lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/random.h>
  29 #include <linux/version.h>
  30 #include <linux/seq_file.h>
  31
  32 #include <linux/lustre_lite.h>
  33 #include <linux/lustre_ha.h>
  34 #include <linux/lustre_dlm.h>
  35 #include <linux/lprocfs_status.h>
  36 #include <linux/lustre_acl.h>
  37 #include <linux/lustre_audit.h>
  38 #include <linux/lustre_gs.h>
  39 #include <linux/lustre_sec.h>
  40 #include "llite_internal.h"
  41
  42 kmem_cache_t *ll_file_data_slab;
  43 kmem_cache_t *ll_intent_slab;
  44
  45 extern struct address_space_operations ll_aops;
  46 extern struct address_space_operations ll_dir_aops;
  47
  48 #ifndef log2
  49 #define log2(n) ffz(~(n))
  50 #endif
  51
  52 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
  53 {
  54         struct ll_sb_info *sbi = NULL;
  55         class_uuid_t uuid;
  56         ENTRY;
  57
  58         OBD_ALLOC(sbi, sizeof(*sbi));
  59         if (!sbi)
  60                 RETURN(NULL);
  61
  62         spin_lock_init(&sbi->ll_lock);
  63         INIT_LIST_HEAD(&sbi->ll_pglist);
  64         sbi->ll_pglist_gen = 0;
  65         if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
  66                 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
  67         else
  68                 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
  69         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  70         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  71         INIT_LIST_HEAD(&sbi->ll_mnt_list);
  72
  73         sema_init(&sbi->ll_gns_sem, 1);
  74         spin_lock_init(&sbi->ll_gns_lock);
  75         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
  76         init_waitqueue_head(&sbi->ll_gns_waitq);
  77         init_completion(&sbi->ll_gns_mount_finished);
  78
  79         /* this later may be reset via /proc/fs/... */
  80         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
  81         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
  82
  83         /* this later may be reset via /proc/fs/... */
  84         memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
  85                strlen("/usr/sbin/gns_upcall"));
  86         sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
  87
  88         /* default values, may be changed via /proc/fs/... */
  89         sbi->ll_gns_state = LL_GNS_IDLE;
  90         sbi->ll_gns_pending_dentry = NULL;
  91         atomic_set(&sbi->ll_gns_enabled, 1);
  92         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  93         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  94
  95         sbi->ll_gns_timer.data = (unsigned long)sbi;
  96         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  97         init_timer(&sbi->ll_gns_timer);
  98         //audit mask
  99         sbi->ll_audit_mask = AUDIT_OFF;
 100         ll_set_sbi(sb, sbi);
 101
 102         generate_random_uuid(uuid);
 103         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 104         RETURN(sbi);
 105 }
 106
 107 void lustre_free_sbi(struct super_block *sb)
 108 {
 109         struct ll_sb_info *sbi = ll_s2sbi(sb);
 110         ENTRY;
 111
 112         if (sbi != NULL) {
 113                 list_del(&sbi->ll_gns_sbi_head);
 114                 del_timer(&sbi->ll_gns_timer);
 115                 OBD_FREE(sbi, sizeof(*sbi));
 116         }
 117         ll_set_sbi(sb, NULL);
 118         EXIT;
 119 }
 120
 121 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 122 {
 123         __u32 valsize;
 124         int rc = 0;
 125         ENTRY;
 126
 127         valsize = sizeof(sbi->ll_dt_desc);
 128         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 129         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 130                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 131         RETURN(rc);
 132 }
 133
 134 static int lustre_connect_mds(struct super_block *sb, char *lmv,
 135                               struct obd_connect_data *data,
 136                               char *mds_security, int async, int pag)
 137 {
 138         struct ll_sb_info *sbi = ll_s2sbi(sb);
 139         struct lustre_handle md_conn = {0, };
 140         struct obd_device *md_obd;
 141         struct obd_statfs osfs;
 142         unsigned long sec_flags;
 143         __u32 valsize;
 144         int    err = 0;
 145         ENTRY;
 146
 147         md_obd = class_name2obd(lmv);
 148         if (!md_obd) {
 149                 CERROR("MDC %s: not setup or attached\n", lmv);
 150                 RETURN(-EINVAL);
 151         }
 152
 153         obd_set_info(md_obd->obd_self_export, strlen("async"), "async",
 154                      sizeof(async), &async);
 155
 156         if (mds_security == NULL)
 157                 mds_security = "null";
 158
 159         err = obd_set_info(md_obd->obd_self_export, strlen("sec"), "sec",
 160                            strlen(mds_security), mds_security);
 161
 162         if (err) {
 163                 CERROR("LMV %s: failed to set security %s, err %d\n",
 164                         lmv, mds_security, err);
 165                 RETURN(err);
 166         }
 167
 168         if (pag) {
 169                 sec_flags = PTLRPC_SEC_FL_PAG;
 170                 err = obd_set_info(md_obd->obd_self_export,
 171                                    strlen("sec_flags"), "sec_flags",
 172                                    sizeof(sec_flags), &sec_flags);
 173                 if (err) {
 174                         OBD_FREE(data, sizeof(*data));
 175                         RETURN(err);
 176                 }
 177         }
 178
 179         err = obd_connect(&md_conn, md_obd, &sbi->ll_sb_uuid, data,
 180                           OBD_OPT_REAL_CLIENT);
 181         if (err == -EBUSY) {
 182                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 183                        " client is not a part.  Please wait for recovery to "
 184                        "complete, abort, or time out.\n", lmv);
 185                 GOTO(out, err);
 186         } else if (err) {
 187                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 188                 GOTO(out, err);
 189         }
 190
 191         sbi->ll_md_exp = class_conn2export(&md_conn);
 192
 193         err = obd_statfs(md_obd, &osfs, jiffies - HZ);
 194         if (err)
 195                 GOTO(out_disconnect, err);
 196
 197         if (!osfs.os_bsize) {
 198                 CERROR("Invalid block size is detected.");
 199                 GOTO(out_disconnect, err);
 200         }
 201
 202         sb->s_magic = LL_SUPER_MAGIC;
 203         sb->s_blocksize = osfs.os_bsize;
 204         sb->s_blocksize_bits = log2(osfs.os_bsize);
 205         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 206
 207         /* in 2.6.x FS is not allowed to form s_dev */
 208 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 209         {
 210                 kdev_t devno;
 211
 212                 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 213                                      strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 214
 215                 sb->s_dev = devno;
 216         }
 217 #endif
 218
 219         /* after statfs, we are supposed to have connected to MDSs,
 220          * so it's ok to check remote flag returned.
 221          */
 222         valsize = sizeof(&sbi->ll_remote);
 223         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 224                            &valsize, &sbi->ll_remote);
 225         if (err) {
 226                 CERROR("fail to obtain remote flag\n");
 227                 GOTO(out_disconnect, err);
 228         }
 229
 230 out_disconnect:
 231         if (err)
 232                 obd_disconnect(sbi->ll_md_exp, 0);
 233 out:
 234         RETURN(err);
 235 }
 236
 237 static int lustre_connect_ost(struct super_block *sb, char *lov,
 238                               struct obd_connect_data *data,
 239                               char *oss_security, int async, int pag)
 240 {
 241         struct ll_sb_info *sbi = ll_s2sbi(sb);
 242         struct lustre_handle dt_conn = {0, };
 243         struct obd_device *obd = NULL;
 244         unsigned long sec_flags;
 245         int err, mdsize;
 246
 247         obd = class_name2obd(lov);
 248         if (!obd) {
 249                 CERROR("OSC %s: not setup or attached\n", lov);
 250                 GOTO(out, err = -EINVAL);
 251         }
 252         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 253                      sizeof(async), &async);
 254
 255        if (oss_security == NULL)
 256                 oss_security = "null";
 257
 258         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 259                            strlen(oss_security), oss_security);
 260         if (err) {
 261                 CERROR("LOV %s: failed to set security %s, err %d\n",
 262                         lov, oss_security, err);
 263                 RETURN(err);
 264         }
 265
 266         if (pag) {
 267                 sec_flags = PTLRPC_SEC_FL_PAG;
 268                 err = obd_set_info(obd->obd_self_export,
 269                                    strlen("sec_flags"), "sec_flags",
 270                                    sizeof(sec_flags), &sec_flags);
 271                 if (err) {
 272                         OBD_FREE(data, sizeof(*data));
 273                         RETURN(err);
 274                 }
 275         }
 276
 277         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
 278         if (err == -EBUSY) {
 279                 CERROR("An OST (lov %s) is performing recovery, of which this"
 280                        " client is not a part.  Please wait for recovery to "
 281                        "complete, abort, or time out.\n", lov);
 282                 GOTO(out, err);
 283         } else if (err) {
 284                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 285                 GOTO(out, err);
 286         }
 287         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 288
 289         err = lustre_init_dt_desc(sbi);
 290
 291         if (err) {
 292                 CWARN("init dt_desc error %d \n", err);
 293                 GOTO(out, err = 0);
 294         }
 295         mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 296         obd_init_ea_size(sbi->ll_md_exp, mdsize, sbi->ll_dt_desc.ld_tgt_count *
 297                          sizeof(struct llog_cookie));
 298 out:
 299         RETURN(err);
 300 }
 301
 302 extern struct dentry_operations ll_d_ops;
 303
 304 static int lustre_init_root_inode(struct super_block *sb)
 305 {
 306         struct ll_sb_info *sbi = ll_s2sbi(sb);
 307         struct ptlrpc_request *request = NULL;
 308         struct inode *root = NULL;
 309         struct lustre_md md;
 310         int err = 0;
 311         ENTRY;
 312
 313         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 314         if (err) {
 315                 CERROR("cannot mds_connect: rc = %d\n", err);
 316                 GOTO(out, err);
 317         }
 318         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 319
 320         sb->s_op = &lustre_super_operations;
 321
 322         /* make root inode */
 323         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 324                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
 325                          NULL, NULL, 0, 0, NULL, &request);
 326         if (err) {
 327                 CERROR("md_getattr failed for root: rc = %d\n", err);
 328                 GOTO(out, err);
 329         }
 330
 331         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 332                                 sbi->ll_dt_exp, &md);
 333         if (err) {
 334                 CERROR("failed to understand root inode md: rc = %d\n", err);
 335                 ptlrpc_req_finished(request);
 336                 GOTO(out, err);
 337         }
 338
 339         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 340         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 341
 342         ptlrpc_req_finished(request);
 343
 344         if (root == NULL || is_bad_inode(root)) {
 345                 if (md.lsm != NULL)
 346                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 347                 if (md.mea != NULL)
 348                     obd_free_memmd(sbi->ll_md_exp,
 349                                    (struct lov_stripe_md**)&md.mea);
 350                 CERROR("lustre_lite: bad iget4 for root\n");
 351                 GOTO(out_root, err = -EBADF);
 352         }
 353         sb->s_root = d_alloc_root(root);
 354         sb->s_root->d_op = &ll_d_ops;
 355 out_root:
 356         if (err)
 357                 iput(root);
 358 out:
 359         RETURN(err);
 360 }
 361
 362 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 363                              char *gkc, int async,  char *mds_security,
 364                              char *oss_security, __u32 *nllu, int pag,
 365                              __u64 *remote)
 366 {
 367         struct ll_sb_info *sbi = ll_s2sbi(sb);
 368         struct obd_connect_data *data;
 369         int err;
 370         ENTRY;
 371
 372         /*process the connect flags*/
 373         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 374                        (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 375                 CERROR("wrong remote flag "LPX64"\n", *remote);
 376                 RETURN(-EINVAL);
 377         }
 378
 379         OBD_ALLOC(data, sizeof(*data));
 380         if (!data)
 381                 RETURN(-ENOMEM);
 382
 383         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 384                                               OBD_CONNECT_REMOTE);
 385         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 386
 387         if (proc_lustre_fs_root) {
 388                 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
 389                                                   sb, lov, lmv);
 390                 if (err < 0)
 391                         CERROR("could not register mount in /proc/lustre");
 392         }
 393
 394         /*connect mds */
 395         err = lustre_connect_mds(sb, lmv, data, mds_security, async, pag);
 396         if (err)
 397                 GOTO(out, err);
 398
 399         /*connect OST*/
 400         err = lustre_connect_ost(sb, lov, data, oss_security, async, pag);
 401         if (err)
 402                 GOTO(out_lmv, err);
 403
 404         err = lustre_init_crypto(sb, gkc, data, async);
 405         if (err) {
 406                 CERROR("Could not connect to GSS err %d\n", err);
 407                 err = 0;
 408         }
 409         /*connect GSS*/
 410         err = lustre_init_root_inode(sb);
 411         if (err)
 412                 GOTO(out_gks, err);
 413
 414         err = ll_close_thread_start(&sbi->ll_lcq);
 415         if (err) {
 416                 CERROR("cannot start close thread: rc %d\n", err);
 417                 GOTO(out_root, err);
 418         }
 419
 420         ll_gns_add_timer(sbi);
 421
 422         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 423            backing dev info assigned to inode mapping is used for
 424            determining maximal readahead. */
 425 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 426     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 427         /* bug 2805 - set VM readahead to zero */
 428         vm_max_readahead = vm_min_readahead = 0;
 429 #endif
 430         sb->s_flags |= MS_POSIXACL;
 431 #ifdef S_PDIROPS
 432         CWARN("Enabling PDIROPS\n");
 433         sb->s_flags |= S_PDIROPS;
 434 #endif
 435         if (data != NULL)
 436                 OBD_FREE(data, sizeof(*data));
 437         RETURN(err);
 438 out_root:
 439         if (sb->s_root)
 440                 dput(sb->s_root);
 441 out_gks:
 442         lustre_destroy_crypto(sb);
 443 out_lmv:
 444         obd_disconnect(sbi->ll_md_exp, 0);
 445 out:
 446         if (data != NULL)
 447                 OBD_FREE(data, sizeof(*data));
 448         lprocfs_unregister_mountpoint(sbi);
 449         RETURN(err);
 450 }
 451
 452 void lustre_common_put_super(struct super_block *sb)
 453 {
 454         struct ll_sb_info *sbi = ll_s2sbi(sb);
 455         struct hlist_node *tmp, *next;
 456         ENTRY;
 457
 458         ll_gns_del_timer(sbi);
 459         ll_close_thread_stop(sbi->ll_lcq);
 460
 461         lustre_destroy_crypto(sb);
 462
 463         list_del(&sbi->ll_conn_chain);
 464         obd_disconnect(sbi->ll_dt_exp, 0);
 465
 466         lprocfs_unregister_mountpoint(sbi);
 467         if (sbi->ll_proc_root) {
 468                 lprocfs_remove(sbi->ll_proc_root);
 469                 sbi->ll_proc_root = NULL;
 470         }
 471
 472         obd_disconnect(sbi->ll_md_exp, 0);
 473
 474         // We do this to get rid of orphaned dentries. That is not really trw.
 475         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 476                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 477                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 478                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 479                 shrink_dcache_parent(dentry);
 480         }
 481         EXIT;
 482 }
 483
 484 char *ll_read_opt(const char *opt, char *data)
 485 {
 486         char *value;
 487         char *retval;
 488         ENTRY;
 489
 490         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 491         if (strncmp(opt, data, strlen(opt)))
 492                 RETURN(NULL);
 493         if ((value = strchr(data, '=')) == NULL)
 494                 RETURN(NULL);
 495
 496         value++;
 497         OBD_ALLOC(retval, strlen(value) + 1);
 498         if (!retval) {
 499                 CERROR("out of memory!\n");
 500                 RETURN(NULL);
 501         }
 502
 503         memcpy(retval, value, strlen(value)+1);
 504         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 505         RETURN(retval);
 506 }
 507
 508 int ll_set_opt(const char *opt, char *data, int fl)
 509 {
 510         ENTRY;
 511
 512         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 513         if (strncmp(opt, data, strlen(opt)))
 514                 RETURN(0);
 515         else
 516                 RETURN(fl);
 517 }
 518
 519 void ll_options(char *options, char **lov, char **lmv, char **gkc,
 520                 char **mds_sec, char **oss_sec, int *async, int *flags)
 521 {
 522         char *this_char;
 523 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 524         char *opt_ptr = options;
 525 #endif
 526         ENTRY;
 527
 528         if (!options) {
 529                 EXIT;
 530                 return;
 531         }
 532
 533         *async = 0;
 534 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 535         for (this_char = strtok (options, ",");
 536              this_char != NULL;
 537              this_char = strtok (NULL, ",")) {
 538 #else
 539         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 540 #endif
 541                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 542                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 543                         continue;
 544                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 545                         continue;
 546                 if (!*gkc && (*gkc = ll_read_opt("gkc", this_char)))
 547                         continue;
 548                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 549                         *async = 1;
 550                         continue;
 551                 }
 552                 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
 553                         continue;
 554                 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
 555                         continue;
 556                 if (!(*flags & LL_SBI_NOLCK) &&
 557                     ((*flags) = (*flags) |
 558                                 ll_set_opt("nolock", this_char,
 559                                            LL_SBI_NOLCK)))
 560                         continue;
 561         }
 562
 563         EXIT;
 564 }
 565
 566 void ll_lli_init(struct ll_inode_info *lli)
 567 {
 568         sema_init(&lli->lli_open_sem, 1);
 569         sema_init(&lli->lli_size_sem, 1);
 570         lli->lli_flags = 0;
 571         lli->lli_size_pid = 0;
 572         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 573         spin_lock_init(&lli->lli_lock);
 574         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 575         INIT_LIST_HEAD(&lli->lli_close_item);
 576         lli->lli_inode_magic = LLI_INODE_MAGIC;
 577         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 578         sema_init(&lli->lli_och_sem, 1);
 579         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 580         lli->lli_mds_exec_och = NULL;
 581         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 582         lli->lli_open_fd_exec_count = 0;
 583         lli->lli_audit_mask = AUDIT_OFF;
 584         lli->lli_key_info = NULL;
 585 }
 586
 587 int ll_fill_super(struct super_block *sb, void *data, int silent)
 588 {
 589         struct ll_sb_info *sbi;
 590         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 591         char *mds_sec = NULL;
 592         char *oss_sec = NULL;
 593         int async, err;
 594         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 595         __u64 remote_flag = 0;
 596         ENTRY;
 597
 598         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 599
 600         sbi = lustre_init_sbi(sb);
 601         if (!sbi)
 602                 RETURN(-ENOMEM);
 603
 604         sbi->ll_flags |= LL_SBI_READAHEAD;
 605         ll_options(data, &lov, &lmv, &gkc, &mds_sec, &oss_sec,
 606                    &async, &sbi->ll_flags);
 607
 608         if (!lov || !lmv) {
 609                 CERROR("no osc %p or no mdc %p\n", lov, lmv);
 610                 GOTO(out, err = -EINVAL);
 611         }
 612
 613         err = lustre_common_fill_super(sb, lmv, lov, gkc, async, mds_sec,
 614                                        oss_sec, nllu, 0, &remote_flag);
 615         EXIT;
 616 out:
 617         if (err)
 618                 lustre_free_sbi(sb);
 619
 620         if (lmv)
 621                 OBD_FREE(lmv, strlen(lmv) + 1);
 622         if (lov)
 623                 OBD_FREE(lov, strlen(lov) + 1);
 624         if (mds_sec)
 625                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 626         if (oss_sec)
 627                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 628         if (gkc)
 629                 OBD_FREE(gkc, strlen(gkc) + 1);
 630
 631         return err;
 632 } /* ll_read_super */
 633
 634 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 635                               struct config_llog_instance *cfg, int allow_recov)
 636 {
 637         struct lustre_cfg *lcfg = NULL;
 638         struct lustre_cfg_bufs bufs;
 639         struct portals_cfg pcfg;
 640         char *peer = "MDS_PEER_UUID";
 641         struct obd_device *obd;
 642         struct lustre_handle md_conn = {0, };
 643         struct obd_export *exp;
 644         char *name = "mdc_dev";
 645         class_uuid_t uuid;
 646         struct obd_uuid lmv_uuid;
 647         struct llog_ctxt *ctxt;
 648         int rc = 0, err = 0;
 649         ENTRY;
 650
 651         if (lmd_bad_magic(lmd))
 652                 RETURN(-EINVAL);
 653
 654         generate_random_uuid(uuid);
 655         class_uuid_unparse(uuid, &lmv_uuid);
 656
 657         if (lmd->lmd_local_nid) {
 658                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 659                 pcfg.pcfg_nal = lmd->lmd_nal;
 660                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 661                 rc = libcfs_nal_cmd(&pcfg);
 662                 if (rc < 0)
 663                         GOTO(out, rc);
 664         }
 665
 666         if (lmd->lmd_nal == SOCKNAL ||
 667             lmd->lmd_nal == OPENIBNAL ||
 668             lmd->lmd_nal == IIBNAL ||
 669             lmd->lmd_nal == VIBNAL ||
 670             lmd->lmd_nal == RANAL) {
 671                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 672                 pcfg.pcfg_nal     = lmd->lmd_nal;
 673                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 674                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 675                 pcfg.pcfg_misc    = lmd->lmd_port;
 676                 rc = libcfs_nal_cmd(&pcfg);
 677                 if (rc < 0)
 678                         GOTO(out, rc);
 679         }
 680         lustre_cfg_bufs_reset(&bufs, name);
 681         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 682
 683         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 684         lcfg->lcfg_nal = lmd->lmd_nal;
 685         lcfg->lcfg_nid = lmd->lmd_server_nid;
 686         LASSERT(lcfg->lcfg_nal);
 687         LASSERT(lcfg->lcfg_nid);
 688         err = class_process_config(lcfg);
 689         lustre_cfg_free(lcfg);
 690         if (err < 0)
 691                 GOTO(out_del_conn, err);
 692
 693         lustre_cfg_bufs_reset(&bufs, name);
 694         lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
 695         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 696
 697         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 698         err = class_process_config(lcfg);
 699         lustre_cfg_free(lcfg);
 700         if (err < 0)
 701                 GOTO(out_del_uuid, err);
 702
 703         lustre_cfg_bufs_reset(&bufs, name);
 704         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 705         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 706
 707         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 708         err = class_process_config(lcfg);
 709         lustre_cfg_free(lcfg);
 710         if (err < 0)
 711                 GOTO(out_detach, err);
 712
 713         obd = class_name2obd(name);
 714         if (obd == NULL)
 715                 GOTO(out_cleanup, rc = -EINVAL);
 716
 717         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 718                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 719         if (rc)
 720                 GOTO(out_cleanup, rc);
 721
 722         if (lmd->lmd_pag) {
 723                 unsigned long sec_flags = PTLRPC_SEC_FL_PAG;
 724                 rc = obd_set_info(obd->obd_self_export,
 725                                   strlen("sec_flags"), "sec_flags",
 726                                   sizeof(sec_flags), &sec_flags);
 727                 if (rc)
 728                         GOTO(out_cleanup, rc);
 729         }
 730
 731         /* Disable initial recovery on this import */
 732         rc = obd_set_info(obd->obd_self_export,
 733                           strlen("initial_recov"), "initial_recov",
 734                           sizeof(allow_recov), &allow_recov);
 735         if (rc)
 736                 GOTO(out_cleanup, rc);
 737
 738         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 739         if (rc) {
 740                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 741                 GOTO(out_cleanup, rc);
 742         }
 743
 744         exp = class_conn2export(&md_conn);
 745
 746         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 747         rc = class_config_process_llog(ctxt, profile, cfg);
 748         if (rc)
 749                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 750
 751         err = obd_disconnect(exp, 0);
 752
 753         EXIT;
 754 out_cleanup:
 755         lustre_cfg_bufs_reset(&bufs, name);
 756         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 757         err = class_process_config(lcfg);
 758         lustre_cfg_free(lcfg);
 759         if (err < 0)
 760                 GOTO(out, err);
 761 out_detach:
 762         lustre_cfg_bufs_reset(&bufs, name);
 763         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 764         err = class_process_config(lcfg);
 765         lustre_cfg_free(lcfg);
 766         if (err < 0)
 767                 GOTO(out, err);
 768
 769 out_del_uuid:
 770         lustre_cfg_bufs_reset(&bufs, name);
 771         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 772         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 773         err = class_process_config(lcfg);
 774         lustre_cfg_free(lcfg);
 775
 776 out_del_conn:
 777         if (lmd->lmd_nal == SOCKNAL ||
 778             lmd->lmd_nal == OPENIBNAL ||
 779             lmd->lmd_nal == IIBNAL ||
 780             lmd->lmd_nal == VIBNAL ||
 781             lmd->lmd_nal == RANAL) {
 782                 int err2;
 783
 784                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 785                 pcfg.pcfg_nal     = lmd->lmd_nal;
 786                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 787                 pcfg.pcfg_flags   = 1;          /* single_share */
 788                 err2 = libcfs_nal_cmd(&pcfg);
 789                 if (err2 && !err)
 790                         err = err2;
 791                 if (err < 0)
 792                         GOTO(out, err);
 793         }
 794 out:
 795         if (rc == 0)
 796                 rc = err;
 797
 798         return rc;
 799 }
 800
 801 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 802 {
 803         struct lustre_cfg *lcfg;
 804         struct lustre_cfg_bufs bufs;
 805         struct obd_device *obd;
 806         int next = 0;
 807
 808         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL) {
 809                 int err;
 810
 811                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 812                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 813                 err = class_process_config(lcfg);
 814                 if (err) {
 815                         CERROR("cleanup failed: %s\n", obd->obd_name);
 816                         //continue;
 817                 }
 818
 819                 lcfg->lcfg_command = LCFG_DETACH;
 820                 err = class_process_config(lcfg);
 821                 lustre_cfg_free(lcfg);
 822                 if (err) {
 823                         CERROR("detach failed: %s\n", obd->obd_name);
 824                         //continue;
 825                 }
 826         }
 827
 828         if (sbi->ll_lmd != NULL)
 829                 class_del_profile(sbi->ll_lmd->lmd_profile);
 830 }
 831
 832 static int lustre_process_profile(struct super_block *sb,
 833                                   struct lustre_mount_data *lmd,
 834                                   char **lov, char **lmv, char **gkc)
 835 {
 836         struct ll_sb_info *sbi = ll_s2sbi(sb);
 837         struct config_llog_instance cfg;
 838         struct lustre_profile *lprof;
 839         int len, err = 0;
 840         ENTRY;
 841
 842         if (!lmd->lmd_profile)
 843                 RETURN(0);
 844
 845         if (lmd->lmd_mds[0] == '\0') {
 846                 CERROR("no mds name\n");
 847                 GOTO(out, err = -EINVAL);
 848         }
 849         lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
 850         lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
 851
 852         OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 853         if (sbi->ll_lmd == NULL)
 854                 GOTO(out, err = -ENOMEM);
 855         memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 856
 857         /* generate a string unique to this super, let's try the address of the
 858          * super itself. */
 859         len = (sizeof(sb) * 2) + 1;
 860         OBD_ALLOC(sbi->ll_instance, len);
 861         if (sbi->ll_instance == NULL)
 862                 GOTO(out, err = -ENOMEM);
 863         sprintf(sbi->ll_instance, "%p", sb);
 864
 865         cfg.cfg_instance = sbi->ll_instance;
 866         cfg.cfg_uuid = sbi->ll_sb_uuid;
 867         cfg.cfg_local_nid = lmd->lmd_local_nid;
 868         err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 869         if (err < 0) {
 870                 CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 871                 GOTO(out, err);
 872         }
 873
 874         lprof = class_get_profile(lmd->lmd_profile);
 875         if (lprof == NULL) {
 876                 CERROR("No profile found: %s\n", lmd->lmd_profile);
 877                 GOTO(out, err = -EINVAL);
 878         }
 879
 880         OBD_ALLOC(*lov, strlen(lprof->lp_lov) +
 881                   strlen(sbi->ll_instance) + 2);
 882         if (*lov = NULL)
 883                 GOTO(out, err = -ENOMEM);
 884
 885         sprintf(*lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 886
 887         OBD_ALLOC(*lmv, strlen(lprof->lp_lmv) +
 888                   strlen(sbi->ll_instance) + 2);
 889         if (*lmv == NULL)
 890                 GOTO(out_free_lov, err = -ENOMEM);
 891
 892         sprintf(*lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 893
 894         if (lprof->lp_gkc) {
 895                OBD_ALLOC(*gkc, strlen(lprof->lp_gkc) +
 896                          strlen(sbi->ll_instance) + 2);
 897                if (*gkc == NULL)
 898                        GOTO(out_free_lmv, err = -ENOMEM);
 899
 900                sprintf(*gkc, "%s-%s", lprof->lp_gkc, sbi->ll_instance);
 901         }
 902
 903         RETURN(err);
 904 out_free_lmv:
 905         OBD_FREE(*lmv, strlen(lprof->lp_lmv) +
 906                  strlen(sbi->ll_instance) + 2);
 907 out_free_lov:
 908         OBD_FREE(*lov, strlen(lprof->lp_lov) +
 909                  strlen(sbi->ll_instance) + 2);
 910 out:
 911         return err;
 912 }
 913
 914 static int lustre_clean_profile(struct ll_sb_info *sbi, int force_umount)
 915 {
 916         struct lustre_mount_data *lmd = sbi->ll_lmd;
 917         struct config_llog_instance cfg;
 918         char *cl_prof;
 919         int len, err = 0;
 920         ENTRY;
 921
 922         if (!lmd)
 923                 RETURN(err);
 924
 925         len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean") + 1;
 926
 927         if (force_umount) {
 928                 CERROR("force umount, doing manual cleanup\n");
 929                 lustre_manual_cleanup(sbi);
 930                 GOTO(free_lmd, 0);
 931
 932         }
 933         if (sbi->ll_instance != NULL) {
 934                 cfg.cfg_instance = sbi->ll_instance;
 935                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 936
 937                 OBD_ALLOC(cl_prof, len);
 938                 if (!cl_prof) {
 939                         CERROR("can't allocate memory, "
 940                                "skipping processing cleanup profile.\n");
 941                         GOTO(free_lmd, err = -ENOMEM);
 942                 }
 943
 944                 sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 945                 err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 946                 if (err < 0) {
 947                         CERROR("Unable to process log: %s\n", cl_prof);
 948                         lustre_manual_cleanup(sbi);
 949                 }
 950                 OBD_FREE(cl_prof, len);
 951         }
 952         EXIT;
 953 free_lmd:
 954         if (sbi->ll_instance)
 955                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 956         OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 957         return err;
 958 }
 959
 960 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 961 {
 962         struct lustre_mount_data * lmd = data;
 963         char *lov = NULL, *lmv = NULL, *gkc = NULL;
 964         struct ll_sb_info *sbi;
 965         int err;
 966         ENTRY;
 967
 968         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 969         if (lmd_bad_magic(lmd))
 970                 RETURN(-EINVAL);
 971
 972         sbi = lustre_init_sbi(sb);
 973         if (!sbi)
 974                 RETURN(-ENOMEM);
 975
 976         sbi->ll_flags |= LL_SBI_READAHEAD;
 977
 978         err = lustre_process_profile(sb, lmd, &lov, &lmv, &gkc);
 979         if (err) {
 980                 CERROR("Can not process the profile err %d \n", err);
 981                 GOTO(out_free, err);
 982         }
 983         if (!lov || !lmv) {
 984                 CERROR("no osc %p or no mdc %p \n", lov, lmv);
 985                 GOTO(out_free, err = -EINVAL);
 986         }
 987
 988         err = lustre_common_fill_super(sb, lmv, lov, gkc, lmd->lmd_async,
 989                                        lmd->lmd_mds_security,
 990                                        lmd->lmd_oss_security,
 991                                        &lmd->lmd_nllu, lmd->lmd_pag,
 992                                        &lmd->lmd_remote_flag);
 993
 994         if (err)
 995                 GOTO(out_free, err);
 996
 997         EXIT;
 998 out_dev:
 999         if (lmv)
1000                 OBD_FREE(lmv, strlen(lmv) + 1);
1001         if (lov)
1002                 OBD_FREE(lov, strlen(lov) + 1);
1003         if (gkc)
1004                 OBD_FREE(gkc, strlen(gkc) + 1);
1005
1006         return err;
1007 out_free:
1008         lustre_clean_profile(sbi, 0);
1009         lustre_free_sbi(sb);
1010         goto out_dev;
1011
1012 } /* lustre_fill_super */
1013
1014 void lustre_put_super(struct super_block *sb)
1015 {
1016         struct obd_device *obd;
1017         struct ll_sb_info *sbi = ll_s2sbi(sb);
1018         int force_umount = 0;
1019         ENTRY;
1020
1021         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1022         obd = class_exp2obd(sbi->ll_md_exp);
1023         if (obd)
1024                 force_umount = obd->obd_no_recov;
1025         obd = NULL;
1026
1027         lustre_common_put_super(sb);
1028         lustre_clean_profile(sbi, force_umount);
1029         lustre_free_sbi(sb);
1030
1031         EXIT;
1032 } /* lustre_put_super */
1033
1034 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
1035 {
1036         struct lustre_mount_data *lmd = sbi->ll_lmd;
1037         char *profile = lmd->lmd_profile, *name = NULL;
1038         struct config_llog_instance cfg;
1039         int rc, namelen =  0, version;
1040         struct llog_ctxt *ctxt;
1041         ENTRY;
1042
1043         if (profile == NULL)
1044                 RETURN(0);
1045         if (lmd == NULL) {
1046                 CERROR("Client not mounted with zero-conf; cannot "
1047                        "process update log.\n");
1048                 RETURN(0);
1049         }
1050
1051         rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
1052                                LDLM_FL_CONFIG_CHANGE, NULL);
1053         if (rc != 0)
1054                 CWARN("obd_cancel_unused(mdc): %d\n", rc);
1055
1056         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
1057                                LDLM_FL_CONFIG_CHANGE, NULL);
1058         if (rc != 0)
1059                 CWARN("obd_cancel_unused(lov): %d\n", rc);
1060
1061         cfg.cfg_instance = sbi->ll_instance;
1062         cfg.cfg_uuid = sbi->ll_sb_uuid;
1063         cfg.cfg_local_nid = lmd->lmd_local_nid;
1064
1065         namelen = strlen(profile) + 20; /* -clean-######### */
1066         OBD_ALLOC(name, namelen);
1067         if (name == NULL)
1068                 RETURN(-ENOMEM);
1069
1070         if (clean) {
1071                 version = sbi->ll_config_version - 1;
1072                 sprintf(name, "%s-clean-%d", profile, version);
1073         } else {
1074                 version = sbi->ll_config_version + 1;
1075                 sprintf(name, "%s-%d", profile, version);
1076         }
1077
1078         CWARN("Applying configuration log %s\n", name);
1079
1080         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
1081                                 LLOG_CONFIG_REPL_CTXT);
1082         rc = class_config_process_llog(ctxt, name, &cfg);
1083         if (rc == 0)
1084                 sbi->ll_config_version = version;
1085         CWARN("Finished applying configuration log %s: %d\n", name, rc);
1086
1087         if (rc == 0 && clean == 0) {
1088                 struct lov_desc desc;
1089                 __u32 valsize;
1090                 int rc = 0;
1091
1092                 valsize = sizeof(desc);
1093                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
1094                                   "lovdesc", &valsize, &desc);
1095
1096                 rc = obd_init_ea_size(sbi->ll_md_exp,
1097                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
1098                                       (desc.ld_tgt_count *
1099                                        sizeof(struct llog_cookie)));
1100         }
1101         OBD_FREE(name, namelen);
1102         RETURN(rc);
1103 }
1104
1105 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1106 {
1107         struct inode *inode = NULL;
1108
1109         /* NOTE: we depend on atomic igrab() -bzzz */
1110         lock_res_and_lock(lock);
1111         if (lock->l_ast_data) {
1112                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1113                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1114                         inode = igrab(lock->l_ast_data);
1115                 } else {
1116                         struct timeval now;
1117                         do_gettimeofday(&now);
1118                         inode = lock->l_ast_data;
1119                         LDLM_ERROR(lock, "granted at %lu.%lu, now %lu.%lu",
1120                                    lock->l_enqueued_time.tv_sec,
1121                                    lock->l_enqueued_time.tv_usec,
1122                                    now.tv_sec, now.tv_usec);
1123                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
1124                                "l_ast_data %p is bogus: magic %0x8\n",
1125                                lock->l_ast_data, lli->lli_inode_magic);
1126                         CDEBUG(D_ERROR, "i_state = 0x%lx, l_ast_data %p is bogus: magic %0x8\n",
1127                                inode->i_state, lock->l_ast_data, lli->lli_inode_magic);
1128                         inode = NULL;
1129                         unlock_res_and_lock(lock);
1130                         LBUG();
1131                 }
1132         }
1133         unlock_res_and_lock(lock);
1134         return inode;
1135 }
1136
1137 int null_if_equal(struct ldlm_lock *lock, void *data)
1138 {
1139         if (data == lock->l_ast_data) {
1140                 lock->l_ast_data = NULL;
1141
1142                 if (lock->l_req_mode != lock->l_granted_mode)
1143                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1144         }
1145
1146         return LDLM_ITER_CONTINUE;
1147 }
1148
1149 static void remote_acl_free(struct remote_acl *racl);
1150
1151 void ll_clear_inode(struct inode *inode)
1152 {
1153         struct lustre_id id;
1154         struct ll_inode_info *lli = ll_i2info(inode);
1155         struct ll_sb_info *sbi = ll_i2sbi(inode);
1156         ENTRY;
1157
1158         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1159                inode->i_generation, inode);
1160
1161         ll_inode2id(&id, inode);
1162
1163         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1164         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1165
1166         LASSERT(!lli->lli_open_fd_write_count);
1167         LASSERT(!lli->lli_open_fd_read_count);
1168         LASSERT(!lli->lli_open_fd_exec_count);
1169         if (lli->lli_mds_write_och)
1170                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1171         if (lli->lli_mds_exec_och)
1172                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1173         if (lli->lli_mds_read_och)
1174                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1175         if (lli->lli_smd)
1176                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1177                                   null_if_equal, inode);
1178
1179         if (lli->lli_smd) {
1180                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1181                 lli->lli_smd = NULL;
1182         }
1183
1184         if (lli->lli_mea) {
1185                 obd_free_memmd(sbi->ll_md_exp,
1186                                (struct lov_stripe_md **) &lli->lli_mea);
1187                 lli->lli_mea = NULL;
1188         }
1189         ll_crypto_destroy_inode_key(inode);
1190         if (lli->lli_symlink_name) {
1191                 OBD_FREE(lli->lli_symlink_name,
1192                          strlen(lli->lli_symlink_name) + 1);
1193                 lli->lli_symlink_name = NULL;
1194         }
1195
1196         if (lli->lli_posix_acl) {
1197                 LASSERT(lli->lli_remote_acl == NULL);
1198                 posix_acl_release(lli->lli_posix_acl);
1199                 lli->lli_posix_acl = NULL;
1200         }
1201
1202         if (lli->lli_remote_acl) {
1203                 LASSERT(lli->lli_posix_acl == NULL);
1204                 remote_acl_free(lli->lli_remote_acl);
1205                 lli->lli_remote_acl = NULL;
1206         }
1207
1208         if (lli->lli_trunc_capa) {
1209                 OBD_FREE(lli->lli_trunc_capa, sizeof(struct lustre_capa));
1210                 lli->lli_trunc_capa = NULL;
1211         }
1212
1213         lli->lli_inode_magic = LLI_INODE_DEAD;
1214         EXIT;
1215 }
1216
1217 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1218  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1219  * keep these values until such a time that objects are allocated for it.
1220  * We do the MDS operations first, as it is checking permissions for us.
1221  * We don't to the MDS RPC if there is nothing that we want to store there,
1222  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1223  * going to do an RPC anyways.
1224  *
1225  * If we are doing a truncate, we will send the mtime and ctime updates
1226  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1227  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1228  * at the same time.
1229  */
1230 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1231 {
1232         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1233         struct ll_inode_info *lli = ll_i2info(inode);
1234         struct ll_sb_info *sbi = ll_i2sbi(inode);
1235         struct ptlrpc_request *request = NULL;
1236         struct mdc_op_data *op_data;
1237         struct lustre_capa *trunc_capa = NULL;
1238         int ia_valid = attr->ia_valid;
1239         int err, rc = 0;
1240         ENTRY;
1241
1242         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1243         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1244
1245         if (ia_valid & ATTR_SIZE) {
1246                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1247                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1248                                attr->ia_size, ll_file_maxbytes(inode));
1249                         RETURN(-EFBIG);
1250                 }
1251
1252                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1253         }
1254
1255         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1256         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1257                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1258                         RETURN(-EPERM);
1259         }
1260
1261         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1262         if (attr->ia_valid & ATTR_CTIME) {
1263                 attr->ia_ctime = CURRENT_TIME;
1264                 attr->ia_valid |= ATTR_CTIME_SET;
1265         }
1266         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1267                 attr->ia_atime = CURRENT_TIME;
1268                 attr->ia_valid |= ATTR_ATIME_SET;
1269         }
1270         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1271                 attr->ia_mtime = CURRENT_TIME;
1272                 attr->ia_valid |= ATTR_MTIME_SET;
1273         }
1274
1275         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1276                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1277                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1278                        LTIME_S(CURRENT_TIME));
1279
1280         /* If only OST attributes being set on objects, don't do MDS RPC.
1281          * In that case, we need to check permissions and update the local
1282          * inode ourselves so we can call obdo_from_inode() always. */
1283         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1284                 struct lustre_md md;
1285                 void *key = NULL;
1286                 int  key_size = 0;
1287
1288                 OBD_ALLOC(op_data, sizeof(*op_data));
1289                 if (op_data == NULL)
1290                         RETURN(-ENOMEM);
1291                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
1292
1293                 if (ia_valid & (ATTR_UID | ATTR_GID)) {
1294                         rc = ll_crypto_get_mac(inode, attr, NULL, 0, &key,
1295                                                &key_size);
1296                 }
1297                 rc = md_setattr(sbi->ll_md_exp, op_data,
1298                                 attr, key, key_size, NULL, 0, NULL,
1299                                 0, &request);
1300                 OBD_FREE(op_data, sizeof(*op_data));
1301
1302                 if (key && key_size)
1303                         OBD_FREE(key, key_size);
1304                 if (rc) {
1305                         ptlrpc_req_finished(request);
1306                         if (rc != -EPERM && rc != -EACCES)
1307                                 CERROR("md_setattr fails: rc = %d\n", rc);
1308                         RETURN(rc);
1309                 }
1310                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1311                                        sbi->ll_dt_exp, &md);
1312                 if (rc) {
1313                         ptlrpc_req_finished(request);
1314                         RETURN(rc);
1315                 }
1316
1317                 if (attr->ia_valid & ATTR_SIZE) {
1318                         /* XXX: hack for truncate capa */
1319                         rc = mdc_req2lustre_capa(request, 0, &trunc_capa);
1320                         if (rc) {
1321                                 ptlrpc_req_finished(request);
1322                                 RETURN(rc);
1323                         }
1324
1325                         spin_lock(&lli->lli_lock);
1326                         if (trunc_capa) {
1327                                 if (lli->lli_trunc_capa)
1328                                         OBD_FREE(lli->lli_trunc_capa,
1329                                                  sizeof(*trunc_capa));
1330                                 lli->lli_trunc_capa = trunc_capa;
1331                         }
1332                         spin_unlock(&lli->lli_lock);
1333                 }
1334
1335                 /* We call inode_setattr to adjust timestamps, but we first
1336                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1337                  *
1338                  * NB: ATTR_SIZE will only be set at this point if the size
1339                  * resides on the MDS, ie, this file has no objects. */
1340                 attr->ia_valid &= ~ATTR_SIZE;
1341
1342                 /*
1343                  * assigning inode_setattr() to @err to disable warning that
1344                  * function's result should be checked by by caller. error is
1345                  * impossible here, as vmtruncate() control path is disabled.
1346                  */
1347                 err = inode_setattr(inode, attr);
1348                 ll_update_inode(inode, &md);
1349                 ptlrpc_req_finished(request);
1350
1351                 if (!lsm || !S_ISREG(inode->i_mode)) {
1352                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1353                         RETURN(0);
1354                 }
1355         } else {
1356                 /* The OST doesn't check permissions, but the alternative is
1357                  * a gratuitous RPC to the MDS.  We already rely on the client
1358                  * to do read/write/truncate permission checks, so is mtime OK?
1359                  */
1360                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1361                         /* from sys_utime() */
1362                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1363                                 if (current->fsuid != inode->i_uid &&
1364                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1365                                         RETURN(rc);
1366                         } else {
1367                                 /* from inode_change_ok() */
1368                                 if (current->fsuid != inode->i_uid &&
1369                                     !capable(CAP_FOWNER))
1370                                         RETURN(-EPERM);
1371                         }
1372                 }
1373
1374                 if (lsm)
1375                         attr->ia_valid &= ~ATTR_SIZE;
1376
1377                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1378                 err = inode_setattr(inode, attr);
1379                 /*
1380                  * assigning inode_setattr() to @err to disable warning that
1381                  * function's result should be checked by by caller. error is
1382                  * impossible here, as vmtruncate() control path is disabled.
1383                  */
1384         }
1385
1386         /* We really need to get our PW lock before we change inode->i_size.
1387          * If we don't we can race with other i_size updaters on our node, like
1388          * ll_file_read.  We can also race with i_size propogation to other
1389          * nodes through dirtying and writeback of final cached pages.  This
1390          * last one is especially bad for racing o_append users on other
1391          * nodes. */
1392         if (ia_valid & ATTR_SIZE) {
1393                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1394                                                            OBD_OBJECT_EOF } };
1395                 struct lustre_handle lockh = { 0 };
1396                 int err, ast_flags = 0;
1397                 /* XXX when we fix the AST intents to pass the discard-range
1398                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1399                  * XXX here. */
1400                 if (attr->ia_size == 0)
1401                         ast_flags = LDLM_AST_DISCARD_DATA;
1402
1403                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1404                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1405
1406                 if (rc != 0)
1407                         RETURN(rc);
1408
1409                 down(&lli->lli_size_sem);
1410                 lli->lli_size_pid = current->pid;
1411                 rc = vmtruncate(inode, attr->ia_size);
1412                 if (rc != 0) {
1413                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1414                         lli->lli_size_pid = 0;
1415                         up(&lli->lli_size_sem);
1416                 }
1417
1418                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1419                 if (err) {
1420                         CERROR("ll_extent_unlock failed: %d\n", err);
1421                         if (!rc)
1422                                 rc = err;
1423                 }
1424         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1425                 struct obdo *oa = NULL;
1426
1427                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1428                        inode->i_ino, LTIME_S(attr->ia_mtime));
1429
1430                 oa = obdo_alloc();
1431                 if (oa == NULL)
1432                         RETURN(-ENOMEM);
1433
1434                 oa->o_id = lsm->lsm_object_id;
1435                 oa->o_gr = lsm->lsm_object_gr;
1436                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1437
1438                 /* adding uid and gid, needed for quota */
1439                 if (ia_valid & ATTR_UID) {
1440                         oa->o_uid = inode->i_uid;
1441                         oa->o_valid |= OBD_MD_FLUID;
1442                 }
1443
1444                 if (ia_valid & ATTR_GID) {
1445                         oa->o_gid = inode->i_gid;
1446                         oa->o_valid |= OBD_MD_FLGID;
1447                 }
1448
1449                 *(obdo_id(oa)) = lli->lli_id;
1450                 oa->o_valid |= OBD_MD_FLIFID;
1451
1452                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1453                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1454                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1455                 obdo_free(oa);
1456                 if (rc)
1457                         CERROR("obd_setattr fails: rc = %d\n", rc);
1458         }
1459
1460         RETURN(rc);
1461 }
1462
1463 int ll_setattr(struct dentry *de, struct iattr *attr)
1464 {
1465         LASSERT(de->d_inode);
1466         return ll_setattr_raw(de->d_inode, attr);
1467 }
1468
1469 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1470                        unsigned long max_age)
1471 {
1472         struct ll_sb_info *sbi = ll_s2sbi(sb);
1473         struct obd_statfs obd_osfs;
1474         int rc;
1475         ENTRY;
1476
1477         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1478         if (rc) {
1479                 CERROR("obd_statfs fails: rc = %d\n", rc);
1480                 RETURN(rc);
1481         }
1482
1483         osfs->os_type = sb->s_magic;
1484
1485         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1486                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1487
1488         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1489         if (rc) {
1490                 CERROR("obd_statfs fails: rc = %d\n", rc);
1491                 RETURN(rc);
1492         }
1493
1494         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1495                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1496                obd_osfs.os_files);
1497
1498         osfs->os_blocks = obd_osfs.os_blocks;
1499         osfs->os_bfree = obd_osfs.os_bfree;
1500         osfs->os_bavail = obd_osfs.os_bavail;
1501
1502         /* If we don't have as many objects free on the OST as inodes
1503          * on the MDS, we reduce the total number of inodes to
1504          * compensate, so that the "inodes in use" number is correct.
1505          */
1506         if (obd_osfs.os_ffree < osfs->os_ffree) {
1507                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1508                         obd_osfs.os_ffree;
1509                 osfs->os_ffree = obd_osfs.os_ffree;
1510         }
1511
1512         RETURN(rc);
1513 }
1514
1515 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1516 {
1517         struct obd_statfs osfs;
1518         int rc;
1519
1520         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1521         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1522
1523         /* For now we will always get up-to-date statfs values, but in the
1524          * future we may allow some amount of caching on the client (e.g.
1525          * from QOS or lprocfs updates). */
1526         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1527         if (rc)
1528                 return rc;
1529
1530         statfs_unpack(sfs, &osfs);
1531
1532         if (sizeof(sfs->f_blocks) == 4) {
1533                 while (osfs.os_blocks > ~0UL) {
1534                         sfs->f_bsize <<= 1;
1535
1536                         osfs.os_blocks >>= 1;
1537                         osfs.os_bfree >>= 1;
1538                         osfs.os_bavail >>= 1;
1539                 }
1540         }
1541
1542         sfs->f_blocks = osfs.os_blocks;
1543         sfs->f_bfree = osfs.os_bfree;
1544         sfs->f_bavail = osfs.os_bavail;
1545
1546         return 0;
1547 }
1548
1549
1550 /********************************
1551  * remote acl                   *
1552  ********************************/
1553
1554 static struct remote_acl *remote_acl_alloc(void)
1555 {
1556         struct remote_acl *racl;
1557         int i;
1558
1559         OBD_ALLOC(racl, sizeof(*racl));
1560         if (!racl)
1561                 return NULL;
1562
1563         spin_lock_init(&racl->ra_lock);
1564         init_MUTEX(&racl->ra_update_sem);
1565
1566         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1567                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1568
1569         return racl;
1570 }
1571
1572 /*
1573  * caller should guarantee no race here.
1574  */
1575 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1576 {
1577         struct remote_perm_setxid *xperm;
1578
1579         while (!list_empty(&perm->lrp_setxid_perms)) {
1580                 xperm = list_entry(perm->lrp_setxid_perms.next,
1581                                    struct remote_perm_setxid,
1582                                    list);
1583                 list_del(&xperm->list);
1584                 OBD_FREE(xperm, sizeof(*xperm));
1585         }
1586 }
1587
1588 /*
1589  * caller should guarantee no race here.
1590  */
1591 static void remote_acl_flush(struct remote_acl *racl)
1592 {
1593         struct list_head *head;
1594         struct lustre_remote_perm *perm, *tmp;
1595         int i;
1596
1597         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1598                 head = &racl->ra_perm_cache[i];
1599
1600                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1601                         remote_perm_flush_xperms(perm);
1602                         list_del(&perm->lrp_list);
1603                         OBD_FREE(perm, sizeof(*perm));
1604                 }
1605         }
1606 }
1607
1608 static void remote_acl_free(struct remote_acl *racl)
1609 {
1610         if (!racl)
1611                 return;
1612
1613         down(&racl->ra_update_sem);
1614         spin_lock(&racl->ra_lock);
1615         remote_acl_flush(racl);
1616         spin_unlock(&racl->ra_lock);
1617         up(&racl->ra_update_sem);
1618
1619         OBD_FREE(racl, sizeof(*racl));
1620 }
1621
1622 static inline int remote_acl_hashfunc(__u32 id)
1623 {
1624         return (id & (REMOTE_ACL_HASHSIZE - 1));
1625 }
1626
1627 static
1628 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1629 {
1630         struct list_head *head;
1631         struct lustre_remote_perm *lperm;
1632         struct remote_perm_setxid *xperm;
1633         int found = 0, rc = -ENOENT;
1634
1635         LASSERT(racl);
1636         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1637         spin_lock(&racl->ra_lock);
1638
1639         list_for_each_entry(lperm, head, lrp_list) {
1640                 if (lperm->lrp_auth_uid == current->uid) {
1641                         found = 1;
1642                         break;
1643                 }
1644         }
1645
1646         if (!found)
1647                 goto out;
1648
1649         if (lperm->lrp_auth_uid == current->fsuid &&
1650             lperm->lrp_auth_gid == current->fsgid) {
1651                 if (lperm->lrp_valid) {
1652                         *perm = lperm->lrp_perm;
1653                         rc = 0;
1654                 }
1655                 goto out;
1656         } else if ((!lperm->lrp_setuid &&
1657                     lperm->lrp_auth_uid != current->fsuid) ||
1658                    (!lperm->lrp_setgid &&
1659                     lperm->lrp_auth_gid != current->fsgid))  {
1660                 *perm = 0;
1661                 rc = 0;
1662                 goto out;
1663         }
1664
1665         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1666                 if (xperm->uid == current->fsuid &&
1667                     xperm->gid == current->fsgid) {
1668                         *perm = xperm->perm;
1669                         rc = 0;
1670                         goto out;
1671                 }
1672         }
1673
1674 out:
1675         spin_unlock(&racl->ra_lock);
1676         return rc;
1677 }
1678
1679 static
1680 int __remote_acl_update(struct remote_acl *racl,
1681                         struct mds_remote_perm *mperm,
1682                         struct lustre_remote_perm *lperm,
1683                         struct remote_perm_setxid *xperm)
1684 {
1685         struct list_head *head;
1686         struct lustre_remote_perm *lp;
1687         struct remote_perm_setxid *xp;
1688         int found = 0, setuid = 0, setgid = 0;
1689
1690         LASSERT(racl);
1691         LASSERT(mperm);
1692         LASSERT(lperm);
1693         LASSERT(current->uid == mperm->mrp_auth_uid);
1694
1695         if (current->fsuid != mperm->mrp_auth_uid)
1696                 setuid = 1;
1697         if (current->fsgid != mperm->mrp_auth_gid)
1698                 setgid = 1;
1699
1700         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1701         spin_lock(&racl->ra_lock);
1702
1703         list_for_each_entry(lp, head, lrp_list) {
1704                 if (lp->lrp_auth_uid == current->uid) {
1705                         found = 1;
1706                         break;
1707                 }
1708         }
1709
1710         if (found) {
1711                 OBD_FREE(lperm, sizeof(*lperm));
1712
1713                 if (!lp->lrp_valid && !setuid && !setgid) {
1714                         lp->lrp_perm = mperm->mrp_perm;
1715                         lp->lrp_valid = 1;
1716                 }
1717
1718                 /* sanity check for changes of setxid rules */
1719                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1720                         CWARN("setuid changes: %d => %d\n",
1721                               (lp->lrp_setuid != 0),
1722                               (mperm->mrp_allow_setuid != 0));
1723                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1724                 }
1725
1726                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1727                         CWARN("setgid changes: %d => %d\n",
1728                               (lp->lrp_setgid != 0),
1729                               (mperm->mrp_allow_setgid != 0));
1730                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1731                 }
1732
1733                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1734                     !list_empty(&lp->lrp_setxid_perms)) {
1735                         remote_perm_flush_xperms(lp);
1736                 }
1737         } else {
1738                 /* initialize lperm and linked into hashtable
1739                  */
1740                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1741                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1742                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1743                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1744                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1745                 list_add(&lperm->lrp_list, head);
1746
1747                 if (!setuid && !setgid) {
1748                         /* in this case, i'm the authenticated user,
1749                          * and mrp_perm is for me.
1750                          */
1751                         lperm->lrp_perm = mperm->mrp_perm;
1752                         lperm->lrp_valid = 1;
1753                         spin_unlock(&racl->ra_lock);
1754
1755                         if (xperm)
1756                                 OBD_FREE(xperm, sizeof(*xperm));
1757                         return 0;
1758                 }
1759
1760                 lp = lperm;
1761                 /* fall through */
1762         }
1763
1764         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1765                 list_empty(&lp->lrp_setxid_perms));
1766
1767         /* if no xperm supplied, we are all done here */
1768         if (!xperm) {
1769                 spin_unlock(&racl->ra_lock);
1770                 return 0;
1771         }
1772
1773         /* whether we allow setuid/setgid */
1774         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1775                 OBD_FREE(xperm, sizeof(*xperm));
1776                 spin_unlock(&racl->ra_lock);
1777                 return 0;
1778         }
1779
1780         /* traverse xperm list */
1781         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1782                 if (xp->uid == current->fsuid &&
1783                     xp->gid == current->fsgid) {
1784                         if (xp->perm != mperm->mrp_perm) {
1785                                 /* actually this should not happen */
1786                                 CWARN("perm changed: %o => %o\n",
1787                                       xp->perm, mperm->mrp_perm);
1788                                 xp->perm = mperm->mrp_perm;
1789                         }
1790                         OBD_FREE(xperm, sizeof(*xperm));
1791                         spin_unlock(&racl->ra_lock);
1792                         return 0;
1793                 }
1794         }
1795
1796         /* finally insert this xperm */
1797         xperm->uid = current->fsuid;
1798         xperm->gid = current->fsgid;
1799         xperm->perm = mperm->mrp_perm;
1800         list_add(&xperm->list, &lp->lrp_setxid_perms);
1801
1802         spin_unlock(&racl->ra_lock);
1803         return 0;
1804 }
1805
1806 /*
1807  * remote_acl semaphore must be held by caller
1808  */
1809 static
1810 int remote_acl_update_locked(struct remote_acl *racl,
1811                              struct mds_remote_perm *mperm)
1812 {
1813         struct lustre_remote_perm *lperm;
1814         struct remote_perm_setxid *xperm;
1815         int setuid = 0, setgid = 0;
1816
1817         might_sleep();
1818
1819         if (current->uid != mperm->mrp_auth_uid) {
1820                 CERROR("current uid %u while authenticated as %u\n",
1821                        current->uid, mperm->mrp_auth_uid);
1822                 return -EINVAL;
1823         }
1824
1825         if (current->fsuid != mperm->mrp_auth_uid)
1826                 setuid = 1;
1827         if (current->fsgid == mperm->mrp_auth_gid)
1828                 setgid = 1;
1829
1830         OBD_ALLOC(lperm, sizeof(*lperm));
1831         if (!lperm)
1832                 return -ENOMEM;
1833
1834         if ((setuid || setgid) &&
1835             !(setuid && !mperm->mrp_allow_setuid) &&
1836             !(setgid && !mperm->mrp_allow_setgid)) {
1837                 OBD_ALLOC(xperm, sizeof(*xperm));
1838                 if (!xperm) {
1839                         OBD_FREE(lperm, sizeof(*lperm));
1840                         return -ENOMEM;
1841                 }
1842         } else
1843                 xperm = NULL;
1844
1845         return __remote_acl_update(racl, mperm, lperm, xperm);
1846 }
1847
1848 /*
1849  * return -EACCES at any error cases
1850  */
1851 int ll_remote_acl_permission(struct inode *inode, int mode)
1852 {
1853         struct ll_sb_info *sbi = ll_i2sbi(inode);
1854         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1855         struct ptlrpc_request *req = NULL;
1856         struct lustre_id id;
1857         struct mds_remote_perm *mperm;
1858         int rc = -EACCES, perm;
1859
1860         if (!racl)
1861                 return -EACCES;
1862
1863         if (__remote_acl_check(racl, &perm) == 0) {
1864                 return ((perm & mode) == mode ? 0 : -EACCES);
1865         }
1866
1867         might_sleep();
1868
1869         /* doing update
1870          */
1871         down(&racl->ra_update_sem);
1872
1873         /* we might lose the race when obtain semaphore,
1874          * so check again.
1875          */
1876         if (__remote_acl_check(racl, &perm) == 0) {
1877                 if ((perm & mode) == mode)
1878                         rc = 0;
1879                 goto out;
1880         }
1881
1882         /* really fetch from mds
1883          */
1884         ll_inode2id(&id, inode);
1885         if (md_access_check(sbi->ll_md_exp, &id, &req))
1886                 goto out;
1887
1888         /* status non-zero indicate there's more apparent error
1889          * detected by mds, e.g. didn't allow this user at all.
1890          * we simply ignore and didn't cache it.
1891          */
1892         if (req->rq_repmsg->status)
1893                 goto out;
1894
1895         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1896                                    lustre_swab_remote_perm);
1897         LASSERT(mperm);
1898         LASSERT_REPSWABBED(req, 1);
1899
1900         if ((mperm->mrp_perm & mode) == mode)
1901                 rc = 0;
1902
1903         remote_acl_update_locked(racl, mperm);
1904 out:
1905         if (req)
1906                 ptlrpc_req_finished(req);
1907
1908         up(&racl->ra_update_sem);
1909         return rc;
1910 }
1911
1912 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1913 {
1914         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1915         int rc;
1916
1917         LASSERT(perm);
1918
1919         if (!racl)
1920                 return -EACCES;
1921
1922         down(&racl->ra_update_sem);
1923         rc = remote_acl_update_locked(racl, perm);
1924         up(&racl->ra_update_sem);
1925
1926         return rc;
1927 }
1928
1929 void ll_inode_invalidate_acl(struct inode *inode)
1930 {
1931         struct ll_sb_info *sbi = ll_i2sbi(inode);
1932         struct ll_inode_info *lli = ll_i2info(inode);
1933
1934         if (sbi->ll_remote) {
1935                 struct remote_acl *racl = lli->lli_remote_acl;
1936
1937                 LASSERT(!lli->lli_posix_acl);
1938                 if (racl) {
1939                         down(&racl->ra_update_sem);
1940                         spin_lock(&racl->ra_lock);
1941                         remote_acl_flush(lli->lli_remote_acl);
1942                         spin_unlock(&racl->ra_lock);
1943                         up(&racl->ra_update_sem);
1944                 }
1945         } else {
1946                 /* we can't invalide acl here: suppose we touch a new file
1947                  * under a dir, blocking ast on dir will lead to open failure
1948                  * on client, although succeed on mds. it's kind of weird,
1949                  * the real fix i think is improve client-vfs interaction.
1950                  *
1951                  * currently we just do nothing here.
1952                  */
1953                 return;
1954
1955                 LASSERT(!lli->lli_remote_acl);
1956                 spin_lock(&lli->lli_lock);
1957                 posix_acl_release(lli->lli_posix_acl);
1958                 lli->lli_posix_acl = NULL;
1959                 spin_unlock(&lli->lli_lock);
1960         }
1961 }
1962
1963 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1964 {
1965         struct ll_inode_info *lli = ll_i2info(inode);
1966         struct lov_stripe_md *lsm = md->lsm;
1967         struct mds_body *body = md->body;
1968         struct mea *mea = md->mea;
1969         struct posix_acl *posix_acl = md->posix_acl;
1970         struct ll_sb_info *sbi = ll_i2sbi(inode);
1971         struct lustre_key *mkey = md->key;
1972         ENTRY;
1973
1974         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1975
1976         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1977                 /* check for default striping info for dir. */
1978                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1979         }
1980
1981         if (lsm != NULL) {
1982                 LASSERT(lsm->lsm_object_gr > 0);
1983                 if (lli->lli_smd == NULL) {
1984                         lli->lli_smd = lsm;
1985                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1986                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1987                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1988                 } else {
1989                         int i;
1990                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1991                                 CERROR("lsm mismatch for inode %ld\n",
1992                                        inode->i_ino);
1993                                 CERROR("lli_smd:\n");
1994                                 dump_lsm(D_ERROR, lli->lli_smd);
1995                                 CERROR("lsm:\n");
1996                                 dump_lsm(D_ERROR, lsm);
1997                                 LBUG();
1998                         }
1999                         /* XXX FIXME -- We should decide on a safer (atomic) and
2000                          * more elegant way to update the lsm */
2001                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
2002                                 lli->lli_smd->lsm_oinfo[i].loi_id =
2003                                         lsm->lsm_oinfo[i].loi_id;
2004                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
2005                                         lsm->lsm_oinfo[i].loi_gr;
2006                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
2007                                         lsm->lsm_oinfo[i].loi_ost_idx;
2008                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
2009                                         lsm->lsm_oinfo[i].loi_ost_gen;
2010                         }
2011                 }
2012                 /* bug 2844 - limit i_blksize for broken user-space apps */
2013                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
2014                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
2015                 if (lli->lli_smd != lsm)
2016                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
2017         }
2018
2019         if (mea != NULL) {
2020                 if (lli->lli_mea == NULL) {
2021                         lli->lli_mea = mea;
2022                 } else {
2023                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
2024                                 CERROR("mea mismatch for inode %lu\n",
2025                                         inode->i_ino);
2026                                 LBUG();
2027                         }
2028                 }
2029                 if (lli->lli_mea != mea)
2030                         obd_free_memmd(ll_i2mdexp(inode),
2031                                        (struct lov_stripe_md **) &mea);
2032         }
2033
2034         if (body->valid & OBD_MD_FID)
2035                 id_assign_fid(&lli->lli_id, &body->id1);
2036
2037         if (body->valid & OBD_MD_FLID)
2038                 id_ino(&lli->lli_id) = id_ino(&body->id1);
2039
2040         if (body->valid & OBD_MD_FLGENER)
2041                 id_gen(&lli->lli_id) = id_gen(&body->id1);
2042
2043         /* local/remote ACL */
2044         if (sbi->ll_remote) {
2045                 LASSERT(md->posix_acl == NULL);
2046                 if (md->remote_perm) {
2047                         ll_remote_acl_update(inode, md->remote_perm);
2048                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
2049                         md->remote_perm = NULL;
2050                 }
2051         } else {
2052                 LASSERT(md->remote_perm == NULL);
2053                 spin_lock(&lli->lli_lock);
2054                 if (posix_acl != NULL) {
2055                         if (lli->lli_posix_acl != NULL)
2056                                 posix_acl_release(lli->lli_posix_acl);
2057                         lli->lli_posix_acl = posix_acl;
2058                 }
2059                 spin_unlock(&lli->lli_lock);
2060         }
2061
2062         if (body->valid & OBD_MD_FLID)
2063                 inode->i_ino = id_ino(&body->id1);
2064         if (body->valid & OBD_MD_FLGENER)
2065                 inode->i_generation = id_gen(&body->id1);
2066         if (body->valid & OBD_MD_FLATIME)
2067                 LTIME_S(inode->i_atime) = body->atime;
2068         if (body->valid & OBD_MD_FLMTIME &&
2069             body->mtime > LTIME_S(inode->i_mtime)) {
2070                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
2071                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
2072                 LTIME_S(inode->i_mtime) = body->mtime;
2073         }
2074         if (body->valid & OBD_MD_FLCTIME &&
2075             body->ctime > LTIME_S(inode->i_ctime))
2076                 LTIME_S(inode->i_ctime) = body->ctime;
2077         if (body->valid & OBD_MD_FLMODE) {
2078                 inode->i_mode = (inode->i_mode & S_IFMT) |
2079                         (body->mode & ~S_IFMT);
2080         }
2081         if (body->valid & OBD_MD_FLTYPE) {
2082                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2083                         (body->mode & S_IFMT);
2084         }
2085         if (body->valid & OBD_MD_FLUID)
2086                 inode->i_uid = body->uid;
2087         if (body->valid & OBD_MD_FLGID)
2088                 inode->i_gid = body->gid;
2089         if (body->valid & OBD_MD_FLFLAGS)
2090                 inode->i_flags = body->flags;
2091         if (body->valid & OBD_MD_FLNLINK)
2092                 inode->i_nlink = body->nlink;
2093         if (body->valid & OBD_MD_FLRDEV)
2094 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2095                 inode->i_rdev = body->rdev;
2096 #else
2097                 inode->i_rdev = old_decode_dev(body->rdev);
2098 #endif
2099         if (body->valid & OBD_MD_FLSIZE)
2100                 inode->i_size = body->size;
2101         if (body->valid & OBD_MD_FLBLOCKS)
2102                 inode->i_blocks = body->blocks;
2103
2104         if (body->valid & OBD_MD_FLSIZE)
2105                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
2106
2107         if (body->valid & OBD_MD_FLAUDIT) {
2108                 struct ll_sb_info * sbi = ll_s2sbi(inode->i_sb);
2109                 if (IS_AUDIT_OP(body->audit, AUDIT_FS))
2110                         sbi->ll_audit_mask = body->audit;
2111                 else
2112                         lli->lli_audit_mask = body->audit;
2113         }
2114
2115         if (mkey != NULL) {
2116                 LASSERT(body->valid & OBD_MD_FLKEY);
2117                 ll_crypto_init_inode_key(inode, mkey);
2118         }
2119
2120 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
2121         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
2122 #endif
2123         LASSERT(id_fid(&lli->lli_id) != 0);
2124 }
2125
2126 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2127 static struct backing_dev_info ll_backing_dev_info = {
2128         .ra_pages       = 0,    /* No readahead */
2129         .memory_backed  = 0,    /* Does contribute to dirty memory */
2130 };
2131 #endif
2132
2133 void ll_read_inode2(struct inode *inode, void *opaque)
2134 {
2135         struct lustre_md *md = opaque;
2136         struct ll_inode_info *lli = ll_i2info(inode);
2137         ENTRY;
2138
2139         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
2140                inode->i_generation, inode);
2141
2142         ll_lli_init(lli);
2143
2144         LASSERT(!lli->lli_smd);
2145
2146         if (ll_i2sbi(inode)->ll_remote) {
2147                 lli->lli_remote_acl = remote_acl_alloc();
2148                 /* if failed alloc, nobody will be able to access this inode */
2149         }
2150
2151         /* Core attributes from the MDS first.  This is a new inode, and
2152          * the VFS doesn't zero times in the core inode so we have to do
2153          * it ourselves.  They will be overwritten by either MDS or OST
2154          * attributes - we just need to make sure they aren't newer. */
2155         LTIME_S(inode->i_mtime) = 0;
2156         LTIME_S(inode->i_atime) = 0;
2157         LTIME_S(inode->i_ctime) = 0;
2158
2159         inode->i_rdev = 0;
2160         ll_update_inode(inode, md);
2161
2162         /* OIDEBUG(inode); */
2163
2164         if (S_ISREG(inode->i_mode)) {
2165                 inode->i_op = &ll_file_inode_operations;
2166                 inode->i_fop = &ll_file_operations;
2167                 inode->i_mapping->a_ops = &ll_aops;
2168                 EXIT;
2169         } else if (S_ISDIR(inode->i_mode)) {
2170                 inode->i_op = &ll_dir_inode_operations;
2171                 inode->i_fop = &ll_dir_operations;
2172                 inode->i_mapping->a_ops = &ll_dir_aops;
2173                 EXIT;
2174         } else if (S_ISLNK(inode->i_mode)) {
2175                 inode->i_op = &ll_fast_symlink_inode_operations;
2176                 EXIT;
2177         } else {
2178                 inode->i_op = &ll_special_inode_operations;
2179
2180 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
2181                 init_special_inode(inode, inode->i_mode,
2182                                    kdev_t_to_nr(inode->i_rdev));
2183
2184                 /* initializing backing dev info. */
2185                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2186 #else
2187                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
2188 #endif
2189                 lli->ll_save_ifop = inode->i_fop;
2190
2191                 if (S_ISCHR(inode->i_mode))
2192                         inode->i_fop = &ll_special_chr_inode_fops;
2193                 else if (S_ISBLK(inode->i_mode))
2194                         inode->i_fop = &ll_special_blk_inode_fops;
2195                 else if (S_ISFIFO(inode->i_mode))
2196                         inode->i_fop = &ll_special_fifo_inode_fops;
2197                 else if (S_ISSOCK(inode->i_mode))
2198                         inode->i_fop = &ll_special_sock_inode_fops;
2199
2200                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2201                       inode->i_fop);
2202
2203                 if (lli->ll_save_ifop->owner) {
2204                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2205                               lli->ll_save_ifop->owner);
2206                 }
2207                 EXIT;
2208         }
2209 }
2210
2211 void ll_delete_inode(struct inode *inode)
2212 {
2213         struct ll_sb_info *sbi = ll_i2sbi(inode);
2214         struct lustre_id id;
2215         int rc;
2216         ENTRY;
2217
2218         ll_inode2id(&id, inode);
2219
2220         rc = md_delete_inode(sbi->ll_md_exp, &id);
2221         if (rc) {
2222                 CERROR("md_delete_inode() failed, error %d\n",
2223                        rc);
2224         }
2225
2226         clear_inode(inode);
2227         EXIT;
2228 }
2229
2230 int ll_iocontrol(struct inode *inode, struct file *file,
2231                  unsigned int cmd, unsigned long arg)
2232 {
2233         struct ll_sb_info *sbi = ll_i2sbi(inode);
2234         struct ptlrpc_request *req = NULL;
2235         int rc, flags = 0;
2236         ENTRY;
2237
2238         switch(cmd) {
2239         case EXT3_IOC_GETFLAGS: {
2240                 struct lustre_id id;
2241                 __u64 valid = OBD_MD_FLFLAGS;
2242                 struct mds_body *body;
2243
2244                 ll_inode2id(&id, inode);
2245                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL,
2246                                 0, 0, NULL, &req);
2247                 if (rc) {
2248                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2249                         RETURN(-abs(rc));
2250                 }
2251
2252                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2253
2254                 if (body->flags & S_APPEND)
2255                         flags |= EXT3_APPEND_FL;
2256                 if (body->flags & S_IMMUTABLE)
2257                         flags |= EXT3_IMMUTABLE_FL;
2258                 if (body->flags & S_NOATIME)
2259                         flags |= EXT3_NOATIME_FL;
2260
2261                 ptlrpc_req_finished (req);
2262
2263                 RETURN(put_user(flags, (int *)arg));
2264         }
2265         case EXT3_IOC_SETFLAGS: {
2266                 struct mdc_op_data *op_data;
2267                 struct iattr attr;
2268                 struct obdo *oa;
2269                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2270
2271                 if (get_user(flags, (int *)arg))
2272                         RETURN(-EFAULT);
2273
2274                 oa = obdo_alloc();
2275                 if (!oa)
2276                         RETURN(-ENOMEM);
2277
2278                 OBD_ALLOC(op_data, sizeof(*op_data));
2279                 if (op_data == NULL) {
2280                         obdo_free(oa);
2281                         RETURN(-ENOMEM);
2282                 }
2283                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
2284
2285                 memset(&attr, 0x0, sizeof(attr));
2286                 attr.ia_attr_flags = flags;
2287                 attr.ia_valid |= ATTR_ATTR_FLAG;
2288
2289                 rc = md_setattr(sbi->ll_md_exp, op_data,
2290                                 &attr, NULL, 0, NULL, 0, NULL, 0, &req);
2291                 OBD_FREE(op_data, sizeof(*op_data));
2292                 if (rc) {
2293                         ptlrpc_req_finished(req);
2294                         if (rc != -EPERM && rc != -EACCES)
2295                                 CERROR("md_setattr fails: rc = %d\n", rc);
2296                         obdo_free(oa);
2297                         RETURN(rc);
2298                 }
2299                 ptlrpc_req_finished(req);
2300
2301                 oa->o_id = lsm->lsm_object_id;
2302                 oa->o_gr = lsm->lsm_object_gr;
2303                 oa->o_flags = flags;
2304                 *(obdo_id(oa)) = ll_i2info(inode)->lli_id;
2305                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP
2306                               | OBD_MD_FLIFID;
2307
2308                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
2309                 obdo_free(oa);
2310                 if (rc) {
2311                         if (rc != -EPERM && rc != -EACCES)
2312                                 CERROR("md_setattr fails: rc = %d\n", rc);
2313                         RETURN(rc);
2314                 }
2315
2316                 if (flags & EXT3_APPEND_FL)
2317                         inode->i_flags |= S_APPEND;
2318                 else
2319                         inode->i_flags &= ~S_APPEND;
2320                 if (flags & EXT3_IMMUTABLE_FL)
2321                         inode->i_flags |= S_IMMUTABLE;
2322                 else
2323                         inode->i_flags &= ~S_IMMUTABLE;
2324                 if (flags & EXT3_NOATIME_FL)
2325                         inode->i_flags |= S_NOATIME;
2326                 else
2327                         inode->i_flags &= ~S_NOATIME;
2328
2329                 RETURN(0);
2330         }
2331         default:
2332                 RETURN(-ENOSYS);
2333         }
2334
2335         RETURN(0);
2336 }
2337
2338 /* this is only called in the case of forced umount. */
2339 void ll_umount_begin(struct super_block *sb)
2340 {
2341         struct ll_sb_info *sbi = ll_s2sbi(sb);
2342         struct obd_ioctl_data ioc_data = { 0 };
2343         struct obd_device *obd;
2344         ENTRY;
2345
2346         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2347                sb->s_count, atomic_read(&sb->s_active));
2348
2349         obd = class_exp2obd(sbi->ll_md_exp);
2350         if (obd == NULL) {
2351                 CERROR("Invalid MDC connection handle "LPX64"\n",
2352                        sbi->ll_md_exp->exp_handle.h_cookie);
2353                 EXIT;
2354                 return;
2355         }
2356         obd->obd_no_recov = 1;
2357         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2358                       sizeof(ioc_data), &ioc_data, NULL);
2359
2360         obd = class_exp2obd(sbi->ll_dt_exp);
2361         if (obd == NULL) {
2362                 CERROR("Invalid LOV connection handle "LPX64"\n",
2363                        sbi->ll_dt_exp->exp_handle.h_cookie);
2364                 EXIT;
2365                 return;
2366         }
2367
2368         obd->obd_no_recov = 1;
2369         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2370                       sizeof(ioc_data), &ioc_data, NULL);
2371
2372         /*
2373          * really, we'd like to wait until there are no requests outstanding,
2374          * and then continue.  For now, we just invalidate the requests,
2375          * schedule, and hope.
2376          */
2377         schedule();
2378
2379         EXIT;
2380 }
2381
2382 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2383                   struct inode **inode, struct ptlrpc_request *req,
2384                   int offset, struct super_block *sb)
2385 {
2386         struct lustre_md md;
2387         int rc = 0;
2388
2389         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2390         if (rc)
2391                 RETURN(rc);
2392
2393         if (*inode) {
2394                 ll_update_inode(*inode, &md);
2395         } else {
2396                 LASSERT(sb);
2397                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2398                 if (*inode == NULL || is_bad_inode(*inode)) {
2399                         /* free the lsm if we allocated one above */
2400                         if (md.lsm != NULL)
2401                                 obd_free_memmd(dt_exp, &md.lsm);
2402                         if (md.mea != NULL)
2403                                 obd_free_memmd(md_exp,
2404                                                (struct lov_stripe_md**)&md.mea);
2405                         rc = -ENOMEM;
2406                         CERROR("new_inode -fatal: rc %d\n", rc);
2407                 }
2408         }
2409
2410         RETURN(rc);
2411 }
2412
2413 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2414 {
2415         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2416         struct lustre_mount_data *lmd = sbi->ll_lmd;
2417
2418         if (lmd) {
2419                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2420                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2421         }
2422         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2423         if (sbi->ll_remote && lmd)
2424                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2425
2426         if (lmd && lmd->lmd_pag)
2427                 seq_printf(m, ",pag");
2428
2429         return 0;
2430 }
2431
2432 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2433                char *filename, struct lustre_id *ret)
2434 {
2435         struct ptlrpc_request *request = NULL;
2436         struct mds_body *body;
2437         int rc;
2438
2439         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2440                              OBD_MD_FID, 0, &request);
2441         if (rc < 0) {
2442                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2443                        filename, rc);
2444                 return rc;
2445         }
2446
2447         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2448         LASSERT(body != NULL);
2449         LASSERT_REPSWABBED(request, 0);
2450
2451         *ret = body->id1;
2452         ptlrpc_req_finished(request);
2453
2454         return rc;
2455 }
2456 int ll_flush_cred(struct inode *inode)
2457 {
2458         struct ll_sb_info *sbi = ll_i2sbi(inode);
2459         int rc = 0;
2460
2461         /* XXX to avoid adding api, we simply use set_info() interface
2462          * to notify underlying obds. set_info() is more like a ioctl() now...
2463          */
2464         if (sbi->ll_md_exp) {
2465                 rc = obd_set_info(sbi->ll_md_exp,
2466                                   strlen("flush_cred"), "flush_cred",
2467                                   0, NULL);
2468                 if (rc)
2469                         return rc;
2470         }
2471
2472         if (sbi->ll_dt_exp) {
2473                 rc = obd_set_info(sbi->ll_dt_exp,
2474                                   strlen("flush_cred"), "flush_cred",
2475                                   0, NULL);
2476                 if (rc)
2477                         return rc;
2478         }
2479
2480         return rc;
2481 }