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         /* FIXME Because of the async nature of file i/o, we never know
 267          * who is actually dirty the pages; and any process have chance
 268          * to trigger dirty-flushing within its own process context. So
 269          * for simplicity we simply use root's credential, we suppose root
 270          * always have credential.
 271          */
 272         if (pag)
 273                 sec_flags = PTLRPC_SEC_FL_PAG;
 274         else
 275                 sec_flags = PTLRPC_SEC_FL_OSS;
 276
 277         err = obd_set_info(obd->obd_self_export,
 278                            strlen("sec_flags"), "sec_flags",
 279                            sizeof(sec_flags), &sec_flags);
 280         if (err) {
 281                 OBD_FREE(data, sizeof(*data));
 282                 RETURN(err);
 283         }
 284
 285         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
 286         if (err == -EBUSY) {
 287                 CERROR("An OST (lov %s) is performing recovery, of which this"
 288                        " client is not a part.  Please wait for recovery to "
 289                        "complete, abort, or time out.\n", lov);
 290                 GOTO(out, err);
 291         } else if (err) {
 292                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 293                 GOTO(out, err);
 294         }
 295         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 296
 297         err = lustre_init_dt_desc(sbi);
 298
 299         if (err) {
 300                 CWARN("init dt_desc error %d \n", err);
 301                 GOTO(out, err = 0);
 302         }
 303         mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 304         obd_init_ea_size(sbi->ll_md_exp, mdsize, sbi->ll_dt_desc.ld_tgt_count *
 305                          sizeof(struct llog_cookie));
 306 out:
 307         RETURN(err);
 308 }
 309
 310 extern struct dentry_operations ll_d_ops;
 311
 312 static int lustre_init_root_inode(struct super_block *sb)
 313 {
 314         struct ll_sb_info *sbi = ll_s2sbi(sb);
 315         struct ptlrpc_request *request = NULL;
 316         struct inode *root = NULL;
 317         struct lustre_md md;
 318         int err = 0;
 319         ENTRY;
 320
 321         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 322         if (err) {
 323                 CERROR("cannot mds_connect: rc = %d\n", err);
 324                 GOTO(out, err);
 325         }
 326         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 327
 328         sb->s_op = &lustre_super_operations;
 329
 330         /* make root inode */
 331         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 332                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
 333                          NULL, NULL, 0, 0, NULL, &request);
 334         if (err) {
 335                 CERROR("md_getattr failed for root: rc = %d\n", err);
 336                 GOTO(out, err);
 337         }
 338
 339         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 340                                 sbi->ll_dt_exp, &md);
 341         if (err) {
 342                 CERROR("failed to understand root inode md: rc = %d\n", err);
 343                 ptlrpc_req_finished(request);
 344                 GOTO(out, err);
 345         }
 346
 347         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 348         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 349
 350         ptlrpc_req_finished(request);
 351
 352         if (root == NULL || is_bad_inode(root)) {
 353                 if (md.lsm != NULL)
 354                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 355                 if (md.mea != NULL)
 356                     obd_free_memmd(sbi->ll_md_exp,
 357                                    (struct lov_stripe_md**)&md.mea);
 358                 CERROR("lustre_lite: bad iget4 for root\n");
 359                 GOTO(out_root, err = -EBADF);
 360         }
 361         sb->s_root = d_alloc_root(root);
 362         sb->s_root->d_op = &ll_d_ops;
 363 out_root:
 364         if (err)
 365                 iput(root);
 366 out:
 367         RETURN(err);
 368 }
 369
 370 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 371                              char *gkc, int async,  char *mds_security,
 372                              char *oss_security, __u32 *nllu, int pag,
 373                              __u64 *remote)
 374 {
 375         struct ll_sb_info *sbi = ll_s2sbi(sb);
 376         struct obd_connect_data *data;
 377         int err;
 378         ENTRY;
 379
 380         /*process the connect flags*/
 381         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 382                        (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 383                 CERROR("wrong remote flag "LPX64"\n", *remote);
 384                 RETURN(-EINVAL);
 385         }
 386
 387         OBD_ALLOC(data, sizeof(*data));
 388         if (!data)
 389                 RETURN(-ENOMEM);
 390
 391         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 392                                               OBD_CONNECT_REMOTE);
 393         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 394
 395         if (proc_lustre_fs_root) {
 396                 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
 397                                                   sb, lov, lmv);
 398                 if (err < 0)
 399                         CERROR("could not register mount in /proc/lustre");
 400         }
 401
 402         /*connect mds */
 403         err = lustre_connect_mds(sb, lmv, data, mds_security, async, pag);
 404         if (err)
 405                 GOTO(out, err);
 406
 407         /*connect OST*/
 408         err = lustre_connect_ost(sb, lov, data, oss_security, async, pag);
 409         if (err)
 410                 GOTO(out_lmv, err);
 411
 412         /*connect GSS*/
 413         err = lustre_init_crypto(sb, gkc, data, async);
 414         if (err) {
 415                 CERROR("Could not connect to GSS err %d\n", err);
 416                 err = 0;
 417         }
 418
 419         err = lustre_init_root_inode(sb);
 420         if (err)
 421                 GOTO(out_gks, err);
 422
 423         err = ll_close_thread_start(&sbi->ll_lcq);
 424         if (err) {
 425                 CERROR("cannot start close thread: rc %d\n", err);
 426                 GOTO(out_root, err);
 427         }
 428
 429         ll_gns_add_timer(sbi);
 430
 431         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 432            backing dev info assigned to inode mapping is used for
 433            determining maximal readahead. */
 434 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 435     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 436         /* bug 2805 - set VM readahead to zero */
 437         vm_max_readahead = vm_min_readahead = 0;
 438 #endif
 439         sb->s_flags |= MS_POSIXACL;
 440 #ifdef S_PDIROPS
 441         CWARN("Enabling PDIROPS\n");
 442         sb->s_flags |= S_PDIROPS;
 443 #endif
 444         if (data != NULL)
 445                 OBD_FREE(data, sizeof(*data));
 446         RETURN(err);
 447 out_root:
 448         if (sb->s_root)
 449                 dput(sb->s_root);
 450 out_gks:
 451         lustre_destroy_crypto(sb);
 452 out_lmv:
 453         obd_disconnect(sbi->ll_md_exp, 0);
 454 out:
 455         if (data != NULL)
 456                 OBD_FREE(data, sizeof(*data));
 457         lprocfs_unregister_mountpoint(sbi);
 458         RETURN(err);
 459 }
 460
 461 void lustre_common_put_super(struct super_block *sb)
 462 {
 463         struct ll_sb_info *sbi = ll_s2sbi(sb);
 464         struct hlist_node *tmp, *next;
 465         ENTRY;
 466
 467         ll_gns_del_timer(sbi);
 468         ll_close_thread_stop(sbi->ll_lcq);
 469
 470         list_del(&sbi->ll_conn_chain);
 471         obd_disconnect(sbi->ll_dt_exp, 0);
 472
 473         lustre_destroy_crypto(sb);
 474
 475         lprocfs_unregister_mountpoint(sbi);
 476         if (sbi->ll_proc_root) {
 477                 lprocfs_remove(sbi->ll_proc_root);
 478                 sbi->ll_proc_root = NULL;
 479         }
 480
 481         obd_disconnect(sbi->ll_md_exp, 0);
 482
 483         // We do this to get rid of orphaned dentries. That is not really trw.
 484         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 485                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 486                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 487                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 488                 shrink_dcache_parent(dentry);
 489         }
 490         EXIT;
 491 }
 492
 493 char *ll_read_opt(const char *opt, char *data)
 494 {
 495         char *value;
 496         char *retval;
 497         ENTRY;
 498
 499         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 500         if (strncmp(opt, data, strlen(opt)))
 501                 RETURN(NULL);
 502         if ((value = strchr(data, '=')) == NULL)
 503                 RETURN(NULL);
 504
 505         value++;
 506         OBD_ALLOC(retval, strlen(value) + 1);
 507         if (!retval) {
 508                 CERROR("out of memory!\n");
 509                 RETURN(NULL);
 510         }
 511
 512         memcpy(retval, value, strlen(value)+1);
 513         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 514         RETURN(retval);
 515 }
 516
 517 int ll_set_opt(const char *opt, char *data, int fl)
 518 {
 519         ENTRY;
 520
 521         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 522         if (strncmp(opt, data, strlen(opt)))
 523                 RETURN(0);
 524         else
 525                 RETURN(fl);
 526 }
 527
 528 void ll_options(char *options, char **lov, char **lmv, char **gkc,
 529                 char **mds_sec, char **oss_sec, int *async, int *flags)
 530 {
 531         char *this_char;
 532 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 533         char *opt_ptr = options;
 534 #endif
 535         ENTRY;
 536
 537         if (!options) {
 538                 EXIT;
 539                 return;
 540         }
 541
 542         *async = 0;
 543 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 544         for (this_char = strtok (options, ",");
 545              this_char != NULL;
 546              this_char = strtok (NULL, ",")) {
 547 #else
 548         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 549 #endif
 550                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 551                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 552                         continue;
 553                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 554                         continue;
 555                 if (!*gkc && (*gkc = ll_read_opt("gkc", this_char)))
 556                         continue;
 557                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 558                         *async = 1;
 559                         continue;
 560                 }
 561                 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
 562                         continue;
 563                 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
 564                         continue;
 565                 if (!(*flags & LL_SBI_NOLCK) &&
 566                     ((*flags) = (*flags) |
 567                                 ll_set_opt("nolock", this_char,
 568                                            LL_SBI_NOLCK)))
 569                         continue;
 570         }
 571
 572         EXIT;
 573 }
 574
 575 void ll_lli_init(struct ll_inode_info *lli)
 576 {
 577         sema_init(&lli->lli_open_sem, 1);
 578         sema_init(&lli->lli_size_sem, 1);
 579         lli->lli_flags = 0;
 580         lli->lli_size_pid = 0;
 581         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 582         spin_lock_init(&lli->lli_lock);
 583         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 584         INIT_LIST_HEAD(&lli->lli_close_item);
 585         lli->lli_inode_magic = LLI_INODE_MAGIC;
 586         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 587         sema_init(&lli->lli_och_sem, 1);
 588         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 589         lli->lli_mds_exec_och = NULL;
 590         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 591         lli->lli_open_fd_exec_count = 0;
 592         lli->lli_audit_mask = AUDIT_OFF;
 593         lli->lli_key_info = NULL;
 594         init_waitqueue_head(&lli->lli_dirty_wait);
 595         lli->lli_io_epoch = 0;
 596         INIT_LIST_HEAD(&lli->lli_capas);
 597 }
 598
 599 int ll_fill_super(struct super_block *sb, void *data, int silent)
 600 {
 601         struct ll_sb_info *sbi;
 602         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 603         char *mds_sec = NULL;
 604         char *oss_sec = NULL;
 605         int async, err;
 606         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 607         __u64 remote_flag = 0;
 608         ENTRY;
 609
 610         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 611
 612         sbi = lustre_init_sbi(sb);
 613         if (!sbi)
 614                 RETURN(-ENOMEM);
 615
 616         sbi->ll_flags |= LL_SBI_READAHEAD;
 617         ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
 618                    &async, &sbi->ll_flags);
 619
 620         if (!lov || !lmv) {
 621                 CERROR("no osc %p or no mdc %p\n", lov, lmv);
 622                 GOTO(out, err = -EINVAL);
 623         }
 624
 625         err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
 626                                        oss_sec, nllu, 0, &remote_flag);
 627         EXIT;
 628 out:
 629         if (err)
 630                 lustre_free_sbi(sb);
 631
 632         if (lmv)
 633                 OBD_FREE(lmv, strlen(lmv) + 1);
 634         if (lov)
 635                 OBD_FREE(lov, strlen(lov) + 1);
 636         if (mds_sec)
 637                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 638         if (oss_sec)
 639                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 640         if (gkc)
 641                 OBD_FREE(gkc, strlen(gkc) + 1);
 642
 643         return err;
 644 } /* ll_read_super */
 645
 646 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 647                               struct config_llog_instance *cfg, int allow_recov)
 648 {
 649         struct lustre_cfg *lcfg = NULL;
 650         struct lustre_cfg_bufs bufs;
 651         struct portals_cfg pcfg;
 652         char *peer = "MDS_PEER_UUID";
 653         struct obd_device *obd;
 654         struct lustre_handle md_conn = {0, };
 655         struct obd_export *exp;
 656         char *name = "mdc_dev";
 657         class_uuid_t uuid;
 658         struct obd_uuid lmv_uuid;
 659         struct llog_ctxt *ctxt;
 660         int rc = 0, err = 0;
 661         ENTRY;
 662
 663         if (lmd_bad_magic(lmd))
 664                 RETURN(-EINVAL);
 665
 666         generate_random_uuid(uuid);
 667         class_uuid_unparse(uuid, &lmv_uuid);
 668
 669         if (lmd->lmd_local_nid) {
 670                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 671                 pcfg.pcfg_nal = lmd->lmd_nal;
 672                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 673                 rc = libcfs_nal_cmd(&pcfg);
 674                 if (rc < 0)
 675                         GOTO(out, rc);
 676         }
 677
 678         if (lmd->lmd_nal == SOCKNAL ||
 679             lmd->lmd_nal == OPENIBNAL ||
 680             lmd->lmd_nal == IIBNAL ||
 681             lmd->lmd_nal == VIBNAL ||
 682             lmd->lmd_nal == RANAL) {
 683                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 684                 pcfg.pcfg_nal     = lmd->lmd_nal;
 685                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 686                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 687                 pcfg.pcfg_misc    = lmd->lmd_port;
 688                 rc = libcfs_nal_cmd(&pcfg);
 689                 if (rc < 0)
 690                         GOTO(out, rc);
 691         }
 692         lustre_cfg_bufs_reset(&bufs, name);
 693         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 694
 695         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 696         lcfg->lcfg_nal = lmd->lmd_nal;
 697         lcfg->lcfg_nid = lmd->lmd_server_nid;
 698         LASSERT(lcfg->lcfg_nal);
 699         LASSERT(lcfg->lcfg_nid);
 700         err = class_process_config(lcfg);
 701         lustre_cfg_free(lcfg);
 702         if (err < 0)
 703                 GOTO(out_del_conn, err);
 704
 705         lustre_cfg_bufs_reset(&bufs, name);
 706         lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
 707         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 708
 709         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 710         err = class_process_config(lcfg);
 711         lustre_cfg_free(lcfg);
 712         if (err < 0)
 713                 GOTO(out_del_uuid, err);
 714
 715         lustre_cfg_bufs_reset(&bufs, name);
 716         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 717         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 718
 719         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 720         err = class_process_config(lcfg);
 721         lustre_cfg_free(lcfg);
 722         if (err < 0)
 723                 GOTO(out_detach, err);
 724
 725         obd = class_name2obd(name);
 726         if (obd == NULL)
 727                 GOTO(out_cleanup, rc = -EINVAL);
 728
 729         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 730                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 731         if (rc)
 732                 GOTO(out_cleanup, rc);
 733
 734         if (lmd->lmd_pag) {
 735                 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
 736                 rc = obd_set_info(obd->obd_self_export,
 737                                   strlen("sec_flags"), "sec_flags",
 738                                   sizeof(sec_flags), &sec_flags);
 739                 if (rc)
 740                         GOTO(out_cleanup, rc);
 741         }
 742
 743         /* Disable initial recovery on this import */
 744         rc = obd_set_info(obd->obd_self_export,
 745                           strlen("initial_recov"), "initial_recov",
 746                           sizeof(allow_recov), &allow_recov);
 747         if (rc)
 748                 GOTO(out_cleanup, rc);
 749
 750         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, OBD_OPT_REAL_CLIENT);
 751         if (rc) {
 752                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 753                 GOTO(out_cleanup, rc);
 754         }
 755
 756         exp = class_conn2export(&md_conn);
 757
 758         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 759         rc = class_config_process_llog(ctxt, profile, cfg);
 760         if (rc)
 761                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 762
 763         err = obd_disconnect(exp, 0);
 764
 765         EXIT;
 766 out_cleanup:
 767         lustre_cfg_bufs_reset(&bufs, name);
 768         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 769         err = class_process_config(lcfg);
 770         lustre_cfg_free(lcfg);
 771         if (err < 0)
 772                 GOTO(out, err);
 773 out_detach:
 774         lustre_cfg_bufs_reset(&bufs, name);
 775         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 776         err = class_process_config(lcfg);
 777         lustre_cfg_free(lcfg);
 778         if (err < 0)
 779                 GOTO(out, err);
 780
 781 out_del_uuid:
 782         lustre_cfg_bufs_reset(&bufs, name);
 783         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 784         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 785         err = class_process_config(lcfg);
 786         lustre_cfg_free(lcfg);
 787
 788 out_del_conn:
 789         if (lmd->lmd_nal == SOCKNAL ||
 790             lmd->lmd_nal == OPENIBNAL ||
 791             lmd->lmd_nal == IIBNAL ||
 792             lmd->lmd_nal == VIBNAL ||
 793             lmd->lmd_nal == RANAL) {
 794                 int err2;
 795
 796                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 797                 pcfg.pcfg_nal     = lmd->lmd_nal;
 798                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 799                 pcfg.pcfg_flags   = 1;          /* single_share */
 800                 err2 = libcfs_nal_cmd(&pcfg);
 801                 if (err2 && !err)
 802                         err = err2;
 803                 if (err < 0)
 804                         GOTO(out, err);
 805         }
 806 out:
 807         if (rc == 0)
 808                 rc = err;
 809
 810         return rc;
 811 }
 812
 813 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 814 {
 815         struct lustre_cfg *lcfg;
 816         struct lustre_cfg_bufs bufs;
 817         struct obd_device *obd;
 818         int next = 0;
 819
 820         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
 821                 int err;
 822
 823                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 824                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 825                 err = class_process_config(lcfg);
 826                 if (err) {
 827                         CERROR("cleanup failed: %s\n", obd->obd_name);
 828                         //continue;
 829                 }
 830
 831                 lcfg->lcfg_command = LCFG_DETACH;
 832                 err = class_process_config(lcfg);
 833                 lustre_cfg_free(lcfg);
 834                 if (err) {
 835                         CERROR("detach failed: %s\n", obd->obd_name);
 836                         //continue;
 837                 }
 838         }
 839
 840         if (sbi->ll_lmd != NULL)
 841                 class_del_profile(sbi->ll_lmd->lmd_profile);
 842 }
 843
 844 static int lustre_process_profile(struct super_block *sb,
 845                                   struct lustre_mount_data *lmd,
 846                                   char **lov, char **lmv, char **gkc)
 847 {
 848         struct ll_sb_info *sbi = ll_s2sbi(sb);
 849         struct config_llog_instance cfg;
 850         struct lustre_profile *lprof;
 851         int len, err = 0;
 852         ENTRY;
 853
 854         if (!lmd->lmd_profile)
 855                 RETURN(0);
 856
 857         if (lmd->lmd_mds[0] == '\0') {
 858                 CERROR("no mds name\n");
 859                 GOTO(out, err = -EINVAL);
 860         }
 861         lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
 862         lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
 863
 864         OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 865         if (sbi->ll_lmd == NULL)
 866                 GOTO(out, err = -ENOMEM);
 867         memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 868
 869         /* generate a string unique to this super, let's try the address of the
 870          * super itself. */
 871         len = (sizeof(sb) * 2) + 1;
 872         OBD_ALLOC(sbi->ll_instance, len);
 873         if (sbi->ll_instance == NULL)
 874                 GOTO(out, err = -ENOMEM);
 875         sprintf(sbi->ll_instance, "%p", sb);
 876
 877         cfg.cfg_instance = sbi->ll_instance;
 878         cfg.cfg_uuid = sbi->ll_sb_uuid;
 879         cfg.cfg_local_nid = lmd->lmd_local_nid;
 880         cfg.cfg_flags |= CFG_MODIFY_UUID_FL;
 881         err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 882         if (err < 0) {
 883                 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 884                 GOTO(out, err);
 885         }
 886
 887         lprof = class_get_profile(lmd->lmd_profile);
 888         if (lprof == NULL) {
 889                 CERROR("No profile found: %s\n", lmd->lmd_profile);
 890                 GOTO(out, err = -EINVAL);
 891         }
 892
 893         OBD_ALLOC(*lov, strlen(lprof->lp_lov) +
 894                   strlen(sbi->ll_instance) + 2);
 895         if (*lov == NULL)
 896                 GOTO(out, err = -ENOMEM);
 897
 898         sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 899
 900         OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
 901                   strlen(sbi->ll_instance) + 2);
 902         if (*lmv == NULL)
 903                 GOTO(out_free_lov, err = -ENOMEM);
 904
 905         sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 906
 907         if (lprof->lp_gkc) {
 908                OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
 909                          strlen(sbi->ll_instance) + 2);
 910                if (*gkc == NULL)
 911                        GOTO(out_free_lmv, err = -ENOMEM);
 912
 913                sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
 914         }
 915
 916         RETURN(err);
 917 out_free_lmv:
 918         OBD_FREE(*lmv, strlen(lprof->lp_lmv) +
 919                  strlen(sbi->ll_instance) + 2);
 920 out_free_lov:
 921         OBD_FREE(*lov, strlen(lprof->lp_lov) +
 922                  strlen(sbi->ll_instance) + 2);
 923 out:
 924         return err;
 925 }
 926
 927 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
 928 {
 929         struct lustre_mount_data *lmd = sbi->ll_lmd;
 930         struct config_llog_instance cfg;
 931         char *cl_prof;
 932         int len, err = 0;
 933         ENTRY;
 934
 935         if (!lmd)
 936                 RETURN(err);
 937
 938         len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
 939
 940         if (force_umount) {
 941                 CERROR("force umount, doing manual cleanup\n");
 942                 lustre_manual_cleanup(sbi);
 943                 GOTO(free_lmd, 0);
 944
 945         }
 946         if (sbi->ll_instance != NULL) {
 947                 cfg.cfg_flags |= CFG_MODIFY_UUID_FL;
 948                 cfg.cfg_instance = sbi->ll_instance;
 949                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 950
 951                 OBD_ALLOC(cl_prof, len);
 952                 if (!cl_prof) {
 953                         CERROR("can't allocate memory, "
 954                                "skipping processing cleanup profile.\n");
 955                         GOTO(free_lmd, err = -ENOMEM);
 956                 }
 957
 958                 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 959                 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 960                 if (err < 0) {
 961                         CERROR("Unable to process log: %s\n", cl_prof);
 962                         lustre_manual_cleanup(sbi);
 963                 }
 964                 OBD_FREE(cl_prof, len);
 965         }
 966         EXIT;
 967 free_lmd:
 968         if (sbi->ll_instance)
 969                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 970         OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 971         return err;
 972 }
 973
 974 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 975 {
 976         struct lustre_mount_data * lmd = data;
 977         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 978         struct ll_sb_info *sbi;
 979         int err;
 980         ENTRY;
 981
 982         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 983         if (lmd_bad_magic(lmd))
 984                 RETURN(-EINVAL);
 985
 986         sbi = lustre_init_sbi(sb);
 987         if (!sbi)
 988                 RETURN(-ENOMEM);
 989
 990         sbi->ll_flags |= LL_SBI_READAHEAD;
 991
 992         err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
 993         if (err) {
 994                 CERROR("Can not process the profile err %d \n", err);
 995                 GOTO(out_free, err);
 996         }
 997         if (!lov || !lmv) {
 998                 CERROR("no osc %p or no mdc %p \n", lov, lmv);
 999                 GOTO(out_free, err = -EINVAL);
1000         }
1001
1002         err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
1003                                        lmd->lmd_mds_security,
1004                                        lmd->lmd_oss_security,
1005                                        &lmd->lmd_nllu, lmd->lmd_pag,
1006                                        &lmd->lmd_remote_flag);
1007
1008         if (err)
1009                 GOTO(out_free, err);
1010
1011         EXIT;
1012 out_dev:
1013         if (lmv)
1014                 OBD_FREE(lmv, strlen(lmv) + 1);
1015         if (lov)
1016                 OBD_FREE(lov, strlen(lov) + 1);
1017         if (gkc)
1018                 OBD_FREE(gkc, strlen(gkc) + 1);
1019
1020         return err;
1021 out_free:
1022         lustre_clean_profile(sbi, 0);
1023         lustre_free_sbi(sb);
1024         goto out_dev;
1025
1026 } /* lustre_fill_super */
1027
1028 void lustre_put_super(struct super_block *sb)
1029 {
1030         struct obd_device *obd;
1031         struct ll_sb_info *sbi = ll_s2sbi(sb);
1032         int force_umount = 0;
1033         ENTRY;
1034
1035         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1036         obd = class_exp2obd(sbi->ll_md_exp);
1037         if (obd)
1038                 force_umount = obd->obd_no_recov;
1039         obd = NULL;
1040
1041         lustre_common_put_super(sb);
1042         lustre_clean_profile(sbi, force_umount);
1043         lustre_free_sbi(sb);
1044
1045         EXIT;
1046 } /* lustre_put_super */
1047
1048 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1049 {
1050         struct lustre_mount_data *lmd = sbi->ll_lmd;
1051         char *profile = lmd->lmd_profile, *name = NULL;
1052         struct config_llog_instance cfg;
1053         int rc, namelen =  0, version;
1054         struct llog_ctxt *ctxt;
1055         ENTRY;
1056
1057         if (profile == NULL)
1058                 RETURN(0);
1059         if (lmd == NULL) {
1060                 CERROR("Client not mounted with zero-conf; cannot "
1061                        "process update log.\n");
1062                 RETURN(0);
1063         }
1064
1065         cfg.cfg_instance = sbi->ll_instance;
1066         cfg.cfg_uuid = sbi->ll_sb_uuid;
1067         cfg.cfg_local_nid = lmd->lmd_local_nid;
1068         cfg.cfg_flags |= CFG_MODIFY_UUID_FL;
1069         namelen = strlen(profile) + 20; /* -clean-######### */
1070         OBD_ALLOC(name, namelen);
1071         if (name == NULL)
1072                 RETURN(-ENOMEM);
1073
1074         if (clean) {
1075                 version = sbi->ll_config_version - 1;
1076                 sprintf(name, "%s-clean-%d", profile, version);
1077         } else {
1078                 version = sbi->ll_config_version + 1;
1079                 sprintf(name, "%s-%d", profile, version);
1080         }
1081
1082         CWARN("Applying configuration log %s\n", name);
1083
1084         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1085                                 LLOG_CONFIG_REPL_CTXT);
1086         rc = class_config_process_llog(ctxt, name, &cfg);
1087         if (rc == 0)
1088                 sbi->ll_config_version = version;
1089         CWARN("Finished applying configuration log %s: %d\n", name, rc);
1090
1091         if (rc == 0 && clean == 0) {
1092                 struct lov_desc desc;
1093                 __u32 valsize;
1094                 int rc = 0;
1095
1096                 valsize = sizeof(desc);
1097                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1098                                   "lovdesc", &valsize, &desc);
1099
1100                 rc = obd_init_ea_size(sbi->ll_md_exp,
1101                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
1102                                       (desc.ld_tgt_count *
1103                                        sizeof(struct llog_cookie)));
1104         }
1105         OBD_FREE(name, namelen);
1106         RETURN(rc);
1107 }
1108
1109 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1110 {
1111         struct inode *inode = NULL;
1112
1113         /* NOTE: we depend on atomic igrab() -bzzz */
1114         lock_res_and_lock(lock);
1115         if (lock->l_ast_data) {
1116                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1117                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1118                         inode = igrab(lock->l_ast_data);
1119                 } else {
1120                         struct timeval now;
1121                         do_gettimeofday(&now);
1122                         inode = lock->l_ast_data;
1123                         LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1124                                    lock->l_enqueued_time.tv_sec,
1125                                    lock->l_enqueued_time.tv_usec,
1126                                    now.tv_sec, now.tv_usec);
1127                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1128                                "l_ast_data %p is bogus: magic %0x8\n",
1129                                lock->l_ast_data, lli->lli_inode_magic);
1130                         CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1131                                inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1132                         inode = NULL;
1133                         unlock_res_and_lock(lock);
1134                         LBUG();
1135                 }
1136         }
1137         unlock_res_and_lock(lock);
1138         return inode;
1139 }
1140
1141 int null_if_equal(struct ldlm_lock *lock, void *data)
1142 {
1143         if (data == lock->l_ast_data) {
1144                 lock->l_ast_data = NULL;
1145
1146                 if (lock->l_req_mode != lock->l_granted_mode)
1147                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1148         }
1149
1150         return LDLM_ITER_CONTINUE;
1151 }
1152
1153 static void remote_acl_free(struct remote_acl *racl);
1154
1155 void ll_clear_inode(struct inode *inode)
1156 {
1157         struct lustre_id id;
1158         struct ll_inode_info *lli = ll_i2info(inode);
1159         struct ll_sb_info *sbi = ll_i2sbi(inode);
1160         struct obd_capa *ocapa, *tmp;
1161         ENTRY;
1162
1163         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1164                inode->i_generation, inode);
1165
1166         LASSERT(ll_is_inode_dirty(inode) == 0);
1167         ll_inode2id(&id, inode);
1168
1169         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1170         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1171
1172         LASSERT(!lli->lli_open_fd_write_count);
1173         LASSERT(!lli->lli_open_fd_read_count);
1174         LASSERT(!lli->lli_open_fd_exec_count);
1175         if (lli->lli_mds_write_och)
1176                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1177         if (lli->lli_mds_exec_och)
1178                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1179         if (lli->lli_mds_read_och)
1180                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1181         if (lli->lli_smd)
1182                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1183                                   null_if_equal, inode);
1184
1185         if (lli->lli_smd) {
1186                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1187                 lli->lli_smd = NULL;
1188         }
1189
1190         if (lli->lli_mea) {
1191                 obd_free_memmd(sbi->ll_md_exp,
1192                                (struct lov_stripe_md **) &lli->lli_mea);
1193                 lli->lli_mea = NULL;
1194         }
1195
1196         LASSERT(sbi->ll_crypto_info != NULL);
1197         ll_crypto_destroy_inode_key(inode);
1198
1199         if (lli->lli_symlink_name) {
1200                 OBD_FREE(lli->lli_symlink_name,
1201                          strlen(lli->lli_symlink_name) + 1);
1202                 lli->lli_symlink_name = NULL;
1203         }
1204
1205         if (lli->lli_posix_acl) {
1206                 LASSERT(lli->lli_remote_acl == NULL);
1207                 posix_acl_release(lli->lli_posix_acl);
1208                 lli->lli_posix_acl = NULL;
1209         }
1210
1211         if (lli->lli_remote_acl) {
1212                 LASSERT(lli->lli_posix_acl == NULL);
1213                 remote_acl_free(lli->lli_remote_acl);
1214                 lli->lli_remote_acl = NULL;
1215         }
1216
1217         list_for_each_entry_safe(ocapa, tmp, &lli->lli_capas, u.client.lli_list)
1218                 capa_put(ocapa);
1219
1220         LASSERT(!mapping_has_pages(inode->i_mapping));
1221         lli->lli_inode_magic = LLI_INODE_DEAD;
1222
1223         EXIT;
1224 }
1225
1226 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1227  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1228  * keep these values until such a time that objects are allocated for it.
1229  * We do the MDS operations first, as it is checking permissions for us.
1230  * We don't to the MDS RPC if there is nothing that we want to store there,
1231  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1232  * going to do an RPC anyways.
1233  *
1234  * If we are doing a truncate, we will send the mtime and ctime updates
1235  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1236  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1237  * at the same time.
1238  */
1239 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1240 {
1241         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1242         struct ll_inode_info *lli = ll_i2info(inode);
1243         struct ll_sb_info *sbi = ll_i2sbi(inode);
1244         struct ptlrpc_request *request = NULL;
1245         struct mdc_op_data *op_data;
1246         int ia_valid = attr->ia_valid;
1247         int err, rc = 0;
1248         ENTRY;
1249
1250         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1251         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1252
1253         if (ia_valid & ATTR_SIZE) {
1254                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1255                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1256                                attr->ia_size, ll_file_maxbytes(inode));
1257                         RETURN(-EFBIG);
1258                 }
1259
1260                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1261         }
1262
1263         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1264         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1265                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1266                         RETURN(-EPERM);
1267         }
1268
1269         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1270         if (attr->ia_valid & ATTR_CTIME) {
1271                 attr->ia_ctime = CURRENT_TIME;
1272                 attr->ia_valid |= ATTR_CTIME_SET;
1273         }
1274         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1275                 attr->ia_atime = CURRENT_TIME;
1276                 attr->ia_valid |= ATTR_ATIME_SET;
1277         }
1278         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1279                 attr->ia_mtime = CURRENT_TIME;
1280                 attr->ia_valid |= ATTR_MTIME_SET;
1281         }
1282
1283         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1284                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1285                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1286                        LTIME_S(CURRENT_TIME));
1287
1288         /* If only OST attributes being set on objects, don't do MDS RPC.
1289          * In that case, we need to check permissions and update the local
1290          * inode ourselves so we can call obdo_from_inode() always. */
1291         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1292                 struct lustre_md md;
1293                 void *key = NULL;
1294                 int  key_size = 0;
1295
1296                 OBD_ALLOC(op_data, sizeof(*op_data));
1297                 if (op_data == NULL)
1298                         RETURN(-ENOMEM);
1299                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
1300
1301                 if (ia_valid & (ATTR_UID | ATTR_GID | ATTR_MODE)) {
1302                         rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1303                                                &key_size);
1304                         if (rc) {
1305                                 CERROR("can not get right mac, rc=%d\n", rc);
1306                                 if (key && key_size)
1307                                         OBD_FREE(key, key_size);
1308                                 RETURN(rc);
1309                         }
1310                 }
1311                 rc = md_setattr(sbi->ll_md_exp, op_data,
1312                                 attr, key, key_size, NULL, 0, NULL,
1313                                 0, &request);
1314                 OBD_FREE(op_data, sizeof(*op_data));
1315
1316                 if (key && key_size)
1317                         OBD_FREE(key, key_size);
1318                 if (rc) {
1319                         ptlrpc_req_finished(request);
1320                         if (rc != -EPERM && rc != -EACCES)
1321                                 CERROR("md_setattr fails: rc = %d\n", rc);
1322                         RETURN(rc);
1323                 }
1324                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1325                                        sbi->ll_dt_exp, &md);
1326                 if (rc) {
1327                         ptlrpc_req_finished(request);
1328                         RETURN(rc);
1329                 }
1330
1331                 if (attr->ia_valid & ATTR_SIZE) {
1332                         rc = ll_set_trunc_capa(request, 0, inode);
1333                         if (rc) {
1334                                 ptlrpc_req_finished(request);
1335                                 RETURN(rc);
1336                         }
1337                 }
1338
1339                 /* We call inode_setattr to adjust timestamps, but we first
1340                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1341                  *
1342                  * NB: ATTR_SIZE will only be set at this point if the size
1343                  * resides on the MDS, ie, this file has no objects. */
1344                 attr->ia_valid &= ~ATTR_SIZE;
1345
1346                 /*
1347                  * assigning inode_setattr() to @err to disable warning that
1348                  * function's result should be checked by by caller. error is
1349                  * impossible here, as vmtruncate() control path is disabled.
1350                  */
1351                 err = inode_setattr(inode, attr);
1352                 ll_update_inode(inode, &md);
1353                 ptlrpc_req_finished(request);
1354
1355                 if (!lsm || !S_ISREG(inode->i_mode)) {
1356                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1357                         RETURN(0);
1358                 }
1359         } else {
1360                 /* The OST doesn't check permissions, but the alternative is
1361                  * a gratuitous RPC to the MDS.  We already rely on the client
1362                  * to do read/write/truncate permission checks, so is mtime OK?
1363                  */
1364                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1365                         /* from sys_utime() */
1366                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1367                                 if (current->fsuid != inode->i_uid &&
1368                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1369                                         RETURN(rc);
1370                         } else {
1371                                 /* from inode_change_ok() */
1372                                 if (current->fsuid != inode->i_uid &&
1373                                     !capable(CAP_FOWNER))
1374                                         RETURN(-EPERM);
1375                         }
1376                 }
1377
1378                 if (lsm)
1379                         attr->ia_valid &= ~ATTR_SIZE;
1380
1381                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1382                 err = inode_setattr(inode, attr);
1383                 /*
1384                  * assigning inode_setattr() to @err to disable warning that
1385                  * function's result should be checked by by caller. error is
1386                  * impossible here, as vmtruncate() control path is disabled.
1387                  */
1388         }
1389
1390         /* We really need to get our PW lock before we change inode->i_size.
1391          * If we don't we can race with other i_size updaters on our node, like
1392          * ll_file_read.  We can also race with i_size propogation to other
1393          * nodes through dirtying and writeback of final cached pages.  This
1394          * last one is especially bad for racing o_append users on other
1395          * nodes. */
1396         if (ia_valid & ATTR_SIZE) {
1397                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1398                                                            OBD_OBJECT_EOF } };
1399                 struct lustre_handle lockh = { 0 };
1400                 int err, ast_flags = 0;
1401                 /* XXX when we fix the AST intents to pass the discard-range
1402                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1403                  * XXX here. */
1404                 if (attr->ia_size == 0)
1405                         ast_flags = LDLM_AST_DISCARD_DATA;
1406
1407                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1408                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1409
1410                 if (rc != 0)
1411                         RETURN(rc);
1412
1413                 down(&lli->lli_size_sem);
1414                 lli->lli_size_pid = current->pid;
1415                 rc = vmtruncate(inode, attr->ia_size);
1416                 if (rc != 0) {
1417                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1418                         lli->lli_size_pid = 0;
1419                         up(&lli->lli_size_sem);
1420                 }
1421
1422                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1423                 if (err) {
1424                         CERROR("ll_extent_unlock failed: %d\n", err);
1425                         if (!rc)
1426                                 rc = err;
1427                 }
1428         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1429                 struct obdo *oa = NULL;
1430
1431                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1432                        inode->i_ino, LTIME_S(attr->ia_mtime));
1433
1434                 oa = obdo_alloc();
1435                 if (oa == NULL)
1436                         RETURN(-ENOMEM);
1437
1438                 oa->o_id = lsm->lsm_object_id;
1439                 oa->o_gr = lsm->lsm_object_gr;
1440                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1441
1442                 /* adding uid and gid, needed for quota */
1443                 if (ia_valid & ATTR_UID) {
1444                         oa->o_uid = inode->i_uid;
1445                         oa->o_valid |= OBD_MD_FLUID;
1446                 }
1447
1448                 if (ia_valid & ATTR_GID) {
1449                         oa->o_gid = inode->i_gid;
1450                         oa->o_valid |= OBD_MD_FLGID;
1451                 }
1452
1453                 *(obdo_id(oa)) = lli->lli_id;
1454                 oa->o_valid |= OBD_MD_FLIFID;
1455
1456                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1457                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1458                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL);
1459                 obdo_free(oa);
1460                 if (rc)
1461                         CERROR("obd_setattr fails: rc = %d\n", rc);
1462         }
1463
1464         RETURN(rc);
1465 }
1466
1467 int ll_setattr(struct dentry *de, struct iattr *attr)
1468 {
1469         LASSERT(de->d_inode);
1470         return ll_setattr_raw(de->d_inode, attr);
1471 }
1472
1473 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1474                        unsigned long max_age)
1475 {
1476         struct ll_sb_info *sbi = ll_s2sbi(sb);
1477         struct obd_statfs obd_osfs;
1478         int rc;
1479         ENTRY;
1480
1481         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1482         if (rc) {
1483                 CERROR("obd_statfs fails: rc = %d\n", rc);
1484                 RETURN(rc);
1485         }
1486
1487         osfs->os_type = sb->s_magic;
1488
1489         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1490                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1491
1492         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1493         if (rc) {
1494                 CERROR("obd_statfs fails: rc = %d\n", rc);
1495                 RETURN(rc);
1496         }
1497
1498         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1499                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1500                obd_osfs.os_files);
1501
1502         osfs->os_blocks = obd_osfs.os_blocks;
1503         osfs->os_bfree = obd_osfs.os_bfree;
1504         osfs->os_bavail = obd_osfs.os_bavail;
1505
1506         /* If we don't have as many objects free on the OST as inodes
1507          * on the MDS, we reduce the total number of inodes to
1508          * compensate, so that the "inodes in use" number is correct.
1509          */
1510         if (obd_osfs.os_ffree < osfs->os_ffree) {
1511                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1512                         obd_osfs.os_ffree;
1513                 osfs->os_ffree = obd_osfs.os_ffree;
1514         }
1515
1516         RETURN(rc);
1517 }
1518
1519 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1520 {
1521         struct obd_statfs osfs;
1522         int rc;
1523
1524         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1525         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1526
1527         /* For now we will always get up-to-date statfs values, but in the
1528          * future we may allow some amount of caching on the client (e.g.
1529          * from QOS or lprocfs updates). */
1530         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1531         if (rc)
1532                 return rc;
1533
1534         statfs_unpack(sfs, &osfs);
1535
1536         if (sizeof(sfs->f_blocks) == 4) {
1537                 while (osfs.os_blocks > ~0UL) {
1538                         sfs->f_bsize <<= 1;
1539
1540                         osfs.os_blocks >>= 1;
1541                         osfs.os_bfree >>= 1;
1542                         osfs.os_bavail >>= 1;
1543                 }
1544         }
1545
1546         sfs->f_blocks = osfs.os_blocks;
1547         sfs->f_bfree = osfs.os_bfree;
1548         sfs->f_bavail = osfs.os_bavail;
1549
1550         return 0;
1551 }
1552
1553
1554 /********************************
1555  * remote acl                   *
1556  ********************************/
1557
1558 static struct remote_acl *remote_acl_alloc(void)
1559 {
1560         struct remote_acl *racl;
1561         int i;
1562
1563         OBD_ALLOC(racl, sizeof(*racl));
1564         if (!racl)
1565                 return NULL;
1566
1567         spin_lock_init(&racl->ra_lock);
1568         init_MUTEX(&racl->ra_update_sem);
1569
1570         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1571                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1572
1573         return racl;
1574 }
1575
1576 /*
1577  * caller should guarantee no race here.
1578  */
1579 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1580 {
1581         struct remote_perm_setxid *xperm;
1582
1583         while (!list_empty(&perm->lrp_setxid_perms)) {
1584                 xperm = list_entry(perm->lrp_setxid_perms.next,
1585                                    struct remote_perm_setxid,
1586                                    list);
1587                 list_del(&xperm->list);
1588                 OBD_FREE(xperm, sizeof(*xperm));
1589         }
1590 }
1591
1592 /*
1593  * caller should guarantee no race here.
1594  */
1595 static void remote_acl_flush(struct remote_acl *racl)
1596 {
1597         struct list_head *head;
1598         struct lustre_remote_perm *perm, *tmp;
1599         int i;
1600
1601         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1602                 head = &racl->ra_perm_cache[i];
1603
1604                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1605                         remote_perm_flush_xperms(perm);
1606                         list_del(&perm->lrp_list);
1607                         OBD_FREE(perm, sizeof(*perm));
1608                 }
1609         }
1610 }
1611
1612 static void remote_acl_free(struct remote_acl *racl)
1613 {
1614         if (!racl)
1615                 return;
1616
1617         down(&racl->ra_update_sem);
1618         spin_lock(&racl->ra_lock);
1619         remote_acl_flush(racl);
1620         spin_unlock(&racl->ra_lock);
1621         up(&racl->ra_update_sem);
1622
1623         OBD_FREE(racl, sizeof(*racl));
1624 }
1625
1626 static inline int remote_acl_hashfunc(__u32 id)
1627 {
1628         return (id & (REMOTE_ACL_HASHSIZE - 1));
1629 }
1630
1631 static
1632 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1633 {
1634         struct list_head *head;
1635         struct lustre_remote_perm *lperm;
1636         struct remote_perm_setxid *xperm;
1637         int found = 0, rc = -ENOENT;
1638
1639         LASSERT(racl);
1640         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1641         spin_lock(&racl->ra_lock);
1642
1643         list_for_each_entry(lperm, head, lrp_list) {
1644                 if (lperm->lrp_auth_uid == current->uid) {
1645                         found = 1;
1646                         break;
1647                 }
1648         }
1649
1650         if (!found)
1651                 goto out;
1652
1653         if (lperm->lrp_auth_uid == current->fsuid &&
1654             lperm->lrp_auth_gid == current->fsgid) {
1655                 if (lperm->lrp_valid) {
1656                         *perm = lperm->lrp_perm;
1657                         rc = 0;
1658                 }
1659                 goto out;
1660         } else if ((!lperm->lrp_setuid &&
1661                     lperm->lrp_auth_uid != current->fsuid) ||
1662                    (!lperm->lrp_setgid &&
1663                     lperm->lrp_auth_gid != current->fsgid))  {
1664                 *perm = 0;
1665                 rc = 0;
1666                 goto out;
1667         }
1668
1669         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1670                 if (xperm->uid == current->fsuid &&
1671                     xperm->gid == current->fsgid) {
1672                         *perm = xperm->perm;
1673                         rc = 0;
1674                         goto out;
1675                 }
1676         }
1677
1678 out:
1679         spin_unlock(&racl->ra_lock);
1680         return rc;
1681 }
1682
1683 static
1684 int __remote_acl_update(struct remote_acl *racl,
1685                         struct mds_remote_perm *mperm,
1686                         struct lustre_remote_perm *lperm,
1687                         struct remote_perm_setxid *xperm)
1688 {
1689         struct list_head *head;
1690         struct lustre_remote_perm *lp;
1691         struct remote_perm_setxid *xp;
1692         int found = 0, setuid = 0, setgid = 0;
1693
1694         LASSERT(racl);
1695         LASSERT(mperm);
1696         LASSERT(lperm);
1697         LASSERT(current->uid == mperm->mrp_auth_uid);
1698
1699         if (current->fsuid != mperm->mrp_auth_uid)
1700                 setuid = 1;
1701         if (current->fsgid != mperm->mrp_auth_gid)
1702                 setgid = 1;
1703
1704         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1705         spin_lock(&racl->ra_lock);
1706
1707         list_for_each_entry(lp, head, lrp_list) {
1708                 if (lp->lrp_auth_uid == current->uid) {
1709                         found = 1;
1710                         break;
1711                 }
1712         }
1713
1714         if (found) {
1715                 OBD_FREE(lperm, sizeof(*lperm));
1716
1717                 if (!lp->lrp_valid && !setuid && !setgid) {
1718                         lp->lrp_perm = mperm->mrp_perm;
1719                         lp->lrp_valid = 1;
1720                 }
1721
1722                 /* sanity check for changes of setxid rules */
1723                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1724                         CWARN("setuid changes: %d => %d\n",
1725                               (lp->lrp_setuid != 0),
1726                               (mperm->mrp_allow_setuid != 0));
1727                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1728                 }
1729
1730                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1731                         CWARN("setgid changes: %d => %d\n",
1732                               (lp->lrp_setgid != 0),
1733                               (mperm->mrp_allow_setgid != 0));
1734                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1735                 }
1736
1737                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1738                     !list_empty(&lp->lrp_setxid_perms)) {
1739                         remote_perm_flush_xperms(lp);
1740                 }
1741         } else {
1742                 /* initialize lperm and linked into hashtable
1743                  */
1744                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1745                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1746                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1747                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1748                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1749                 list_add(&lperm->lrp_list, head);
1750
1751                 if (!setuid && !setgid) {
1752                         /* in this case, i'm the authenticated user,
1753                          * and mrp_perm is for me.
1754                          */
1755                         lperm->lrp_perm = mperm->mrp_perm;
1756                         lperm->lrp_valid = 1;
1757                         spin_unlock(&racl->ra_lock);
1758
1759                         if (xperm)
1760                                 OBD_FREE(xperm, sizeof(*xperm));
1761                         return 0;
1762                 }
1763
1764                 lp = lperm;
1765                 /* fall through */
1766         }
1767
1768         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1769                 list_empty(&lp->lrp_setxid_perms));
1770
1771         /* if no xperm supplied, we are all done here */
1772         if (!xperm) {
1773                 spin_unlock(&racl->ra_lock);
1774                 return 0;
1775         }
1776
1777         /* whether we allow setuid/setgid */
1778         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1779                 OBD_FREE(xperm, sizeof(*xperm));
1780                 spin_unlock(&racl->ra_lock);
1781                 return 0;
1782         }
1783
1784         /* traverse xperm list */
1785         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1786                 if (xp->uid == current->fsuid &&
1787                     xp->gid == current->fsgid) {
1788                         if (xp->perm != mperm->mrp_perm) {
1789                                 /* actually this should not happen */
1790                                 CWARN("perm changed: %o => %o\n",
1791                                       xp->perm, mperm->mrp_perm);
1792                                 xp->perm = mperm->mrp_perm;
1793                         }
1794                         OBD_FREE(xperm, sizeof(*xperm));
1795                         spin_unlock(&racl->ra_lock);
1796                         return 0;
1797                 }
1798         }
1799
1800         /* finally insert this xperm */
1801         xperm->uid = current->fsuid;
1802         xperm->gid = current->fsgid;
1803         xperm->perm = mperm->mrp_perm;
1804         list_add(&xperm->list, &lp->lrp_setxid_perms);
1805
1806         spin_unlock(&racl->ra_lock);
1807         return 0;
1808 }
1809
1810 /*
1811  * remote_acl semaphore must be held by caller
1812  */
1813 static
1814 int remote_acl_update_locked(struct remote_acl *racl,
1815                              struct mds_remote_perm *mperm)
1816 {
1817         struct lustre_remote_perm *lperm;
1818         struct remote_perm_setxid *xperm;
1819         int setuid = 0, setgid = 0;
1820
1821         might_sleep();
1822
1823         if (current->uid != mperm->mrp_auth_uid) {
1824                 CERROR("current uid %u while authenticated as %u\n",
1825                        current->uid, mperm->mrp_auth_uid);
1826                 return -EINVAL;
1827         }
1828
1829         if (current->fsuid != mperm->mrp_auth_uid)
1830                 setuid = 1;
1831         if (current->fsgid == mperm->mrp_auth_gid)
1832                 setgid = 1;
1833
1834         OBD_ALLOC(lperm, sizeof(*lperm));
1835         if (!lperm)
1836                 return -ENOMEM;
1837
1838         if ((setuid || setgid) &&
1839             !(setuid && !mperm->mrp_allow_setuid) &&
1840             !(setgid && !mperm->mrp_allow_setgid)) {
1841                 OBD_ALLOC(xperm, sizeof(*xperm));
1842                 if (!xperm) {
1843                         OBD_FREE(lperm, sizeof(*lperm));
1844                         return -ENOMEM;
1845                 }
1846         } else
1847                 xperm = NULL;
1848
1849         return __remote_acl_update(racl, mperm, lperm, xperm);
1850 }
1851
1852 /*
1853  * return -EACCES at any error cases
1854  */
1855 int ll_remote_acl_permission(struct inode *inode, int mode)
1856 {
1857         struct ll_sb_info *sbi = ll_i2sbi(inode);
1858         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1859         struct ptlrpc_request *req = NULL;
1860         struct lustre_id id;
1861         struct mds_remote_perm *mperm;
1862         int rc = -EACCES, perm;
1863
1864         if (!racl)
1865                 return -EACCES;
1866
1867         if (__remote_acl_check(racl, &perm) == 0) {
1868                 return ((perm & mode) == mode ? 0 : -EACCES);
1869         }
1870
1871         might_sleep();
1872
1873         /* doing update
1874          */
1875         down(&racl->ra_update_sem);
1876
1877         /* we might lose the race when obtain semaphore,
1878          * so check again.
1879          */
1880         if (__remote_acl_check(racl, &perm) == 0) {
1881                 if ((perm & mode) == mode)
1882                         rc = 0;
1883                 goto out;
1884         }
1885
1886         /* really fetch from mds
1887          */
1888         ll_inode2id(&id, inode);
1889         if (md_access_check(sbi->ll_md_exp, &id, &req))
1890                 goto out;
1891
1892         /* status non-zero indicate there's more apparent error
1893          * detected by mds, e.g. didn't allow this user at all.
1894          * we simply ignore and didn't cache it.
1895          */
1896         if (req->rq_repmsg->status)
1897                 goto out;
1898
1899         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1900                                    lustre_swab_remote_perm);
1901         LASSERT(mperm);
1902         LASSERT_REPSWABBED(req, 1);
1903
1904         if ((mperm->mrp_perm & mode) == mode)
1905                 rc = 0;
1906
1907         remote_acl_update_locked(racl, mperm);
1908 out:
1909         if (req)
1910                 ptlrpc_req_finished(req);
1911
1912         up(&racl->ra_update_sem);
1913         return rc;
1914 }
1915
1916 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1917 {
1918         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1919         int rc;
1920
1921         LASSERT(perm);
1922
1923         if (!racl)
1924                 return -EACCES;
1925
1926         down(&racl->ra_update_sem);
1927         rc = remote_acl_update_locked(racl, perm);
1928         up(&racl->ra_update_sem);
1929
1930         return rc;
1931 }
1932
1933 void ll_inode_invalidate_acl(struct inode *inode)
1934 {
1935         struct ll_sb_info *sbi = ll_i2sbi(inode);
1936         struct ll_inode_info *lli = ll_i2info(inode);
1937
1938         if (sbi->ll_remote) {
1939                 struct remote_acl *racl = lli->lli_remote_acl;
1940
1941                 LASSERT(!lli->lli_posix_acl);
1942                 if (racl) {
1943                         down(&racl->ra_update_sem);
1944                         spin_lock(&racl->ra_lock);
1945                         remote_acl_flush(lli->lli_remote_acl);
1946                         spin_unlock(&racl->ra_lock);
1947                         up(&racl->ra_update_sem);
1948                 }
1949         } else {
1950                 /* we can't invalide acl here: suppose we touch a new file
1951                  * under a dir, blocking ast on dir will lead to open failure
1952                  * on client, although succeed on mds. it's kind of weird,
1953                  * the real fix i think is improve client-vfs interaction.
1954                  *
1955                  * currently we just do nothing here.
1956                  */
1957                 return;
1958
1959                 LASSERT(!lli->lli_remote_acl);
1960                 spin_lock(&lli->lli_lock);
1961                 posix_acl_release(lli->lli_posix_acl);
1962                 lli->lli_posix_acl = NULL;
1963                 spin_unlock(&lli->lli_lock);
1964         }
1965 }
1966
1967 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1968 {
1969         struct ll_inode_info *lli = ll_i2info(inode);
1970         struct lov_stripe_md *lsm = md->lsm;
1971         struct mds_body *body = md->body;
1972         struct mea *mea = md->mea;
1973         struct posix_acl *posix_acl = md->posix_acl;
1974         struct ll_sb_info *sbi = ll_i2sbi(inode);
1975         struct lustre_key *mkey = md->key;
1976         ENTRY;
1977
1978         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1979
1980         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1981                 /* check for default striping info for dir. */
1982                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1983         }
1984
1985         if (lsm != NULL) {
1986                 LASSERT(lsm->lsm_object_gr > 0);
1987                 if (lli->lli_smd == NULL) {
1988                         lli->lli_smd = lsm;
1989                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1990                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1991                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1992                 } else {
1993                         int i;
1994                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1995                                 CERROR("lsm mismatch for inode %ld\n",
1996                                        inode->i_ino);
1997                                 CERROR("lli_smd:\n");
1998                                 dump_lsm(D_ERROR, lli->lli_smd);
1999                                 CERROR("lsm:\n");
2000                                 dump_lsm(D_ERROR, lsm);
2001                                 LBUG();
2002                         }
2003                         /* XXX FIXME -- We should decide on a safer (atomic) and
2004                          * more elegant way to update the lsm */
2005                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2006                                 lli->lli_smd->lsm_oinfo[i].loi_id =
2007                                         lsm->lsm_oinfo[i].loi_id;
2008                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
2009                                         lsm->lsm_oinfo[i].loi_gr;
2010                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
2011                                         lsm->lsm_oinfo[i].loi_ost_idx;
2012                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
2013                                         lsm->lsm_oinfo[i].loi_ost_gen;
2014                         }
2015                 }
2016                 /* bug 2844 - limit i_blksize for broken user-space apps */
2017                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
2018                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
2019                 if (lli->lli_smd != lsm)
2020                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
2021         }
2022
2023         if (mea != NULL) {
2024                 if (lli->lli_mea == NULL) {
2025                         lli->lli_mea = mea;
2026                 } else {
2027                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
2028                                 CERROR("mea mismatch for inode %lu\n",
2029                                         inode->i_ino);
2030                                 LBUG();
2031                         }
2032                 }
2033                 if (lli->lli_mea != mea)
2034                         obd_free_memmd(ll_i2mdexp(inode),
2035                                        (struct lov_stripe_md **) &mea);
2036         }
2037
2038         if (body->valid & OBD_MD_FID)
2039                 id_assign_fid(&lli->lli_id, &body->id1);
2040
2041         if (body->valid & OBD_MD_FLID)
2042                 id_ino(&lli->lli_id) = id_ino(&body->id1);
2043
2044         if (body->valid & OBD_MD_FLGENER)
2045                 id_gen(&lli->lli_id) = id_gen(&body->id1);
2046
2047         /* local/remote ACL */
2048         if (sbi->ll_remote) {
2049                 LASSERT(md->posix_acl == NULL);
2050                 if (md->remote_perm) {
2051                         ll_remote_acl_update(inode, md->remote_perm);
2052                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2053                         md->remote_perm = NULL;
2054                 }
2055         } else {
2056                 LASSERT(md->remote_perm == NULL);
2057                 spin_lock(&lli->lli_lock);
2058                 if (posix_acl != NULL) {
2059                         if (lli->lli_posix_acl != NULL)
2060                                 posix_acl_release(lli->lli_posix_acl);
2061                         lli->lli_posix_acl = posix_acl;
2062                 }
2063                 spin_unlock(&lli->lli_lock);
2064         }
2065
2066         if (body->valid & OBD_MD_FLID)
2067                 inode->i_ino = id_ino(&body->id1);
2068         if (body->valid & OBD_MD_FLGENER)
2069                 inode->i_generation = id_gen(&body->id1);
2070         if (body->valid & OBD_MD_FLATIME)
2071                 LTIME_S(inode->i_atime) = body->atime;
2072         if (body->valid & OBD_MD_FLMTIME &&
2073             body->mtime > LTIME_S(inode->i_mtime)) {
2074                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2075                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2076                 LTIME_S(inode->i_mtime) = body->mtime;
2077         }
2078         if (body->valid & OBD_MD_FLCTIME &&
2079             body->ctime > LTIME_S(inode->i_ctime))
2080                 LTIME_S(inode->i_ctime) = body->ctime;
2081         if (body->valid & OBD_MD_FLMODE) {
2082                 inode->i_mode = (inode->i_mode & S_IFMT) |
2083                         (body->mode & ~S_IFMT);
2084         }
2085         if (body->valid & OBD_MD_FLTYPE) {
2086                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2087                         (body->mode & S_IFMT);
2088         }
2089         if (body->valid & OBD_MD_FLUID)
2090                 inode->i_uid = body->uid;
2091         if (body->valid & OBD_MD_FLGID)
2092                 inode->i_gid = body->gid;
2093         if (body->valid & OBD_MD_FLFLAGS)
2094                 inode->i_flags = body->flags;
2095         if (body->valid & OBD_MD_FLNLINK)
2096                 inode->i_nlink = body->nlink;
2097         if (body->valid & OBD_MD_FLRDEV)
2098 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2099                 inode->i_rdev = body->rdev;
2100 #else
2101                 inode->i_rdev = old_decode_dev(body->rdev);
2102 #endif
2103         if (body->valid & OBD_MD_FLSIZE)
2104                 inode->i_size = body->size;
2105         if (body->valid & OBD_MD_FLBLOCKS)
2106                 inode->i_blocks = body->blocks;
2107
2108         if (body->valid & OBD_MD_FLSIZE)
2109                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2110
2111         if (body->valid & OBD_MD_FLAUDIT) {
2112                 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2113                 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2114                         sbi->ll_audit_mask = body->audit;
2115                 else
2116                         lli->lli_audit_mask = body->audit;
2117         }
2118
2119         if (mkey != NULL) {
2120                 LASSERT(body->valid & OBD_MD_FLKEY);
2121                 ll_crypto_init_inode_key(inode, mkey);
2122         }
2123
2124 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2125         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2126 #endif
2127         LASSERT(id_fid(&lli->lli_id) != 0);
2128 }
2129
2130 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2131 static struct backing_dev_info ll_backing_dev_info = {
2132         .ra_pages       = 0,    /* No readahead */
2133         .memory_backed  = 0,    /* Does contribute to dirty memory */
2134 };
2135 #endif
2136
2137 void ll_read_inode2(struct inode *inode, void *opaque)
2138 {
2139         struct lustre_md *md = opaque;
2140         struct ll_inode_info *lli = ll_i2info(inode);
2141         ENTRY;
2142
2143         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2144                inode->i_generation, inode);
2145
2146         ll_lli_init(lli);
2147
2148         LASSERT(!lli->lli_smd);
2149
2150         if (ll_i2sbi(inode)->ll_remote) {
2151                 lli->lli_remote_acl = remote_acl_alloc();
2152                 /* if failed alloc, nobody will be able to access this inode */
2153         }
2154
2155         /* Core attributes from the MDS first.  This is a new inode, and
2156          * the VFS doesn't zero times in the core inode so we have to do
2157          * it ourselves.  They will be overwritten by either MDS or OST
2158          * attributes - we just need to make sure they aren't newer. */
2159         LTIME_S(inode->i_mtime) = 0;
2160         LTIME_S(inode->i_atime) = 0;
2161         LTIME_S(inode->i_ctime) = 0;
2162
2163         inode->i_rdev = 0;
2164         ll_update_inode(inode, md);
2165
2166         /* OIDEBUG(inode); */
2167
2168         if (S_ISREG(inode->i_mode)) {
2169                 inode->i_op = &ll_file_inode_operations;
2170                 inode->i_fop = &ll_file_operations;
2171                 inode->i_mapping->a_ops = &ll_aops;
2172                 EXIT;
2173         } else if (S_ISDIR(inode->i_mode)) {
2174                 inode->i_op = &ll_dir_inode_operations;
2175                 inode->i_fop = &ll_dir_operations;
2176                 inode->i_mapping->a_ops = &ll_dir_aops;
2177                 EXIT;
2178         } else if (S_ISLNK(inode->i_mode)) {
2179                 inode->i_op = &ll_fast_symlink_inode_operations;
2180                 EXIT;
2181         } else {
2182                 inode->i_op = &ll_special_inode_operations;
2183
2184 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2185                 init_special_inode(inode, inode->i_mode,
2186                                    kdev_t_to_nr(inode->i_rdev));
2187
2188                 /* initializing backing dev info. */
2189                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2190 #else
2191                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2192 #endif
2193                 lli->ll_save_ifop = inode->i_fop;
2194
2195                 if (S_ISCHR(inode->i_mode))
2196                         inode->i_fop = &ll_special_chr_inode_fops;
2197                 else if (S_ISBLK(inode->i_mode))
2198                         inode->i_fop = &ll_special_blk_inode_fops;
2199                 else if (S_ISFIFO(inode->i_mode))
2200                         inode->i_fop = &ll_special_fifo_inode_fops;
2201                 else if (S_ISSOCK(inode->i_mode))
2202                         inode->i_fop = &ll_special_sock_inode_fops;
2203
2204                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2205                       inode->i_fop);
2206
2207                 if (lli->ll_save_ifop->owner) {
2208                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2209                               lli->ll_save_ifop->owner);
2210                 }
2211                 EXIT;
2212         }
2213 }
2214
2215 void ll_delete_inode(struct inode *inode)
2216 {
2217         struct ll_sb_info *sbi = ll_i2sbi(inode);
2218         struct lustre_id id;
2219         int rc;
2220         ENTRY;
2221
2222         ll_inode2id(&id, inode);
2223
2224         rc = md_delete_inode(sbi->ll_md_exp, &id);
2225         if (rc) {
2226                 CERROR("md_delete_inode() failed, error %d\n",
2227                        rc);
2228         }
2229
2230         clear_inode(inode);
2231         EXIT;
2232 }
2233
2234 int ll_iocontrol(struct inode *inode, struct file *file,
2235                  unsigned int cmd, unsigned long arg)
2236 {
2237         struct ll_sb_info *sbi = ll_i2sbi(inode);
2238         struct ptlrpc_request *req = NULL;
2239         int rc, flags = 0;
2240         ENTRY;
2241
2242         switch(cmd) {
2243         case EXT3_IOC_GETFLAGS: {
2244                 struct lustre_id id;
2245                 __u64 valid = OBD_MD_FLFLAGS;
2246                 struct mds_body *body;
2247
2248                 ll_inode2id(&id, inode);
2249                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2250                                 0, 0, NULL, &req);
2251                 if (rc) {
2252                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2253                         RETURN(-abs(rc));
2254                 }
2255
2256                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2257
2258                 if (body->flags & S_APPEND)
2259                         flags |= EXT3_APPEND_FL;
2260                 if (body->flags & S_IMMUTABLE)
2261                         flags |= EXT3_IMMUTABLE_FL;
2262                 if (body->flags & S_NOATIME)
2263                         flags |= EXT3_NOATIME_FL;
2264
2265                 ptlrpc_req_finished (req);
2266
2267                 RETURN(put_user(flags, (int *)arg));
2268         }
2269         case EXT3_IOC_SETFLAGS: {
2270                 struct mdc_op_data *op_data;
2271                 struct iattr attr;
2272                 struct obdo *oa;
2273                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2274
2275                 if (get_user(flags, (int *)arg))
2276                         RETURN(-EFAULT);
2277
2278                 oa = obdo_alloc();
2279                 if (!oa)
2280                         RETURN(-ENOMEM);
2281
2282                 OBD_ALLOC(op_data, sizeof(*op_data));
2283                 if (op_data == NULL) {
2284                         obdo_free(oa);
2285                         RETURN(-ENOMEM);
2286                 }
2287                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
2288
2289                 memset(&attr, 0x0, sizeof(attr));
2290                 attr.ia_attr_flags = flags;
2291                 attr.ia_valid |= ATTR_ATTR_FLAG;
2292
2293                 rc = md_setattr(sbi->ll_md_exp, op_data,
2294                                 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2295                 OBD_FREE(op_data, sizeof(*op_data));
2296                 if (rc) {
2297                         ptlrpc_req_finished(req);
2298                         if (rc != -EPERM && rc != -EACCES)
2299                                 CERROR("md_setattr fails: rc = %d\n", rc);
2300                         obdo_free(oa);
2301                         RETURN(rc);
2302                 }
2303                 ptlrpc_req_finished(req);
2304
2305                 oa->o_id = lsm->lsm_object_id;
2306                 oa->o_gr = lsm->lsm_object_gr;
2307                 oa->o_flags = flags;
2308                 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2309                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2310                               | OBD_MD_FLIFID;
2311
2312                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL, NULL);
2313                 obdo_free(oa);
2314                 if (rc) {
2315                         if (rc != -EPERM && rc != -EACCES)
2316                                 CERROR("md_setattr fails: rc = %d\n", rc);
2317                         RETURN(rc);
2318                 }
2319
2320                 if (flags & EXT3_APPEND_FL)
2321                         inode->i_flags |= S_APPEND;
2322                 else
2323                         inode->i_flags &= ~S_APPEND;
2324                 if (flags & EXT3_IMMUTABLE_FL)
2325                         inode->i_flags |= S_IMMUTABLE;
2326                 else
2327                         inode->i_flags &= ~S_IMMUTABLE;
2328                 if (flags & EXT3_NOATIME_FL)
2329                         inode->i_flags |= S_NOATIME;
2330                 else
2331                         inode->i_flags &= ~S_NOATIME;
2332
2333                 RETURN(0);
2334         }
2335         default:
2336                 RETURN(-ENOSYS);
2337         }
2338
2339         RETURN(0);
2340 }
2341
2342 /* this is only called in the case of forced umount. */
2343 void ll_umount_begin(struct super_block *sb)
2344 {
2345         struct ll_sb_info *sbi = ll_s2sbi(sb);
2346         struct obd_ioctl_data ioc_data = { 0 };
2347         struct obd_device *obd;
2348         ENTRY;
2349
2350         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2351                sb->s_count, atomic_read(&sb->s_active));
2352
2353         obd = class_exp2obd(sbi->ll_md_exp);
2354         if (obd == NULL) {
2355                 CERROR("Invalid MDC connection handle "LPX64"\n",
2356                        sbi->ll_md_exp->exp_handle.h_cookie);
2357                 EXIT;
2358                 return;
2359         }
2360         obd->obd_no_recov = 1;
2361         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2362                       sizeof(ioc_data), &ioc_data, NULL);
2363
2364         obd = class_exp2obd(sbi->ll_dt_exp);
2365         if (obd == NULL) {
2366                 CERROR("Invalid LOV connection handle "LPX64"\n",
2367                        sbi->ll_dt_exp->exp_handle.h_cookie);
2368                 EXIT;
2369                 return;
2370         }
2371
2372         obd->obd_no_recov = 1;
2373         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2374                       sizeof(ioc_data), &ioc_data, NULL);
2375
2376         /*
2377          * really, we'd like to wait until there are no requests outstanding,
2378          * and then continue.  For now, we just invalidate the requests,
2379          * schedule, and hope.
2380          */
2381         schedule();
2382
2383         EXIT;
2384 }
2385
2386 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2387                   struct inode **inode, struct ptlrpc_request *req,
2388                   int offset, struct super_block *sb)
2389 {
2390         struct lustre_md md;
2391         int rc = 0;
2392
2393         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2394         if (rc)
2395                 RETURN(rc);
2396
2397         if (*inode) {
2398                 ll_update_inode(*inode, &md);
2399         } else {
2400                 LASSERT(sb);
2401                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2402                 if (*inode == NULL || is_bad_inode(*inode)) {
2403                         /* free the lsm if we allocated one above */
2404                         if (md.lsm != NULL)
2405                                 obd_free_memmd(dt_exp, &md.lsm);
2406                         if (md.mea != NULL)
2407                                 obd_free_memmd(md_exp,
2408                                                (struct lov_stripe_md**)&md.mea);
2409                         rc = -ENOMEM;
2410                         CERROR("new_inode -fatal: rc %d\n", rc);
2411                 }
2412         }
2413
2414         RETURN(rc);
2415 }
2416
2417 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2418 {
2419         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2420         struct lustre_mount_data *lmd = sbi->ll_lmd;
2421
2422         if (lmd) {
2423                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2424                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2425         }
2426         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2427         if (sbi->ll_remote && lmd)
2428                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2429
2430         if (lmd && lmd->lmd_pag)
2431                 seq_printf(m, ",pag");
2432
2433         return 0;
2434 }
2435
2436 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2437                char *filename, struct lustre_id *ret)
2438 {
2439         struct ptlrpc_request *request = NULL;
2440         struct mds_body *body;
2441         int rc;
2442
2443         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2444                              OBD_MD_FID, 0, &request);
2445         if (rc < 0) {
2446                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2447                        filename, rc);
2448                 return rc;
2449         }
2450
2451         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2452         LASSERT(body != NULL);
2453         LASSERT_REPSWABBED(request, 0);
2454
2455         *ret = body->id1;
2456         ptlrpc_req_finished(request);
2457
2458         return rc;
2459 }
2460 int ll_flush_cred(struct inode *inode)
2461 {
2462         struct ll_sb_info *sbi = ll_i2sbi(inode);
2463         int rc = 0;
2464
2465         /* XXX to avoid adding api, we simply use set_info() interface
2466          * to notify underlying obds. set_info() is more like a ioctl() now...
2467          */
2468         if (sbi->ll_md_exp) {
2469                 rc = obd_set_info(sbi->ll_md_exp,
2470                                   strlen("flush_cred"), "flush_cred",
2471                                   0, NULL);
2472                 if (rc)
2473                         return rc;
2474         }
2475
2476         if (sbi->ll_dt_exp) {
2477                 rc = obd_set_info(sbi->ll_dt_exp,
2478                                   strlen("flush_cred"), "flush_cred",
2479                                   0, NULL);
2480                 if (rc)
2481                         return rc;
2482         }
2483
2484         return rc;
2485 }