mea != NULL)
2405                                 obd_free_memmd(md_exp,
2406                                                (struct lov_stripe_md**)&md.mea);
2407                         rc = -ENOMEM;
2408                         CERROR("new_inode -fatal: rc %d\n", rc);
2409                 }
2410         }
2411
2412         RETURN(rc);
2413 }
2414
2415 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2416 {
2417         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2418         struct lustre_mount_data *lmd = sbi->ll_lmd;
2419
2420         if (lmd) {
2421                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2422                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2423         }
2424         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2425         if (sbi->ll_remote && lmd)
2426                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2427
2428         if (lmd && lmd->lmd_pag)
2429                 seq_printf(m, ",pag");
2430
2431         return 0;
2432 }
2433
2434 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2435                char *filename, struct lustre_id *ret)
2436 {
2437         struct ptlrpc_request *request = NULL;
2438         struct mds_body *body;
2439         int rc;
2440
2441         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2442                              OBD_MD_FID, 0, &request);
2443         if (rc < 0) {
2444                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2445                        filename, rc);
2446                 return rc;
2447         }
2448
2449         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2450         LASSERT(body != NULL);
2451         LASSERT_REPSWABBED(request, 0);
2452
2453         *ret = body->id1;
2454         ptlrpc_req_finished(request);
2455
2456         return rc;
2457 }
2458 int ll_flush_cred(struct inode *inode)
2459 {
2460         struct ll_sb_info *sbi = ll_i2sbi(inode);
2461         int rc = 0;
2462
2463         /* XXX to avoid adding api, we simply use set_info() interface
2464          * to notify underlying obds. set_info() is more like a ioctl() now...
2465          */
2466         if (sbi->ll_md_exp) {
2467                 rc = obd_set_info(sbi->ll_md_exp,
2468                                   strlen("flush_cred"), "flush_cred",
2469                                   0, NULL);
2470                 if (rc)
2471                         return rc;
2472         }
2473
2474         if (sbi->ll_dt_exp) {
2475                 rc = obd_set_info(sbi->ll_dt_exp,
2476                                   strlen("flush_cred"), "flush_cred",
2477                                   0, NULL);
2478                 if (rc)
2479                         return rc;
2480         }
2481
2482         return rc;
2483 }