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 }
 586
 587 int ll_fill_super(struct super_block *sb, void *data, int silent)
 588 {
 589         struct ll_sb_info *sbi;
 590         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 591         char *mds_sec = NULL;
 592         char *oss_sec = NULL;
 593         int async, err;
 594         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 595         __u64 remote_flag = 0;
 596         ENTRY;
 597
 598         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 599
 600         sbi = lustre_init_sbi(sb);
 601         if (!sbi)
 602                 RETURN(-ENOMEM);
 603
 604         sbi->ll_flags |= LL_SBI_READAHEAD;
 605         ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
 606                    &async, &sbi->ll_flags);
 607
 608         if (!lov || !lmv) {
 609                 CERROR("no osc %p or no mdc %p\n", lov, lmv);
 610                 GOTO(out, err = -EINVAL);
 611         }
 612
 613         err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
 614                                        oss_sec, nllu, 0, &remote_flag);
 615         EXIT;
 616 out:
 617         if (err)
 618                 lustre_free_sbi(sb);
 619
 620         if (lmv)
 621                 OBD_FREE(lmv, strlen(lmv) + 1);
 622         if (lov)
 623                 OBD_FREE(lov, strlen(lov) + 1);
 624         if (mds_sec)
 625                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 626         if (oss_sec)
 627                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 628         if (gkc)
 629                 OBD_FREE(gkc, strlen(gkc) + 1);
 630
 631         return err;
 632 } /* ll_read_super */
 633
 634 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 635                               struct config_llog_instance *cfg, int allow_recov)
 636 {
 637         struct lustre_cfg *lcfg = NULL;
 638         struct lustre_cfg_bufs bufs;
 639         struct portals_cfg pcfg;
 640         char *peer = "MDS_PEER_UUID";
 641         struct obd_device *obd;
 642         struct lustre_handle md_conn = {0, };
 643         struct obd_export *exp;
 644         char *name = "mdc_dev";
 645         class_uuid_t uuid;
 646         struct obd_uuid lmv_uuid;
 647         struct llog_ctxt *ctxt;
 648         int rc = 0, err = 0;
 649         ENTRY;
 650
 651         if (lmd_bad_magic(lmd))
 652                 RETURN(-EINVAL);
 653
 654         generate_random_uuid(uuid);
 655         class_uuid_unparse(uuid, &lmv_uuid);
 656
 657         if (lmd->lmd_local_nid) {
 658                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 659                 pcfg.pcfg_nal = lmd->lmd_nal;
 660                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 661                 rc = libcfs_nal_cmd(&pcfg);
 662                 if (rc < 0)
 663                         GOTO(out, rc);
 664         }
 665
 666         if (lmd->lmd_nal == SOCKNAL ||
 667             lmd->lmd_nal == OPENIBNAL ||
 668             lmd->lmd_nal == IIBNAL ||
 669             lmd->lmd_nal == VIBNAL ||
 670             lmd->lmd_nal == RANAL) {
 671                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 672                 pcfg.pcfg_nal     = lmd->lmd_nal;
 673                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 674                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 675                 pcfg.pcfg_misc    = lmd->lmd_port;
 676                 rc = libcfs_nal_cmd(&pcfg);
 677                 if (rc < 0)
 678                         GOTO(out, rc);
 679         }
 680         lustre_cfg_bufs_reset(&bufs, name);
 681         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 682
 683         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 684         lcfg->lcfg_nal = lmd->lmd_nal;
 685         lcfg->lcfg_nid = lmd->lmd_server_nid;
 686         LASSERT(lcfg->lcfg_nal);
 687         LASSERT(lcfg->lcfg_nid);
 688         err = class_process_config(lcfg);
 689         lustre_cfg_free(lcfg);
 690         if (err < 0)
 691                 GOTO(out_del_conn, err);
 692
 693         lustre_cfg_bufs_reset(&bufs, name);
 694         lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
 695         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 696
 697         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 698         err = class_process_config(lcfg);
 699         lustre_cfg_free(lcfg);
 700         if (err < 0)
 701                 GOTO(out_del_uuid, err);
 702
 703         lustre_cfg_bufs_reset(&bufs, name);
 704         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 705         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 706
 707         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 708         err = class_process_config(lcfg);
 709         lustre_cfg_free(lcfg);
 710         if (err < 0)
 711                 GOTO(out_detach, err);
 712
 713         obd = class_name2obd(name);
 714         if (obd == NULL)
 715                 GOTO(out_cleanup, rc = -EINVAL);
 716
 717         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 718                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 719         if (rc)
 720                 GOTO(out_cleanup, rc);
 721
 722         if (lmd->lmd_pag) {
 723                 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
 724                 rc = obd_set_info(obd->obd_self_export,
 725                                   strlen("sec_flags"), "sec_flags",
 726                                   sizeof(sec_flags), &sec_flags);
 727                 if (rc)
 728                         GOTO(out_cleanup, rc);
 729         }
 730
 731         /* Disable initial recovery on this import */
 732         rc = obd_set_info(obd->obd_self_export,
 733                           strlen("initial_recov"), "initial_recov",
 734                           sizeof(allow_recov), &allow_recov);
 735         if (rc)
 736                 GOTO(out_cleanup, rc);
 737
 738         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 739         if (rc) {
 740                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 741                 GOTO(out_cleanup, rc);
 742         }
 743
 744         exp = class_conn2export(&md_conn);
 745
 746         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 747         rc = class_config_process_llog(ctxt, profile, cfg);
 748         if (rc)
 749                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 750
 751         err = obd_disconnect(exp, 0);
 752
 753         EXIT;
 754 out_cleanup:
 755         lustre_cfg_bufs_reset(&bufs, name);
 756         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 757         err = class_process_config(lcfg);
 758         lustre_cfg_free(lcfg);
 759         if (err < 0)
 760                 GOTO(out, err);
 761 out_detach:
 762         lustre_cfg_bufs_reset(&bufs, name);
 763         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 764         err = class_process_config(lcfg);
 765         lustre_cfg_free(lcfg);
 766         if (err < 0)
 767                 GOTO(out, err);
 768
 769 out_del_uuid:
 770         lustre_cfg_bufs_reset(&bufs, name);
 771         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 772         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 773         err = class_process_config(lcfg);
 774         lustre_cfg_free(lcfg);
 775
 776 out_del_conn:
 777         if (lmd->lmd_nal == SOCKNAL ||
 778             lmd->lmd_nal == OPENIBNAL ||
 779             lmd->lmd_nal == IIBNAL ||
 780             lmd->lmd_nal == VIBNAL ||
 781             lmd->lmd_nal == RANAL) {
 782                 int err2;
 783
 784                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 785                 pcfg.pcfg_nal     = lmd->lmd_nal;
 786                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 787                 pcfg.pcfg_flags   = 1;          /* single_share */
 788                 err2 = libcfs_nal_cmd(&pcfg);
 789                 if (err2 && !err)
 790                         err = err2;
 791                 if (err < 0)
 792                         GOTO(out, err);
 793         }
 794 out:
 795         if (rc == 0)
 796                 rc = err;
 797
 798         return rc;
 799 }
 800
 801 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 802 {
 803         struct lustre_cfg *lcfg;
 804         struct lustre_cfg_bufs bufs;
 805         struct obd_device *obd;
 806         int next = 0;
 807
 808         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
 809                 int err;
 810
 811                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 812                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 813                 err = class_process_config(lcfg);
 814                 if (err) {
 815                         CERROR("cleanup failed: %s\n", obd->obd_name);
 816                         //continue;
 817                 }
 818
 819                 lcfg->lcfg_command = LCFG_DETACH;
 820                 err = class_process_config(lcfg);
 821                 lustre_cfg_free(lcfg);
 822                 if (err) {
 823                         CERROR("detach failed: %s\n", obd->obd_name);
 824                         //continue;
 825                 }
 826         }
 827
 828         if (sbi->ll_lmd != NULL)
 829                 class_del_profile(sbi->ll_lmd->lmd_profile);
 830 }
 831
 832 static int lustre_process_profile(struct super_block *sb,
 833                                   struct lustre_mount_data *lmd,
 834                                   char **lov, char **lmv, char **gkc)
 835 {
 836         struct ll_sb_info *sbi = ll_s2sbi(sb);
 837         struct lustre_profile *lprof;
 838         struct config_llog_instance cfg;
 839         int len, err = 0;
 840
 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
 859          the address of the 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         sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 884
 885         OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
 886                   strlen(sbi->ll_instance) + 2);
 887         sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 888
 889         if (lprof->lp_gkc) {
 890                OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
 891                                strlen(sbi->ll_instance) + 2);
 892                sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
 893         }
 894 out:
 895         RETURN(err);
 896 }
 897
 898 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
 899 {
 900         struct lustre_mount_data *lmd = sbi->ll_lmd;
 901         struct config_llog_instance cfg;
 902         char *cl_prof;
 903         int len, err = 0;
 904         ENTRY;
 905
 906         if (!lmd)
 907                 RETURN(err);
 908
 909         len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
 910
 911         if (force_umount) {
 912                 CERROR("force umount, doing manual cleanup\n");
 913                 lustre_manual_cleanup(sbi);
 914                 GOTO(free_lmd, 0);
 915
 916         }
 917         if (sbi->ll_instance != NULL) {
 918                 cfg.cfg_instance = sbi->ll_instance;
 919                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 920
 921                 OBD_ALLOC(cl_prof, len);
 922                 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 923                 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 924                 if (err < 0) {
 925                         CERROR("Unable to process log: %s\n", cl_prof);
 926                         lustre_manual_cleanup(sbi);
 927                 }
 928                 OBD_FREE(cl_prof, len);
 929         }
 930 free_lmd:
 931         if (sbi->ll_instance)
 932                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 933         OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 934         RETURN(err);
 935 }
 936
 937 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 938 {
 939         struct lustre_mount_data * lmd = data;
 940         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 941         struct ll_sb_info *sbi;
 942         int err;
 943         ENTRY;
 944
 945         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 946         if (lmd_bad_magic(lmd))
 947                 RETURN(-EINVAL);
 948
 949         sbi = lustre_init_sbi(sb);
 950         if (!sbi)
 951                 RETURN(-ENOMEM);
 952
 953         sbi->ll_flags |= LL_SBI_READAHEAD;
 954
 955         err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
 956         if (err) {
 957                 CERROR("Can not process the profile err %d \n", err);
 958                 GOTO(out_free, err);
 959         }
 960         if (!lov || !lmv) {
 961                 CERROR("no osc %p or no mdc %p \n", lov, lmv);
 962                 GOTO(out_free, err = -EINVAL);
 963         }
 964
 965         err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
 966                                        lmd->lmd_mds_security,
 967                                        lmd->lmd_oss_security,
 968                                        &lmd->lmd_nllu, lmd->lmd_pag,
 969                                        &lmd->lmd_remote_flag);
 970
 971         if (err)
 972                 GOTO(out_free, err);
 973
 974 out_dev:
 975         if (lmv)
 976                 OBD_FREE(lmv, strlen(lmv) + 1);
 977         if (lov)
 978                 OBD_FREE(lov, strlen(lov) + 1);
 979         if (gkc)
 980                 OBD_FREE(gkc, strlen(gkc) + 1);
 981
 982         RETURN(err);
 983 out_free:
 984         lustre_clean_profile(sbi, 0);
 985         lustre_free_sbi(sb);
 986         GOTO(out_dev, err);
 987
 988 } /* lustre_fill_super */
 989
 990 void lustre_put_super(struct super_block *sb)
 991 {
 992         struct obd_device *obd;
 993         struct ll_sb_info *sbi = ll_s2sbi(sb);
 994         int force_umount = 0;
 995         ENTRY;
 996
 997         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 998         obd = class_exp2obd(sbi->ll_md_exp);
 999         if (obd)
1000                 force_umount = obd->obd_no_recov;
1001         obd = NULL;
1002
1003         lustre_common_put_super(sb);
1004         lustre_clean_profile(sbi, force_umount);
1005         lustre_free_sbi(sb);
1006
1007         EXIT;
1008 } /* lustre_put_super */
1009
1010 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1011 {
1012         struct lustre_mount_data *lmd = sbi->ll_lmd;
1013         char *profile = lmd->lmd_profile, *name = NULL;
1014         struct config_llog_instance cfg;
1015         int rc, namelen =  0, version;
1016         struct llog_ctxt *ctxt;
1017         ENTRY;
1018
1019         if (profile == NULL)
1020                 RETURN(0);
1021         if (lmd == NULL) {
1022                 CERROR("Client not mounted with zero-conf; cannot "
1023                        "process update log.\n");
1024                 RETURN(0);
1025         }
1026
1027         rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
1028                                LDLM_FL_CONFIG_CHANGE, NULL);
1029         if (rc != 0)
1030                 CWARN("obd_cancel_unused(mdc): %d\n", rc);
1031
1032         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
1033                                LDLM_FL_CONFIG_CHANGE, NULL);
1034         if (rc != 0)
1035                 CWARN("obd_cancel_unused(lov): %d\n", rc);
1036
1037         cfg.cfg_instance = sbi->ll_instance;
1038         cfg.cfg_uuid = sbi->ll_sb_uuid;
1039         cfg.cfg_local_nid = lmd->lmd_local_nid;
1040
1041         namelen = strlen(profile) + 20; /* -clean-######### */
1042         OBD_ALLOC(name, namelen);
1043         if (name == NULL)
1044                 RETURN(-ENOMEM);
1045
1046         if (clean) {
1047                 version = sbi->ll_config_version - 1;
1048                 sprintf(name, "%s-clean-%d", profile, version);
1049         } else {
1050                 version = sbi->ll_config_version + 1;
1051                 sprintf(name, "%s-%d", profile, version);
1052         }
1053
1054         CWARN("Applying configuration log %s\n", name);
1055
1056         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1057                                 LLOG_CONFIG_REPL_CTXT);
1058         rc = class_config_process_llog(ctxt, name, &cfg);
1059         if (rc == 0)
1060                 sbi->ll_config_version = version;
1061         CWARN("Finished applying configuration log %s: %d\n", name, rc);
1062
1063         if (rc == 0 && clean == 0) {
1064                 struct lov_desc desc;
1065                 __u32 valsize;
1066                 int rc = 0;
1067
1068                 valsize = sizeof(desc);
1069                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1070                                   "lovdesc", &valsize, &desc);
1071
1072                 rc = obd_init_ea_size(sbi->ll_md_exp,
1073                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
1074                                       (desc.ld_tgt_count *
1075                                        sizeof(struct llog_cookie)));
1076         }
1077         OBD_FREE(name, namelen);
1078         RETURN(rc);
1079 }
1080
1081 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1082 {
1083         struct inode *inode = NULL;
1084
1085         /* NOTE: we depend on atomic igrab() -bzzz */
1086         lock_res_and_lock(lock);
1087         if (lock->l_ast_data) {
1088                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1089                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1090                         inode = igrab(lock->l_ast_data);
1091                 } else {
1092                         struct timeval now;
1093                         do_gettimeofday(&now);
1094                         inode = lock->l_ast_data;
1095                         LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1096                                    lock->l_enqueued_time.tv_sec,
1097                                    lock->l_enqueued_time.tv_usec,
1098                                    now.tv_sec, now.tv_usec);
1099                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1100                                "l_ast_data %p is bogus: magic %0x8\n",
1101                                lock->l_ast_data, lli->lli_inode_magic);
1102                         CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1103                                inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1104                         inode = NULL;
1105                         unlock_res_and_lock(lock);
1106                         LBUG();
1107                 }
1108         }
1109         unlock_res_and_lock(lock);
1110         return inode;
1111 }
1112
1113 int null_if_equal(struct ldlm_lock *lock, void *data)
1114 {
1115         if (data == lock->l_ast_data) {
1116                 lock->l_ast_data = NULL;
1117
1118                 if (lock->l_req_mode != lock->l_granted_mode)
1119                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1120         }
1121
1122         return LDLM_ITER_CONTINUE;
1123 }
1124
1125 static void remote_acl_free(struct remote_acl *racl);
1126
1127 void ll_clear_inode(struct inode *inode)
1128 {
1129         struct lustre_id id;
1130         struct ll_inode_info *lli = ll_i2info(inode);
1131         struct ll_sb_info *sbi = ll_i2sbi(inode);
1132         ENTRY;
1133
1134         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1135                inode->i_generation, inode);
1136
1137         ll_inode2id(&id, inode);
1138
1139         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1140         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1141
1142         LASSERT(!lli->lli_open_fd_write_count);
1143         LASSERT(!lli->lli_open_fd_read_count);
1144         LASSERT(!lli->lli_open_fd_exec_count);
1145         if (lli->lli_mds_write_och)
1146                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1147         if (lli->lli_mds_exec_och)
1148                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1149         if (lli->lli_mds_read_och)
1150                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1151         if (lli->lli_smd)
1152                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1153                                   null_if_equal, inode);
1154
1155         if (lli->lli_smd) {
1156                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1157                 lli->lli_smd = NULL;
1158         }
1159
1160         if (lli->lli_mea) {
1161                 obd_free_memmd(sbi->ll_md_exp,
1162                                (struct lov_stripe_md **) &lli->lli_mea);
1163                 lli->lli_mea = NULL;
1164         }
1165         ll_crypto_destroy_inode_key(inode);
1166         if (lli->lli_symlink_name) {
1167                 OBD_FREE(lli->lli_symlink_name,
1168                          strlen(lli->lli_symlink_name) + 1);
1169                 lli->lli_symlink_name = NULL;
1170         }
1171
1172         if (lli->lli_posix_acl) {
1173                 LASSERT(lli->lli_remote_acl == NULL);
1174                 posix_acl_release(lli->lli_posix_acl);
1175                 lli->lli_posix_acl = NULL;
1176         }
1177
1178         if (lli->lli_remote_acl) {
1179                 LASSERT(lli->lli_posix_acl == NULL);
1180                 remote_acl_free(lli->lli_remote_acl);
1181                 lli->lli_remote_acl = NULL;
1182         }
1183
1184         if (lli->lli_trunc_capa) {
1185                 OBD_FREE(lli->lli_trunc_capa, sizeof(struct lustre_capa));
1186                 lli->lli_trunc_capa = NULL;
1187         }
1188
1189         lli->lli_inode_magic = LLI_INODE_DEAD;
1190         EXIT;
1191 }
1192
1193 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1194  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1195  * keep these values until such a time that objects are allocated for it.
1196  * We do the MDS operations first, as it is checking permissions for us.
1197  * We don't to the MDS RPC if there is nothing that we want to store there,
1198  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1199  * going to do an RPC anyways.
1200  *
1201  * If we are doing a truncate, we will send the mtime and ctime updates
1202  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1203  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1204  * at the same time.
1205  */
1206 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1207 {
1208         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1209         struct ll_inode_info *lli = ll_i2info(inode);
1210         struct ll_sb_info *sbi = ll_i2sbi(inode);
1211         struct ptlrpc_request *request = NULL;
1212         struct mdc_op_data *op_data;
1213         struct lustre_capa *trunc_capa = NULL;
1214         int ia_valid = attr->ia_valid;
1215         int err, rc = 0;
1216         ENTRY;
1217
1218         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1219         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1220
1221         if (ia_valid & ATTR_SIZE) {
1222                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1223                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1224                                attr->ia_size, ll_file_maxbytes(inode));
1225                         RETURN(-EFBIG);
1226                 }
1227
1228                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1229         }
1230
1231         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1232         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1233                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1234                         RETURN(-EPERM);
1235         }
1236
1237         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1238         if (attr->ia_valid & ATTR_CTIME) {
1239                 attr->ia_ctime = CURRENT_TIME;
1240                 attr->ia_valid |= ATTR_CTIME_SET;
1241         }
1242         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1243                 attr->ia_atime = CURRENT_TIME;
1244                 attr->ia_valid |= ATTR_ATIME_SET;
1245         }
1246         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1247                 attr->ia_mtime = CURRENT_TIME;
1248                 attr->ia_valid |= ATTR_MTIME_SET;
1249         }
1250
1251         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1252                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1253                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1254                        LTIME_S(CURRENT_TIME));
1255
1256         /* If only OST attributes being set on objects, don't do MDS RPC.
1257          * In that case, we need to check permissions and update the local
1258          * inode ourselves so we can call obdo_from_inode() always. */
1259         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1260                 struct lustre_md md;
1261                 void *key = NULL;
1262                 int  key_size = 0;
1263
1264                 OBD_ALLOC(op_data, sizeof(*op_data));
1265                 if (op_data == NULL)
1266                         RETURN(-ENOMEM);
1267                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1268
1269                 if (ia_valid & (ATTR_UID | ATTR_GID)) {
1270                         rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1271                                                &key_size);
1272                 }
1273                 rc = md_setattr(sbi->ll_md_exp, op_data,
1274                                 attr, key, key_size, NULL, 0, NULL,
1275                                 0, &request);
1276                 OBD_FREE(op_data, sizeof(*op_data));
1277
1278                 if (key && key_size)
1279                         OBD_FREE(key, key_size);
1280                 if (rc) {
1281                         ptlrpc_req_finished(request);
1282                         if (rc != -EPERM && rc != -EACCES)
1283                                 CERROR("md_setattr fails: rc = %d\n", rc);
1284                         RETURN(rc);
1285                 }
1286                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1287                                        sbi->ll_dt_exp, &md);
1288                 if (rc) {
1289                         ptlrpc_req_finished(request);
1290                         RETURN(rc);
1291                 }
1292
1293                 if (attr->ia_valid & ATTR_SIZE) {
1294                         /* XXX: hack for truncate capa */
1295                         rc = mdc_req2lustre_capa(request, 0, &trunc_capa);
1296                         if (rc) {
1297                                 ptlrpc_req_finished(request);
1298                                 RETURN(rc);
1299                         }
1300
1301                         spin_lock(&lli->lli_lock);
1302                         if (trunc_capa) {
1303                                 if (lli->lli_trunc_capa)
1304                                         OBD_FREE(lli->lli_trunc_capa,
1305                                                  sizeof(*trunc_capa));
1306                                 lli->lli_trunc_capa = trunc_capa;
1307                         }
1308                         spin_unlock(&lli->lli_lock);
1309                 }
1310
1311                 /* We call inode_setattr to adjust timestamps, but we first
1312                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1313                  *
1314                  * NB: ATTR_SIZE will only be set at this point if the size
1315                  * resides on the MDS, ie, this file has no objects. */
1316                 attr->ia_valid &= ~ATTR_SIZE;
1317
1318                 /*
1319                  * assigning inode_setattr() to @err to disable warning that
1320                  * function's result should be checked by by caller. error is
1321                  * impossible here, as vmtruncate() control path is disabled.
1322                  */
1323                 err = inode_setattr(inode, attr);
1324                 ll_update_inode(inode, &md);
1325                 ptlrpc_req_finished(request);
1326
1327                 if (!lsm || !S_ISREG(inode->i_mode)) {
1328                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1329                         RETURN(0);
1330                 }
1331         } else {
1332                 /* The OST doesn't check permissions, but the alternative is
1333                  * a gratuitous RPC to the MDS.  We already rely on the client
1334                  * to do read/write/truncate permission checks, so is mtime OK?
1335                  */
1336                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1337                         /* from sys_utime() */
1338                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1339                                 if (current->fsuid != inode->i_uid &&
1340                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1341                                         RETURN(rc);
1342                         } else {
1343                                 /* from inode_change_ok() */
1344                                 if (current->fsuid != inode->i_uid &&
1345                                     !capable(CAP_FOWNER))
1346                                         RETURN(-EPERM);
1347                         }
1348                 }
1349
1350                 if (lsm)
1351                         attr->ia_valid &= ~ATTR_SIZE;
1352
1353                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1354                 err = inode_setattr(inode, attr);
1355                 /*
1356                  * assigning inode_setattr() to @err to disable warning that
1357                  * function's result should be checked by by caller. error is
1358                  * impossible here, as vmtruncate() control path is disabled.
1359                  */
1360         }
1361
1362         /* We really need to get our PW lock before we change inode->i_size.
1363          * If we don't we can race with other i_size updaters on our node, like
1364          * ll_file_read.  We can also race with i_size propogation to other
1365          * nodes through dirtying and writeback of final cached pages.  This
1366          * last one is especially bad for racing o_append users on other
1367          * nodes. */
1368         if (ia_valid & ATTR_SIZE) {
1369                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1370                                                            OBD_OBJECT_EOF } };
1371                 struct lustre_handle lockh = { 0 };
1372                 int err, ast_flags = 0;
1373                 /* XXX when we fix the AST intents to pass the discard-range
1374                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1375                  * XXX here. */
1376                 if (attr->ia_size == 0)
1377                         ast_flags = LDLM_AST_DISCARD_DATA;
1378
1379                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1380                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1381
1382                 if (rc != 0)
1383                         RETURN(rc);
1384
1385                 down(&lli->lli_size_sem);
1386                 lli->lli_size_pid = current->pid;
1387                 rc = vmtruncate(inode, attr->ia_size);
1388                 if (rc != 0) {
1389                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1390                         lli->lli_size_pid = 0;
1391                         up(&lli->lli_size_sem);
1392                 }
1393
1394                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1395                 if (err) {
1396                         CERROR("ll_extent_unlock failed: %d\n", err);
1397                         if (!rc)
1398                                 rc = err;
1399                 }
1400         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1401                 struct obdo *oa = NULL;
1402
1403                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1404                        inode->i_ino, LTIME_S(attr->ia_mtime));
1405
1406                 oa = obdo_alloc();
1407                 if (oa == NULL)
1408                         RETURN(-ENOMEM);
1409
1410                 oa->o_id = lsm->lsm_object_id;
1411                 oa->o_gr = lsm->lsm_object_gr;
1412                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1413
1414                 /* adding uid and gid, needed for quota */
1415                 if (ia_valid & ATTR_UID) {
1416                         oa->o_uid = inode->i_uid;
1417                         oa->o_valid |= OBD_MD_FLUID;
1418                 }
1419
1420                 if (ia_valid & ATTR_GID) {
1421                         oa->o_gid = inode->i_gid;
1422                         oa->o_valid |= OBD_MD_FLGID;
1423                 }
1424
1425                 *(obdo_id(oa)) = lli->lli_id;
1426                 oa->o_valid |= OBD_MD_FLIFID;
1427
1428                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1429                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1430                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1431                 obdo_free(oa);
1432                 if (rc)
1433                         CERROR("obd_setattr fails: rc = %d\n", rc);
1434         }
1435
1436         RETURN(rc);
1437 }
1438
1439 int ll_setattr(struct dentry *de, struct iattr *attr)
1440 {
1441         LASSERT(de->d_inode);
1442         return ll_setattr_raw(de->d_inode, attr);
1443 }
1444
1445 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1446                        unsigned long max_age)
1447 {
1448         struct ll_sb_info *sbi = ll_s2sbi(sb);
1449         struct obd_statfs obd_osfs;
1450         int rc;
1451         ENTRY;
1452
1453         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1454         if (rc) {
1455                 CERROR("obd_statfs fails: rc = %d\n", rc);
1456                 RETURN(rc);
1457         }
1458
1459         osfs->os_type = sb->s_magic;
1460
1461         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1462                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1463
1464         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1465         if (rc) {
1466                 CERROR("obd_statfs fails: rc = %d\n", rc);
1467                 RETURN(rc);
1468         }
1469
1470         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1471                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1472                obd_osfs.os_files);
1473
1474         osfs->os_blocks = obd_osfs.os_blocks;
1475         osfs->os_bfree = obd_osfs.os_bfree;
1476         osfs->os_bavail = obd_osfs.os_bavail;
1477
1478         /* If we don't have as many objects free on the OST as inodes
1479          * on the MDS, we reduce the total number of inodes to
1480          * compensate, so that the "inodes in use" number is correct.
1481          */
1482         if (obd_osfs.os_ffree < osfs->os_ffree) {
1483                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1484                         obd_osfs.os_ffree;
1485                 osfs->os_ffree = obd_osfs.os_ffree;
1486         }
1487
1488         RETURN(rc);
1489 }
1490
1491 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1492 {
1493         struct obd_statfs osfs;
1494         int rc;
1495
1496         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1497         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1498
1499         /* For now we will always get up-to-date statfs values, but in the
1500          * future we may allow some amount of caching on the client (e.g.
1501          * from QOS or lprocfs updates). */
1502         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1503         if (rc)
1504                 return rc;
1505
1506         statfs_unpack(sfs, &osfs);
1507
1508         if (sizeof(sfs->f_blocks) == 4) {
1509                 while (osfs.os_blocks > ~0UL) {
1510                         sfs->f_bsize <<= 1;
1511
1512                         osfs.os_blocks >>= 1;
1513                         osfs.os_bfree >>= 1;
1514                         osfs.os_bavail >>= 1;
1515                 }
1516         }
1517
1518         sfs->f_blocks = osfs.os_blocks;
1519         sfs->f_bfree = osfs.os_bfree;
1520         sfs->f_bavail = osfs.os_bavail;
1521
1522         return 0;
1523 }
1524
1525
1526 /********************************
1527  * remote acl                   *
1528  ********************************/
1529
1530 static struct remote_acl *remote_acl_alloc(void)
1531 {
1532         struct remote_acl *racl;
1533         int i;
1534
1535         OBD_ALLOC(racl, sizeof(*racl));
1536         if (!racl)
1537                 return NULL;
1538
1539         spin_lock_init(&racl->ra_lock);
1540         init_MUTEX(&racl->ra_update_sem);
1541
1542         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1543                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1544
1545         return racl;
1546 }
1547
1548 /*
1549  * caller should guarantee no race here.
1550  */
1551 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1552 {
1553         struct remote_perm_setxid *xperm;
1554
1555         while (!list_empty(&perm->lrp_setxid_perms)) {
1556                 xperm = list_entry(perm->lrp_setxid_perms.next,
1557                                    struct remote_perm_setxid,
1558                                    list);
1559                 list_del(&xperm->list);
1560                 OBD_FREE(xperm, sizeof(*xperm));
1561         }
1562 }
1563
1564 /*
1565  * caller should guarantee no race here.
1566  */
1567 static void remote_acl_flush(struct remote_acl *racl)
1568 {
1569         struct list_head *head;
1570         struct lustre_remote_perm *perm, *tmp;
1571         int i;
1572
1573         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1574                 head = &racl->ra_perm_cache[i];
1575
1576                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1577                         remote_perm_flush_xperms(perm);
1578                         list_del(&perm->lrp_list);
1579                         OBD_FREE(perm, sizeof(*perm));
1580                 }
1581         }
1582 }
1583
1584 static void remote_acl_free(struct remote_acl *racl)
1585 {
1586         if (!racl)
1587                 return;
1588
1589         down(&racl->ra_update_sem);
1590         spin_lock(&racl->ra_lock);
1591         remote_acl_flush(racl);
1592         spin_unlock(&racl->ra_lock);
1593         up(&racl->ra_update_sem);
1594
1595         OBD_FREE(racl, sizeof(*racl));
1596 }
1597
1598 static inline int remote_acl_hashfunc(__u32 id)
1599 {
1600         return (id & (REMOTE_ACL_HASHSIZE - 1));
1601 }
1602
1603 static
1604 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1605 {
1606         struct list_head *head;
1607         struct lustre_remote_perm *lperm;
1608         struct remote_perm_setxid *xperm;
1609         int found = 0, rc = -ENOENT;
1610
1611         LASSERT(racl);
1612         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1613         spin_lock(&racl->ra_lock);
1614
1615         list_for_each_entry(lperm, head, lrp_list) {
1616                 if (lperm->lrp_auth_uid == current->uid) {
1617                         found = 1;
1618                         break;
1619                 }
1620         }
1621
1622         if (!found)
1623                 goto out;
1624
1625         if (lperm->lrp_auth_uid == current->fsuid &&
1626             lperm->lrp_auth_gid == current->fsgid) {
1627                 if (lperm->lrp_valid) {
1628                         *perm = lperm->lrp_perm;
1629                         rc = 0;
1630                 }
1631                 goto out;
1632         } else if ((!lperm->lrp_setuid &&
1633                     lperm->lrp_auth_uid != current->fsuid) ||
1634                    (!lperm->lrp_setgid &&
1635                     lperm->lrp_auth_gid != current->fsgid))  {
1636                 *perm = 0;
1637                 rc = 0;
1638                 goto out;
1639         }
1640
1641         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1642                 if (xperm->uid == current->fsuid &&
1643                     xperm->gid == current->fsgid) {
1644                         *perm = xperm->perm;
1645                         rc = 0;
1646                         goto out;
1647                 }
1648         }
1649
1650 out:
1651         spin_unlock(&racl->ra_lock);
1652         return rc;
1653 }
1654
1655 static
1656 int __remote_acl_update(struct remote_acl *racl,
1657                         struct mds_remote_perm *mperm,
1658                         struct lustre_remote_perm *lperm,
1659                         struct remote_perm_setxid *xperm)
1660 {
1661         struct list_head *head;
1662         struct lustre_remote_perm *lp;
1663         struct remote_perm_setxid *xp;
1664         int found = 0, setuid = 0, setgid = 0;
1665
1666         LASSERT(racl);
1667         LASSERT(mperm);
1668         LASSERT(lperm);
1669         LASSERT(current->uid == mperm->mrp_auth_uid);
1670
1671         if (current->fsuid != mperm->mrp_auth_uid)
1672                 setuid = 1;
1673         if (current->fsgid != mperm->mrp_auth_gid)
1674                 setgid = 1;
1675
1676         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1677         spin_lock(&racl->ra_lock);
1678
1679         list_for_each_entry(lp, head, lrp_list) {
1680                 if (lp->lrp_auth_uid == current->uid) {
1681                         found = 1;
1682                         break;
1683                 }
1684         }
1685
1686         if (found) {
1687                 OBD_FREE(lperm, sizeof(*lperm));
1688
1689                 if (!lp->lrp_valid && !setuid && !setgid) {
1690                         lp->lrp_perm = mperm->mrp_perm;
1691                         lp->lrp_valid = 1;
1692                 }
1693
1694                 /* sanity check for changes of setxid rules */
1695                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1696                         CWARN("setuid changes: %d => %d\n",
1697                               (lp->lrp_setuid != 0),
1698                               (mperm->mrp_allow_setuid != 0));
1699                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1700                 }
1701
1702                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1703                         CWARN("setgid changes: %d => %d\n",
1704                               (lp->lrp_setgid != 0),
1705                               (mperm->mrp_allow_setgid != 0));
1706                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1707                 }
1708
1709                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1710                     !list_empty(&lp->lrp_setxid_perms)) {
1711                         remote_perm_flush_xperms(lp);
1712                 }
1713         } else {
1714                 /* initialize lperm and linked into hashtable
1715                  */
1716                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1717                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1718                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1719                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1720                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1721                 list_add(&lperm->lrp_list, head);
1722
1723                 if (!setuid && !setgid) {
1724                         /* in this case, i'm the authenticated user,
1725                          * and mrp_perm is for me.
1726                          */
1727                         lperm->lrp_perm = mperm->mrp_perm;
1728                         lperm->lrp_valid = 1;
1729                         spin_unlock(&racl->ra_lock);
1730
1731                         if (xperm)
1732                                 OBD_FREE(xperm, sizeof(*xperm));
1733                         return 0;
1734                 }
1735
1736                 lp = lperm;
1737                 /* fall through */
1738         }
1739
1740         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1741                 list_empty(&lp->lrp_setxid_perms));
1742
1743         /* if no xperm supplied, we are all done here */
1744         if (!xperm) {
1745                 spin_unlock(&racl->ra_lock);
1746                 return 0;
1747         }
1748
1749         /* whether we allow setuid/setgid */
1750         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1751                 OBD_FREE(xperm, sizeof(*xperm));
1752                 spin_unlock(&racl->ra_lock);
1753                 return 0;
1754         }
1755
1756         /* traverse xperm list */
1757         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1758                 if (xp->uid == current->fsuid &&
1759                     xp->gid == current->fsgid) {
1760                         if (xp->perm != mperm->mrp_perm) {
1761                                 /* actually this should not happen */
1762                                 CWARN("perm changed: %o => %o\n",
1763                                       xp->perm, mperm->mrp_perm);
1764                                 xp->perm = mperm->mrp_perm;
1765                         }
1766                         OBD_FREE(xperm, sizeof(*xperm));
1767                         spin_unlock(&racl->ra_lock);
1768                         return 0;
1769                 }
1770         }
1771
1772         /* finally insert this xperm */
1773         xperm->uid = current->fsuid;
1774         xperm->gid = current->fsgid;
1775         xperm->perm = mperm->mrp_perm;
1776         list_add(&xperm->list, &lp->lrp_setxid_perms);
1777
1778         spin_unlock(&racl->ra_lock);
1779         return 0;
1780 }
1781
1782 /*
1783  * remote_acl semaphore must be held by caller
1784  */
1785 static
1786 int remote_acl_update_locked(struct remote_acl *racl,
1787                              struct mds_remote_perm *mperm)
1788 {
1789         struct lustre_remote_perm *lperm;
1790         struct remote_perm_setxid *xperm;
1791         int setuid = 0, setgid = 0;
1792
1793         might_sleep();
1794
1795         if (current->uid != mperm->mrp_auth_uid) {
1796                 CERROR("current uid %u while authenticated as %u\n",
1797                        current->uid, mperm->mrp_auth_uid);
1798                 return -EINVAL;
1799         }
1800
1801         if (current->fsuid != mperm->mrp_auth_uid)
1802                 setuid = 1;
1803         if (current->fsgid == mperm->mrp_auth_gid)
1804                 setgid = 1;
1805
1806         OBD_ALLOC(lperm, sizeof(*lperm));
1807         if (!lperm)
1808                 return -ENOMEM;
1809
1810         if ((setuid || setgid) &&
1811             !(setuid && !mperm->mrp_allow_setuid) &&
1812             !(setgid && !mperm->mrp_allow_setgid)) {
1813                 OBD_ALLOC(xperm, sizeof(*xperm));
1814                 if (!xperm) {
1815                         OBD_FREE(lperm, sizeof(*lperm));
1816                         return -ENOMEM;
1817                 }
1818         } else
1819                 xperm = NULL;
1820
1821         return __remote_acl_update(racl, mperm, lperm, xperm);
1822 }
1823
1824 /*
1825  * return -EACCES at any error cases
1826  */
1827 int ll_remote_acl_permission(struct inode *inode, int mode)
1828 {
1829         struct ll_sb_info *sbi = ll_i2sbi(inode);
1830         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1831         struct ptlrpc_request *req = NULL;
1832         struct lustre_id id;
1833         struct mds_remote_perm *mperm;
1834         int rc = -EACCES, perm;
1835
1836         if (!racl)
1837                 return -EACCES;
1838
1839         if (__remote_acl_check(racl, &perm) == 0) {
1840                 return ((perm & mode) == mode ? 0 : -EACCES);
1841         }
1842
1843         might_sleep();
1844
1845         /* doing update
1846          */
1847         down(&racl->ra_update_sem);
1848
1849         /* we might lose the race when obtain semaphore,
1850          * so check again.
1851          */
1852         if (__remote_acl_check(racl, &perm) == 0) {
1853                 if ((perm & mode) == mode)
1854                         rc = 0;
1855                 goto out;
1856         }
1857
1858         /* really fetch from mds
1859          */
1860         ll_inode2id(&id, inode);
1861         if (md_access_check(sbi->ll_md_exp, &id, &req))
1862                 goto out;
1863
1864         /* status non-zero indicate there's more apparent error
1865          * detected by mds, e.g. didn't allow this user at all.
1866          * we simply ignore and didn't cache it.
1867          */
1868         if (req->rq_repmsg->status)
1869                 goto out;
1870
1871         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1872                                    lustre_swab_remote_perm);
1873         LASSERT(mperm);
1874         LASSERT_REPSWABBED(req, 1);
1875
1876         if ((mperm->mrp_perm & mode) == mode)
1877                 rc = 0;
1878
1879         remote_acl_update_locked(racl, mperm);
1880 out:
1881         if (req)
1882                 ptlrpc_req_finished(req);
1883
1884         up(&racl->ra_update_sem);
1885         return rc;
1886 }
1887
1888 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1889 {
1890         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1891         int rc;
1892
1893         LASSERT(perm);
1894
1895         if (!racl)
1896                 return -EACCES;
1897
1898         down(&racl->ra_update_sem);
1899         rc = remote_acl_update_locked(racl, perm);
1900         up(&racl->ra_update_sem);
1901
1902         return rc;
1903 }
1904
1905 void ll_inode_invalidate_acl(struct inode *inode)
1906 {
1907         struct ll_sb_info *sbi = ll_i2sbi(inode);
1908         struct ll_inode_info *lli = ll_i2info(inode);
1909
1910         if (sbi->ll_remote) {
1911                 struct remote_acl *racl = lli->lli_remote_acl;
1912
1913                 LASSERT(!lli->lli_posix_acl);
1914                 if (racl) {
1915                         down(&racl->ra_update_sem);
1916                         spin_lock(&racl->ra_lock);
1917                         remote_acl_flush(lli->lli_remote_acl);
1918                         spin_unlock(&racl->ra_lock);
1919                         up(&racl->ra_update_sem);
1920                 }
1921         } else {
1922                 /* we can't invalide acl here: suppose we touch a new file
1923                  * under a dir, blocking ast on dir will lead to open failure
1924                  * on client, although succeed on mds. it's kind of weird,
1925                  * the real fix i think is improve client-vfs interaction.
1926                  *
1927                  * currently we just do nothing here.
1928                  */
1929                 return;
1930
1931                 LASSERT(!lli->lli_remote_acl);
1932                 spin_lock(&lli->lli_lock);
1933                 posix_acl_release(lli->lli_posix_acl);
1934                 lli->lli_posix_acl = NULL;
1935                 spin_unlock(&lli->lli_lock);
1936         }
1937 }
1938
1939 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1940 {
1941         struct ll_inode_info *lli = ll_i2info(inode);
1942         struct lov_stripe_md *lsm = md->lsm;
1943         struct mds_body *body = md->body;
1944         struct mea *mea = md->mea;
1945         struct posix_acl *posix_acl = md->posix_acl;
1946         struct ll_sb_info *sbi = ll_i2sbi(inode);
1947         struct lustre_key *mkey = md->key;
1948         ENTRY;
1949
1950         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1951
1952         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1953                 /* check for default striping info for dir. */
1954                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1955         }
1956
1957         if (lsm != NULL) {
1958                 LASSERT(lsm->lsm_object_gr > 0);
1959                 if (lli->lli_smd == NULL) {
1960                         lli->lli_smd = lsm;
1961                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1962                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1963                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1964                 } else {
1965                         int i;
1966                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1967                                 CERROR("lsm mismatch for inode %ld\n",
1968                                        inode->i_ino);
1969                                 CERROR("lli_smd:\n");
1970                                 dump_lsm(D_ERROR, lli->lli_smd);
1971                                 CERROR("lsm:\n");
1972                                 dump_lsm(D_ERROR, lsm);
1973                                 LBUG();
1974                         }
1975                         /* XXX FIXME -- We should decide on a safer (atomic) and
1976                          * more elegant way to update the lsm */
1977                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1978                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1979                                         lsm->lsm_oinfo[i].loi_id;
1980                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1981                                         lsm->lsm_oinfo[i].loi_gr;
1982                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1983                                         lsm->lsm_oinfo[i].loi_ost_idx;
1984                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1985                                         lsm->lsm_oinfo[i].loi_ost_gen;
1986                         }
1987                 }
1988                 /* bug 2844 - limit i_blksize for broken user-space apps */
1989                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1990                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1991                 if (lli->lli_smd != lsm)
1992                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1993         }
1994
1995         if (mea != NULL) {
1996                 if (lli->lli_mea == NULL) {
1997                         lli->lli_mea = mea;
1998                 } else {
1999                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
2000                                 CERROR("mea mismatch for inode %lu\n",
2001                                         inode->i_ino);
2002                                 LBUG();
2003                         }
2004                 }
2005                 if (lli->lli_mea != mea)
2006                         obd_free_memmd(ll_i2mdexp(inode),
2007                                        (struct lov_stripe_md **) &mea);
2008         }
2009
2010         if (body->valid & OBD_MD_FID)
2011                 id_assign_fid(&lli->lli_id, &body->id1);
2012
2013         if (body->valid & OBD_MD_FLID)
2014                 id_ino(&lli->lli_id) = id_ino(&body->id1);
2015
2016         if (body->valid & OBD_MD_FLGENER)
2017                 id_gen(&lli->lli_id) = id_gen(&body->id1);
2018
2019         /* local/remote ACL */
2020         if (sbi->ll_remote) {
2021                 LASSERT(md->posix_acl == NULL);
2022                 if (md->remote_perm) {
2023                         ll_remote_acl_update(inode, md->remote_perm);
2024                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2025                         md->remote_perm = NULL;
2026                 }
2027         } else {
2028                 LASSERT(md->remote_perm == NULL);
2029                 spin_lock(&lli->lli_lock);
2030                 if (posix_acl != NULL) {
2031                         if (lli->lli_posix_acl != NULL)
2032                                 posix_acl_release(lli->lli_posix_acl);
2033                         lli->lli_posix_acl = posix_acl;
2034                 }
2035                 spin_unlock(&lli->lli_lock);
2036         }
2037
2038         if (body->valid & OBD_MD_FLID)
2039                 inode->i_ino = id_ino(&body->id1);
2040         if (body->valid & OBD_MD_FLGENER)
2041                 inode->i_generation = id_gen(&body->id1);
2042         if (body->valid & OBD_MD_FLATIME)
2043                 LTIME_S(inode->i_atime) = body->atime;
2044         if (body->valid & OBD_MD_FLMTIME &&
2045             body->mtime > LTIME_S(inode->i_mtime)) {
2046                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2047                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2048                 LTIME_S(inode->i_mtime) = body->mtime;
2049         }
2050         if (body->valid & OBD_MD_FLCTIME &&
2051             body->ctime > LTIME_S(inode->i_ctime))
2052                 LTIME_S(inode->i_ctime) = body->ctime;
2053         if (body->valid & OBD_MD_FLMODE) {
2054                 inode->i_mode = (inode->i_mode & S_IFMT) |
2055                         (body->mode & ~S_IFMT);
2056         }
2057         if (body->valid & OBD_MD_FLTYPE) {
2058                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2059                         (body->mode & S_IFMT);
2060         }
2061         if (body->valid & OBD_MD_FLUID)
2062                 inode->i_uid = body->uid;
2063         if (body->valid & OBD_MD_FLGID)
2064                 inode->i_gid = body->gid;
2065         if (body->valid & OBD_MD_FLFLAGS)
2066                 inode->i_flags = body->flags;
2067         if (body->valid & OBD_MD_FLNLINK)
2068                 inode->i_nlink = body->nlink;
2069         if (body->valid & OBD_MD_FLRDEV)
2070 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2071                 inode->i_rdev = body->rdev;
2072 #else
2073                 inode->i_rdev = old_decode_dev(body->rdev);
2074 #endif
2075         if (body->valid & OBD_MD_FLSIZE)
2076                 inode->i_size = body->size;
2077         if (body->valid & OBD_MD_FLBLOCKS)
2078                 inode->i_blocks = body->blocks;
2079
2080         if (body->valid & OBD_MD_FLSIZE)
2081                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2082
2083         if (body->valid & OBD_MD_FLAUDIT) {
2084                 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2085                 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2086                         sbi->ll_audit_mask = body->audit;
2087                 else
2088                         lli->lli_audit_mask = body->audit;
2089         }
2090
2091         if (mkey != NULL) {
2092                 LASSERT(body->valid & OBD_MD_FLKEY);
2093                 ll_crypto_init_inode_key(inode, mkey);
2094         }
2095
2096 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2097         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2098 #endif
2099         LASSERT(id_fid(&lli->lli_id) != 0);
2100 }
2101
2102 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2103 static struct backing_dev_info ll_backing_dev_info = {
2104         .ra_pages       = 0,    /* No readahead */
2105         .memory_backed  = 0,    /* Does contribute to dirty memory */
2106 };
2107 #endif
2108
2109 void ll_read_inode2(struct inode *inode, void *opaque)
2110 {
2111         struct lustre_md *md = opaque;
2112         struct ll_inode_info *lli = ll_i2info(inode);
2113         ENTRY;
2114
2115         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2116                inode->i_generation, inode);
2117
2118         ll_lli_init(lli);
2119
2120         LASSERT(!lli->lli_smd);
2121
2122         if (ll_i2sbi(inode)->ll_remote) {
2123                 lli->lli_remote_acl = remote_acl_alloc();
2124                 /* if failed alloc, nobody will be able to access this inode */
2125         }
2126
2127         /* Core attributes from the MDS first.  This is a new inode, and
2128          * the VFS doesn't zero times in the core inode so we have to do
2129          * it ourselves.  They will be overwritten by either MDS or OST
2130          * attributes - we just need to make sure they aren't newer. */
2131         LTIME_S(inode->i_mtime) = 0;
2132         LTIME_S(inode->i_atime) = 0;
2133         LTIME_S(inode->i_ctime) = 0;
2134
2135         inode->i_rdev = 0;
2136         ll_update_inode(inode, md);
2137
2138         /* OIDEBUG(inode); */
2139
2140         if (S_ISREG(inode->i_mode)) {
2141                 inode->i_op = &ll_file_inode_operations;
2142                 inode->i_fop = &ll_file_operations;
2143                 inode->i_mapping->a_ops = &ll_aops;
2144                 EXIT;
2145         } else if (S_ISDIR(inode->i_mode)) {
2146                 inode->i_op = &ll_dir_inode_operations;
2147                 inode->i_fop = &ll_dir_operations;
2148                 inode->i_mapping->a_ops = &ll_dir_aops;
2149                 EXIT;
2150         } else if (S_ISLNK(inode->i_mode)) {
2151                 inode->i_op = &ll_fast_symlink_inode_operations;
2152                 EXIT;
2153         } else {
2154                 inode->i_op = &ll_special_inode_operations;
2155
2156 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2157                 init_special_inode(inode, inode->i_mode,
2158                                    kdev_t_to_nr(inode->i_rdev));
2159
2160                 /* initializing backing dev info. */
2161                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2162 #else
2163                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2164 #endif
2165                 lli->ll_save_ifop = inode->i_fop;
2166
2167                 if (S_ISCHR(inode->i_mode))
2168                         inode->i_fop = &ll_special_chr_inode_fops;
2169                 else if (S_ISBLK(inode->i_mode))
2170                         inode->i_fop = &ll_special_blk_inode_fops;
2171                 else if (S_ISFIFO(inode->i_mode))
2172                         inode->i_fop = &ll_special_fifo_inode_fops;
2173                 else if (S_ISSOCK(inode->i_mode))
2174                         inode->i_fop = &ll_special_sock_inode_fops;
2175
2176                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2177                       inode->i_fop);
2178
2179                 if (lli->ll_save_ifop->owner) {
2180                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2181                               lli->ll_save_ifop->owner);
2182                 }
2183                 EXIT;
2184         }
2185 }
2186
2187 void ll_delete_inode(struct inode *inode)
2188 {
2189         struct ll_sb_info *sbi = ll_i2sbi(inode);
2190         struct lustre_id id;
2191         int rc;
2192         ENTRY;
2193
2194         ll_inode2id(&id, inode);
2195
2196         rc = md_delete_inode(sbi->ll_md_exp, &id);
2197         if (rc) {
2198                 CERROR("md_delete_inode() failed, error %d\n",
2199                        rc);
2200         }
2201
2202         clear_inode(inode);
2203         EXIT;
2204 }
2205
2206 int ll_iocontrol(struct inode *inode, struct file *file,
2207                  unsigned int cmd, unsigned long arg)
2208 {
2209         struct ll_sb_info *sbi = ll_i2sbi(inode);
2210         struct ptlrpc_request *req = NULL;
2211         int rc, flags = 0;
2212         ENTRY;
2213
2214         switch(cmd) {
2215         case EXT3_IOC_GETFLAGS: {
2216                 struct lustre_id id;
2217                 __u64 valid = OBD_MD_FLFLAGS;
2218                 struct mds_body *body;
2219
2220                 ll_inode2id(&id, inode);
2221                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2222                                 0, 0, NULL, &req);
2223                 if (rc) {
2224                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2225                         RETURN(-abs(rc));
2226                 }
2227
2228                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2229
2230                 if (body->flags & S_APPEND)
2231                         flags |= EXT3_APPEND_FL;
2232                 if (body->flags & S_IMMUTABLE)
2233                         flags |= EXT3_IMMUTABLE_FL;
2234                 if (body->flags & S_NOATIME)
2235                         flags |= EXT3_NOATIME_FL;
2236
2237                 ptlrpc_req_finished (req);
2238
2239                 RETURN(put_user(flags, (int *)arg));
2240         }
2241         case EXT3_IOC_SETFLAGS: {
2242                 struct mdc_op_data *op_data;
2243                 struct iattr attr;
2244                 struct obdo *oa;
2245                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2246
2247                 if (get_user(flags, (int *)arg))
2248                         RETURN(-EFAULT);
2249
2250                 oa = obdo_alloc();
2251                 if (!oa)
2252                         RETURN(-ENOMEM);
2253
2254                 OBD_ALLOC(op_data, sizeof(*op_data));
2255                 if (op_data == NULL) {
2256                         obdo_free(oa);
2257                         RETURN(-ENOMEM);
2258                 }
2259                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
2260
2261                 memset(&attr, 0x0, sizeof(attr));
2262                 attr.ia_attr_flags = flags;
2263                 attr.ia_valid |= ATTR_ATTR_FLAG;
2264
2265                 rc = md_setattr(sbi->ll_md_exp, op_data,
2266                                 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2267                 OBD_FREE(op_data, sizeof(*op_data));
2268                 if (rc) {
2269                         ptlrpc_req_finished(req);
2270                         if (rc != -EPERM && rc != -EACCES)
2271                                 CERROR("md_setattr fails: rc = %d\n", rc);
2272                         obdo_free(oa);
2273                         RETURN(rc);
2274                 }
2275                 ptlrpc_req_finished(req);
2276
2277                 oa->o_id = lsm->lsm_object_id;
2278                 oa->o_gr = lsm->lsm_object_gr;
2279                 oa->o_flags = flags;
2280                 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2281                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2282                               | OBD_MD_FLIFID;
2283
2284                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
2285                 obdo_free(oa);
2286                 if (rc) {
2287                         if (rc != -EPERM && rc != -EACCES)
2288                                 CERROR("md_setattr fails: rc = %d\n", rc);
2289                         RETURN(rc);
2290                 }
2291
2292                 if (flags & EXT3_APPEND_FL)
2293                         inode->i_flags |= S_APPEND;
2294                 else
2295                         inode->i_flags &= ~S_APPEND;
2296                 if (flags & EXT3_IMMUTABLE_FL)
2297                         inode->i_flags |= S_IMMUTABLE;
2298                 else
2299                         inode->i_flags &= ~S_IMMUTABLE;
2300                 if (flags & EXT3_NOATIME_FL)
2301                         inode->i_flags |= S_NOATIME;
2302                 else
2303                         inode->i_flags &= ~S_NOATIME;
2304
2305                 RETURN(0);
2306         }
2307         default:
2308                 RETURN(-ENOSYS);
2309         }
2310
2311         RETURN(0);
2312 }
2313
2314 /* this is only called in the case of forced umount. */
2315 void ll_umount_begin(struct super_block *sb)
2316 {
2317         struct ll_sb_info *sbi = ll_s2sbi(sb);
2318         struct obd_ioctl_data ioc_data = { 0 };
2319         struct obd_device *obd;
2320         ENTRY;
2321
2322         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2323                sb->s_count, atomic_read(&sb->s_active));
2324
2325         obd = class_exp2obd(sbi->ll_md_exp);
2326         if (obd == NULL) {
2327                 CERROR("Invalid MDC connection handle "LPX64"\n",
2328                        sbi->ll_md_exp->exp_handle.h_cookie);
2329                 EXIT;
2330                 return;
2331         }
2332         obd->obd_no_recov = 1;
2333         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2334                       sizeof(ioc_data), &ioc_data, NULL);
2335
2336         obd = class_exp2obd(sbi->ll_dt_exp);
2337         if (obd == NULL) {
2338                 CERROR("Invalid LOV connection handle "LPX64"\n",
2339                        sbi->ll_dt_exp->exp_handle.h_cookie);
2340                 EXIT;
2341                 return;
2342         }
2343
2344         obd->obd_no_recov = 1;
2345         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2346                       sizeof(ioc_data), &ioc_data, NULL);
2347
2348         /*
2349          * really, we'd like to wait until there are no requests outstanding,
2350          * and then continue.  For now, we just invalidate the requests,
2351          * schedule, and hope.
2352          */
2353         schedule();
2354
2355         EXIT;
2356 }
2357
2358 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2359                   struct inode **inode, struct ptlrpc_request *req,
2360                   int offset, struct super_block *sb)
2361 {
2362         struct lustre_md md;
2363         int rc = 0;
2364
2365         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2366         if (rc)
2367                 RETURN(rc);
2368
2369         if (*inode) {
2370                 ll_update_inode(*inode, &md);
2371         } else {
2372                 LASSERT(sb);
2373                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2374                 if (*inode == NULL || is_bad_inode(*inode)) {
2375                         /* free the lsm if we allocated one above */
2376                         if (md.lsm != NULL)
2377                                 obd_free_memmd(dt_exp, &md.lsm);
2378                         if (md.mea != NULL)
2379                                 obd_free_memmd(md_exp,
2380                                                (struct lov_stripe_md**)&md.mea);
2381                         rc = -ENOMEM;
2382                         CERROR("new_inode -fatal: rc %d\n", rc);
2383                 }
2384         }
2385
2386         RETURN(rc);
2387 }
2388
2389 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2390 {
2391         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2392         struct lustre_mount_data *lmd = sbi->ll_lmd;
2393
2394         if (lmd) {
2395                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2396                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2397         }
2398         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2399         if (sbi->ll_remote && lmd)
2400                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2401
2402         if (lmd && lmd->lmd_pag)
2403                 seq_printf(m, ",pag");
2404
2405         return 0;
2406 }
2407
2408 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2409                char *filename, struct lustre_id *ret)
2410 {
2411         struct ptlrpc_request *request = NULL;
2412         struct mds_body *body;
2413         int rc;
2414
2415         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2416                              OBD_MD_FID, 0, &request);
2417         if (rc < 0) {
2418                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2419                        filename, rc);
2420                 return rc;
2421         }
2422
2423         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2424         LASSERT(body != NULL);
2425         LASSERT_REPSWABBED(request, 0);
2426
2427         *ret = body->id1;
2428         ptlrpc_req_finished(request);
2429
2430         return rc;
2431 }
2432 int ll_flush_cred(struct inode *inode)
2433 {
2434         struct ll_sb_info *sbi = ll_i2sbi(inode);
2435         int rc = 0;
2436
2437         /* XXX to avoid adding api, we simply use set_info() interface
2438          * to notify underlying obds. set_info() is more like a ioctl() now...
2439          */
2440         if (sbi->ll_md_exp) {
2441                 rc = obd_set_info(sbi->ll_md_exp,
2442                                   strlen("flush_cred"), "flush_cred",
2443                                   0, NULL);
2444                 if (rc)
2445                         return rc;
2446         }
2447
2448         if (sbi->ll_dt_exp) {
2449                 rc = obd_set_info(sbi->ll_dt_exp,
2450                                   strlen("flush_cred"), "flush_cred",
2451                                   0, NULL);
2452                 if (rc)
2453                         return rc;
2454         }
2455
2456         return rc;
2457 }