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
  31 #include <linux/lustre_lite.h>
  32 #include <linux/lustre_ha.h>
  33 #include <linux/lustre_dlm.h>
  34 #include <linux/lprocfs_status.h>
  35 #include <linux/lustre_acl.h>
  36 #include "llite_internal.h"
  37
  38 kmem_cache_t *ll_file_data_slab;
  39 kmem_cache_t *ll_intent_slab;
  40
  41 extern struct address_space_operations ll_aops;
  42 extern struct address_space_operations ll_dir_aops;
  43
  44 #ifndef log2
  45 #define log2(n) ffz(~(n))
  46 #endif
  47
  48 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
  49 {
  50         struct ll_sb_info *sbi = NULL;
  51         class_uuid_t uuid;
  52         ENTRY;
  53
  54         OBD_ALLOC(sbi, sizeof(*sbi));
  55         if (!sbi)
  56                 RETURN(NULL);
  57
  58         spin_lock_init(&sbi->ll_lock);
  59         INIT_LIST_HEAD(&sbi->ll_pglist);
  60         sbi->ll_pglist_gen = 0;
  61         if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
  62                 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
  63         else
  64                 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
  65         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  66         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  67         INIT_LIST_HEAD(&sbi->ll_mnt_list);
  68
  69         sema_init(&sbi->ll_gns_sem, 1);
  70         spin_lock_init(&sbi->ll_gns_lock);
  71         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
  72         init_waitqueue_head(&sbi->ll_gns_waitq);
  73         init_completion(&sbi->ll_gns_mount_finished);
  74
  75         /* this later may be reset via /proc/fs/... */
  76         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
  77         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
  78
  79         /* this later may be reset via /proc/fs/... */
  80         memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
  81                strlen("/usr/sbin/gns_upcall"));
  82         sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
  83
  84         /* default values, may be changed via /proc/fs/... */
  85         sbi->ll_gns_state = LL_GNS_IDLE;
  86         sbi->ll_gns_pending_dentry = NULL;
  87         atomic_set(&sbi->ll_gns_enabled, 1);
  88         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  89         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  90
  91         sbi->ll_gns_timer.data = (unsigned long)sbi;
  92         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  93         init_timer(&sbi->ll_gns_timer);
  94
  95         ll_set_sbi(sb, sbi);
  96
  97         generate_random_uuid(uuid);
  98         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  99         RETURN(sbi);
 100 }
 101
 102 void lustre_free_sbi(struct super_block *sb)
 103 {
 104         struct ll_sb_info *sbi = ll_s2sbi(sb);
 105         ENTRY;
 106
 107         if (sbi != NULL) {
 108                 list_del(&sbi->ll_gns_sbi_head);
 109                 del_timer(&sbi->ll_gns_timer);
 110                 OBD_FREE(sbi, sizeof(*sbi));
 111         }
 112         ll_set_sbi(sb, NULL);
 113         EXIT;
 114 }
 115
 116 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 117 {
 118         __u32 valsize;
 119         int rc = 0;
 120         ENTRY;
 121
 122         valsize = sizeof(sbi->ll_dt_desc);
 123         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 124         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 125                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 126         RETURN(rc);
 127 }
 128
 129 extern struct dentry_operations ll_d_ops;
 130
 131 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 132                              int async,  char *mds_security,  char *oss_security,
 133                              __u32 *nllu, __u64 *remote)
 134 {
 135         struct ll_sb_info *sbi = ll_s2sbi(sb);
 136         struct ptlrpc_request *request = NULL;
 137         struct lustre_handle dt_conn = {0, };
 138         struct lustre_handle md_conn = {0, };
 139         struct obd_connect_data *data;
 140         struct inode *root = NULL;
 141         struct obd_device *obd;
 142         struct obd_statfs osfs;
 143         struct lustre_md md;
 144         kdev_t devno;
 145         int err;
 146         __u32 valsize;
 147         ENTRY;
 148
 149         obd = class_name2obd(lmv);
 150         if (!obd) {
 151                 CERROR("MDC %s: not setup or attached\n", lmv);
 152                 RETURN(-EINVAL);
 153         }
 154         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 155                      sizeof(async), &async);
 156
 157         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 158             (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 159                 CERROR("wrong remote flag "LPX64"\n", *remote);
 160                 RETURN(-EINVAL);
 161         }
 162
 163         OBD_ALLOC(data, sizeof(*data));
 164         if (!data)
 165                 RETURN(-ENOMEM);
 166
 167         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 168                                               OBD_CONNECT_REMOTE);
 169         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 170
 171         if (mds_security == NULL)
 172                 mds_security = "null";
 173
 174         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 175                            strlen(mds_security), mds_security);
 176         if (err) {
 177                 CERROR("LMV %s: failed to set security %s, err %d\n",
 178                         lmv, mds_security, err);
 179                 OBD_FREE(data, sizeof(*data));
 180                 RETURN(err);
 181         }
 182
 183         if (proc_lustre_fs_root) {
 184                 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
 185                                                   sb, lov, lmv);
 186                 if (err < 0)
 187                         CERROR("could not register mount in /proc/lustre");
 188         }
 189
 190         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
 191                           OBD_OPT_REAL_CLIENT);
 192         if (err == -EBUSY) {
 193                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 194                        " client is not a part.  Please wait for recovery to "
 195                        "complete, abort, or time out.\n", lmv);
 196                 GOTO(out, err);
 197         } else if (err) {
 198                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 199                 GOTO(out, err);
 200         }
 201         sbi->ll_md_exp = class_conn2export(&md_conn);
 202         err = obd_statfs(obd, &osfs, jiffies - HZ);
 203         if (err)
 204                 GOTO(out_lmv, err);
 205
 206         if (!osfs.os_bsize) {
 207                 CERROR("Invalid block size is detected.");
 208                 GOTO(out_lmv, err);
 209         }
 210
 211         sb->s_magic = LL_SUPER_MAGIC;
 212         sb->s_blocksize = osfs.os_bsize;
 213         sb->s_blocksize_bits = log2(osfs.os_bsize);
 214         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 215
 216         devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 217                              strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 218
 219         sb->s_dev = devno;
 220
 221         /* after statfs, we are supposed to have connected to MDSs,
 222          * so it's ok to check remote flag returned.
 223          */
 224         valsize = sizeof(&sbi->ll_remote);
 225         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 226                            &valsize, &sbi->ll_remote);
 227         if (err) {
 228                 CERROR("fail to obtain remote flag\n");
 229                 GOTO(out, err);
 230         }
 231
 232         obd = class_name2obd(lov);
 233         if (!obd) {
 234                 CERROR("OSC %s: not setup or attached\n", lov);
 235                 GOTO(out_lmv, err);
 236         }
 237         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 238                      sizeof(async), &async);
 239
 240        if (oss_security == NULL)
 241                 oss_security = "null";
 242
 243         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 244                            strlen(oss_security), oss_security);
 245         if (err) {
 246                 CERROR("LOV %s: failed to set security %s, err %d\n",
 247                         lov, oss_security, err);
 248                 OBD_FREE(data, sizeof(*data));
 249                 RETURN(err);
 250         }
 251
 252         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
 253         if (err == -EBUSY) {
 254                 CERROR("An OST (lov %s) is performing recovery, of which this"
 255                        " client is not a part.  Please wait for recovery to "
 256                        "complete, abort, or time out.\n", lov);
 257                 GOTO(out, err);
 258         } else if (err) {
 259                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 260                 GOTO(out_lmv, err);
 261         }
 262         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 263
 264         err = lustre_init_dt_desc(sbi);
 265         if (err == 0) {
 266                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 267                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
 268                                  sbi->ll_dt_desc.ld_tgt_count *
 269                                  sizeof(struct llog_cookie));
 270         }
 271
 272         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 273         if (err) {
 274                 CERROR("cannot mds_connect: rc = %d\n", err);
 275                 GOTO(out_lov, err);
 276         }
 277         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 278
 279         sb->s_op = &lustre_super_operations;
 280
 281         /* make root inode */
 282         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 283                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
 284                          NULL, 0, 0, &request);
 285         if (err) {
 286                 CERROR("md_getattr failed for root: rc = %d\n", err);
 287                 GOTO(out_lov, err);
 288         }
 289
 290         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 291                                 sbi->ll_dt_exp, &md);
 292         if (err) {
 293                 CERROR("failed to understand root inode md: rc = %d\n", err);
 294                 ptlrpc_req_finished(request);
 295                 GOTO(out_lov, err);
 296         }
 297
 298         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 299         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 300
 301         ptlrpc_req_finished(request);
 302
 303         if (root == NULL || is_bad_inode(root)) {
 304                 if (md.lsm != NULL)
 305                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 306                 if (md.mea != NULL)
 307                     obd_free_memmd(sbi->ll_md_exp,
 308                                    (struct lov_stripe_md**)&md.mea);
 309                 CERROR("lustre_lite: bad iget4 for root\n");
 310                 GOTO(out_root, err = -EBADF);
 311         }
 312
 313         err = ll_close_thread_start(&sbi->ll_lcq);
 314         if (err) {
 315                 CERROR("cannot start close thread: rc %d\n", err);
 316                 GOTO(out_root, err);
 317         }
 318
 319         ll_gns_add_timer(sbi);
 320
 321         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 322            backing dev info assigned to inode mapping is used for
 323            determining maximal readahead. */
 324 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 325     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 326         /* bug 2805 - set VM readahead to zero */
 327         vm_max_readahead = vm_min_readahead = 0;
 328 #endif
 329
 330         sb->s_root = d_alloc_root(root);
 331         sb->s_root->d_op = &ll_d_ops;
 332
 333         sb->s_flags |= MS_POSIXACL;
 334 #ifdef S_PDIROPS
 335         CWARN("Enabling PDIROPS\n");
 336         sb->s_flags |= S_PDIROPS;
 337 #endif
 338
 339         if (data != NULL)
 340                 OBD_FREE(data, sizeof(*data));
 341         RETURN(err);
 342 out_root:
 343         if (root)
 344                 iput(root);
 345 out_lov:
 346         obd_disconnect(sbi->ll_dt_exp, 0);
 347 out_lmv:
 348         obd_disconnect(sbi->ll_md_exp, 0);
 349 out:
 350         if (data != NULL)
 351                 OBD_FREE(data, sizeof(*data));
 352         lprocfs_unregister_mountpoint(sbi);
 353         return err;
 354 }
 355
 356 void lustre_common_put_super(struct super_block *sb)
 357 {
 358         struct ll_sb_info *sbi = ll_s2sbi(sb);
 359         struct hlist_node *tmp, *next;
 360         ENTRY;
 361
 362         ll_gns_del_timer(sbi);
 363         ll_close_thread_shutdown(sbi->ll_lcq);
 364
 365         list_del(&sbi->ll_conn_chain);
 366         obd_disconnect(sbi->ll_dt_exp, 0);
 367
 368         lprocfs_unregister_mountpoint(sbi);
 369         if (sbi->ll_proc_root) {
 370                 lprocfs_remove(sbi->ll_proc_root);
 371                 sbi->ll_proc_root = NULL;
 372         }
 373
 374         obd_disconnect(sbi->ll_md_exp, 0);
 375
 376         // We do this to get rid of orphaned dentries. That is not really trw.
 377         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 378                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 379                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 380                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 381                 shrink_dcache_parent(dentry);
 382         }
 383         EXIT;
 384 }
 385
 386 char *ll_read_opt(const char *opt, char *data)
 387 {
 388         char *value;
 389         char *retval;
 390         ENTRY;
 391
 392         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 393         if (strncmp(opt, data, strlen(opt)))
 394                 RETURN(NULL);
 395         if ((value = strchr(data, '=')) == NULL)
 396                 RETURN(NULL);
 397
 398         value++;
 399         OBD_ALLOC(retval, strlen(value) + 1);
 400         if (!retval) {
 401                 CERROR("out of memory!\n");
 402                 RETURN(NULL);
 403         }
 404
 405         memcpy(retval, value, strlen(value)+1);
 406         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 407         RETURN(retval);
 408 }
 409
 410 int ll_set_opt(const char *opt, char *data, int fl)
 411 {
 412         ENTRY;
 413
 414         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 415         if (strncmp(opt, data, strlen(opt)))
 416                 RETURN(0);
 417         else
 418                 RETURN(fl);
 419 }
 420
 421 void ll_options(char *options, char **lov, char **lmv, char **mds_sec,
 422                 char **oss_sec, int *async, int *flags)
 423 {
 424         char *this_char;
 425 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 426         char *opt_ptr = options;
 427 #endif
 428         ENTRY;
 429
 430         if (!options) {
 431                 EXIT;
 432                 return;
 433         }
 434
 435         *async = 0;
 436 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 437         for (this_char = strtok (options, ",");
 438              this_char != NULL;
 439              this_char = strtok (NULL, ",")) {
 440 #else
 441         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 442 #endif
 443                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 444                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 445                         continue;
 446                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 447                         continue;
 448                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 449                         *async = 1;
 450                         continue;
 451                 }
 452                 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
 453                         continue;
 454                 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
 455                         continue;
 456                 if (!(*flags & LL_SBI_NOLCK) &&
 457                     ((*flags) = (*flags) |
 458                                 ll_set_opt("nolock", this_char,
 459                                            LL_SBI_NOLCK)))
 460                         continue;
 461         }
 462
 463         EXIT;
 464 }
 465
 466 void ll_lli_init(struct ll_inode_info *lli)
 467 {
 468         sema_init(&lli->lli_open_sem, 1);
 469         sema_init(&lli->lli_size_sem, 1);
 470         lli->lli_flags = 0;
 471         lli->lli_size_pid = 0;
 472         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 473         spin_lock_init(&lli->lli_lock);
 474         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 475         lli->lli_inode_magic = LLI_INODE_MAGIC;
 476         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 477         sema_init(&lli->lli_och_sem, 1);
 478         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 479         lli->lli_mds_exec_och = NULL;
 480         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 481         lli->lli_open_fd_exec_count = 0;
 482 }
 483
 484 int ll_fill_super(struct super_block *sb, void *data, int silent)
 485 {
 486         struct ll_sb_info *sbi;
 487         char *lov = NULL;
 488         char *lmv = NULL;
 489         char *mds_sec = NULL;
 490         char *oss_sec = NULL;
 491         int async, err;
 492         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 493         __u64 remote_flag = 0;
 494         ENTRY;
 495
 496         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 497
 498         sbi = lustre_init_sbi(sb);
 499         if (!sbi)
 500                 RETURN(-ENOMEM);
 501
 502         sbi->ll_flags |= LL_SBI_READAHEAD;
 503         ll_options(data, &lov, &lmv, &mds_sec, &oss_sec,
 504                    &async, &sbi->ll_flags);
 505
 506         if (!lov) {
 507                 CERROR("no osc\n");
 508                 GOTO(out, err = -EINVAL);
 509         }
 510
 511         if (!lmv) {
 512                 CERROR("no mdc\n");
 513                 GOTO(out, err = -EINVAL);
 514         }
 515
 516         err = lustre_common_fill_super(sb, lmv, lov, async, mds_sec, oss_sec,
 517                                        nllu, &remote_flag);
 518         EXIT;
 519 out:
 520         if (err)
 521                 lustre_free_sbi(sb);
 522
 523         if (lmv)
 524                 OBD_FREE(lmv, strlen(lmv) + 1);
 525         if (lov)
 526                 OBD_FREE(lov, strlen(lov) + 1);
 527         if (mds_sec)
 528                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 529         if (oss_sec)
 530                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 531
 532         return err;
 533 } /* ll_read_super */
 534
 535 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 536                               struct config_llog_instance *cfg, int allow_recov)
 537 {
 538         struct lustre_cfg *lcfg = NULL;
 539         struct lustre_cfg_bufs bufs;
 540         struct portals_cfg pcfg;
 541         char *peer = "MDS_PEER_UUID";
 542         struct obd_device *obd;
 543         struct lustre_handle md_conn = {0, };
 544         struct obd_export *exp;
 545         char *name = "mdc_dev";
 546         class_uuid_t uuid;
 547         struct obd_uuid lmv_uuid;
 548         struct llog_ctxt *ctxt;
 549         int rc = 0, err = 0;
 550         ENTRY;
 551
 552         if (lmd_bad_magic(lmd))
 553                 RETURN(-EINVAL);
 554
 555         generate_random_uuid(uuid);
 556         class_uuid_unparse(uuid, &lmv_uuid);
 557
 558         if (lmd->lmd_local_nid) {
 559                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 560                 pcfg.pcfg_nal = lmd->lmd_nal;
 561                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 562                 rc = libcfs_nal_cmd(&pcfg);
 563                 if (rc < 0)
 564                         GOTO(out, rc);
 565         }
 566
 567         if (lmd->lmd_nal == SOCKNAL ||
 568             lmd->lmd_nal == OPENIBNAL ||
 569             lmd->lmd_nal == IIBNAL ||
 570             lmd->lmd_nal == VIBNAL ||
 571             lmd->lmd_nal == RANAL) {
 572                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 573                 pcfg.pcfg_nal     = lmd->lmd_nal;
 574                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 575                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 576                 pcfg.pcfg_misc    = lmd->lmd_port;
 577                 rc = libcfs_nal_cmd(&pcfg);
 578                 if (rc < 0)
 579                         GOTO(out, rc);
 580         }
 581         lustre_cfg_bufs_reset(&bufs, name);
 582         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 583
 584         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 585         lcfg->lcfg_nal = lmd->lmd_nal;
 586         lcfg->lcfg_nid = lmd->lmd_server_nid;
 587         LASSERT(lcfg->lcfg_nal);
 588         LASSERT(lcfg->lcfg_nid);
 589         err = class_process_config(lcfg);
 590         lustre_cfg_free(lcfg);
 591         if (err < 0)
 592                 GOTO(out_del_conn, err);
 593
 594         lustre_cfg_bufs_reset(&bufs, name);
 595         lustre_cfg_bufs_set_string(&bufs, 1, LUSTRE_MDC_NAME);
 596         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 597
 598         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 599         err = class_process_config(lcfg);
 600         lustre_cfg_free(lcfg);
 601         if (err < 0)
 602                 GOTO(out_del_uuid, err);
 603
 604         lustre_cfg_bufs_reset(&bufs, name);
 605         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 606         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 607
 608         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 609         err = class_process_config(lcfg);
 610         lustre_cfg_free(lcfg);
 611         if (err < 0)
 612                 GOTO(out_detach, err);
 613
 614         obd = class_name2obd(name);
 615         if (obd == NULL)
 616                 GOTO(out_cleanup, rc = -EINVAL);
 617
 618         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 619                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 620         if (rc)
 621                 GOTO(out_cleanup, rc);
 622
 623         /* Disable initial recovery on this import */
 624         rc = obd_set_info(obd->obd_self_export,
 625                           strlen("initial_recov"), "initial_recov",
 626                           sizeof(allow_recov), &allow_recov);
 627         if (rc)
 628                 GOTO(out_cleanup, rc);
 629
 630         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 631         if (rc) {
 632                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 633                 GOTO(out_cleanup, rc);
 634         }
 635
 636         exp = class_conn2export(&md_conn);
 637
 638         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 639         rc = class_config_process_llog(ctxt, profile, cfg);
 640         if (rc)
 641                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 642
 643         err = obd_disconnect(exp, 0);
 644
 645         EXIT;
 646 out_cleanup:
 647         lustre_cfg_bufs_reset(&bufs, name);
 648         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 649         err = class_process_config(lcfg);
 650         lustre_cfg_free(lcfg);
 651         if (err < 0)
 652                 GOTO(out, err);
 653 out_detach:
 654         lustre_cfg_bufs_reset(&bufs, name);
 655         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 656         err = class_process_config(lcfg);
 657         lustre_cfg_free(lcfg);
 658         if (err < 0)
 659                 GOTO(out, err);
 660
 661 out_del_uuid:
 662         lustre_cfg_bufs_reset(&bufs, name);
 663         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 664         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 665         err = class_process_config(lcfg);
 666         lustre_cfg_free(lcfg);
 667
 668 out_del_conn:
 669         if (lmd->lmd_nal == SOCKNAL ||
 670             lmd->lmd_nal == OPENIBNAL ||
 671             lmd->lmd_nal == IIBNAL ||
 672             lmd->lmd_nal == VIBNAL ||
 673             lmd->lmd_nal == RANAL) {
 674                 int err2;
 675
 676                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 677                 pcfg.pcfg_nal     = lmd->lmd_nal;
 678                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 679                 pcfg.pcfg_flags   = 1;          /* single_share */
 680                 err2 = libcfs_nal_cmd(&pcfg);
 681                 if (err2 && !err)
 682                         err = err2;
 683                 if (err < 0)
 684                         GOTO(out, err);
 685         }
 686 out:
 687         if (rc == 0)
 688                 rc = err;
 689
 690         return rc;
 691 }
 692
 693 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 694 {
 695         struct lustre_cfg *lcfg;
 696         struct lustre_cfg_bufs bufs;
 697         struct obd_device *obd;
 698         int next = 0;
 699
 700         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 701         {
 702                 int err;
 703
 704                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 705                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 706                 err = class_process_config(lcfg);
 707                 if (err) {
 708                         CERROR("cleanup failed: %s\n", obd->obd_name);
 709                         //continue;
 710                 }
 711
 712                 lcfg->lcfg_command = LCFG_DETACH;
 713                 err = class_process_config(lcfg);
 714                 lustre_cfg_free(lcfg);
 715                 if (err) {
 716                         CERROR("detach failed: %s\n", obd->obd_name);
 717                         //continue;
 718                 }
 719         }
 720
 721         if (sbi->ll_lmd != NULL)
 722                 class_del_profile(sbi->ll_lmd->lmd_profile);
 723 }
 724
 725 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 726 {
 727         struct lustre_mount_data * lmd = data;
 728         char *lov = NULL, *lmv = NULL;
 729         struct ll_sb_info *sbi;
 730         int err;
 731         ENTRY;
 732
 733         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 734         if (lmd_bad_magic(lmd))
 735                 RETURN(-EINVAL);
 736
 737         sbi = lustre_init_sbi(sb);
 738         if (!sbi)
 739                 RETURN(-ENOMEM);
 740
 741         sbi->ll_flags |= LL_SBI_READAHEAD;
 742
 743         if (lmd->lmd_profile) {
 744                 struct lustre_profile *lprof;
 745                 struct config_llog_instance cfg;
 746                 int len;
 747
 748                 if (lmd->lmd_mds[0] == '\0') {
 749                         CERROR("no mds name\n");
 750                         GOTO(out_free, err = -EINVAL);
 751                 }
 752                 lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
 753                 lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
 754
 755                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 756                 if (sbi->ll_lmd == NULL)
 757                         GOTO(out_free, err = -ENOMEM);
 758                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 759
 760                 /* generate a string unique to this super, let's try
 761                  the address of the super itself.*/
 762                 len = (sizeof(sb) * 2) + 1;
 763                 OBD_ALLOC(sbi->ll_instance, len);
 764                 if (sbi->ll_instance == NULL)
 765                         GOTO(out_free, err = -ENOMEM);
 766                 sprintf(sbi->ll_instance, "%p", sb);
 767
 768                 cfg.cfg_instance = sbi->ll_instance;
 769                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 770                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 771                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 772                 if (err < 0) {
 773                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 774                         GOTO(out_free, err);
 775                 }
 776
 777                 lprof = class_get_profile(lmd->lmd_profile);
 778                 if (lprof == NULL) {
 779                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 780                         GOTO(out_free, err = -EINVAL);
 781                 }
 782                 if (lov)
 783                         OBD_FREE(lov, strlen(lov) + 1);
 784                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
 785                           strlen(sbi->ll_instance) + 2);
 786                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 787
 788                 if (lmv)
 789                         OBD_FREE(lmv, strlen(lmv) + 1);
 790                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
 791                           strlen(sbi->ll_instance) + 2);
 792                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 793         }
 794
 795         if (!lov) {
 796                 CERROR("no osc\n");
 797                 GOTO(out_free, err = -EINVAL);
 798         }
 799
 800         if (!lmv) {
 801                 CERROR("no mdc\n");
 802                 GOTO(out_free, err = -EINVAL);
 803         }
 804
 805         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
 806                                        lmd->lmd_mds_security,
 807                                        lmd->lmd_oss_security,
 808                                        &lmd->lmd_nllu, &lmd->lmd_remote_flag);
 809
 810         if (err)
 811                 GOTO(out_free, err);
 812
 813 out_dev:
 814         if (lmv)
 815                 OBD_FREE(lmv, strlen(lmv) + 1);
 816         if (lov)
 817                 OBD_FREE(lov, strlen(lov) + 1);
 818
 819         RETURN(err);
 820
 821 out_free:
 822         if (sbi->ll_lmd) {
 823                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 824                 int err;
 825
 826                 if (sbi->ll_instance != NULL) {
 827                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 828                         struct config_llog_instance cfg;
 829                         char *cl_prof;
 830
 831                         cfg.cfg_instance = sbi->ll_instance;
 832                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 833
 834                         OBD_ALLOC(cl_prof, len);
 835                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 836                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 837                         if (err < 0) {
 838                                 CERROR("Unable to process log: %s\n", cl_prof);
 839                                 lustre_manual_cleanup(sbi);
 840                         }
 841                         OBD_FREE(cl_prof, len);
 842                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 843                 }
 844                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 845         }
 846         lustre_free_sbi(sb);
 847         goto out_dev;
 848 } /* lustre_fill_super */
 849
 850 void lustre_put_super(struct super_block *sb)
 851 {
 852         struct obd_device *obd;
 853         struct ll_sb_info *sbi = ll_s2sbi(sb);
 854         int force_umount = 0;
 855         ENTRY;
 856
 857         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 858         obd = class_exp2obd(sbi->ll_md_exp);
 859         if (obd)
 860                 force_umount = obd->obd_no_recov;
 861         obd = NULL;
 862
 863         lustre_common_put_super(sb);
 864         if (sbi->ll_lmd != NULL) {
 865                 char *cl_prof;
 866                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 867                 int err;
 868                 struct config_llog_instance cfg;
 869
 870                 if (force_umount) {
 871                         CERROR("force umount, doing manual cleanup\n");
 872                         lustre_manual_cleanup(sbi);
 873                         GOTO(free_lmd, 0);
 874                 }
 875
 876                 cfg.cfg_instance = sbi->ll_instance;
 877                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 878
 879                 OBD_ALLOC(cl_prof, len);
 880                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 881                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
 882                 if (err < 0) {
 883                         CERROR("Unable to process log: %s, doing manual cleanup"
 884                                "\n", cl_prof);
 885                         lustre_manual_cleanup(sbi);
 886                 }
 887
 888                 OBD_FREE(cl_prof, len);
 889         free_lmd:
 890                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 891                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 892         }
 893
 894         lustre_free_sbi(sb);
 895
 896         EXIT;
 897 } /* lustre_put_super */
 898
 899 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 900 {
 901         struct lustre_mount_data *lmd = sbi->ll_lmd;
 902         char *profile = lmd->lmd_profile, *name = NULL;
 903         struct config_llog_instance cfg;
 904         int rc, namelen =  0, version;
 905         struct llog_ctxt *ctxt;
 906         ENTRY;
 907
 908         if (profile == NULL)
 909                 RETURN(0);
 910         if (lmd == NULL) {
 911                 CERROR("Client not mounted with zero-conf; cannot "
 912                        "process update log.\n");
 913                 RETURN(0);
 914         }
 915
 916         rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
 917                                LDLM_FL_CONFIG_CHANGE, NULL);
 918         if (rc != 0)
 919                 CWARN("obd_cancel_unused(mdc): %d\n", rc);
 920
 921         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
 922                                LDLM_FL_CONFIG_CHANGE, NULL);
 923         if (rc != 0)
 924                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 925
 926         cfg.cfg_instance = sbi->ll_instance;
 927         cfg.cfg_uuid = sbi->ll_sb_uuid;
 928         cfg.cfg_local_nid = lmd->lmd_local_nid;
 929
 930         namelen = strlen(profile) + 20; /* -clean-######### */
 931         OBD_ALLOC(name, namelen);
 932         if (name == NULL)
 933                 RETURN(-ENOMEM);
 934
 935         if (clean) {
 936                 version = sbi->ll_config_version - 1;
 937                 sprintf(name, "%s-clean-%d", profile, version);
 938         } else {
 939                 version = sbi->ll_config_version + 1;
 940                 sprintf(name, "%s-%d", profile, version);
 941         }
 942
 943         CWARN("Applying configuration log %s\n", name);
 944
 945         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
 946                                 LLOG_CONFIG_REPL_CTXT);
 947         rc = class_config_process_llog(ctxt, name, &cfg);
 948         if (rc == 0)
 949                 sbi->ll_config_version = version;
 950         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 951
 952         if (rc == 0 && clean == 0) {
 953                 struct lov_desc desc;
 954                 __u32 valsize;
 955                 int rc = 0;
 956
 957                 valsize = sizeof(desc);
 958                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 959                                   "lovdesc", &valsize, &desc);
 960
 961                 rc = obd_init_ea_size(sbi->ll_md_exp,
 962                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
 963                                       (desc.ld_tgt_count *
 964                                        sizeof(struct llog_cookie)));
 965         }
 966         OBD_FREE(name, namelen);
 967         RETURN(rc);
 968 }
 969
 970 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 971 {
 972         struct inode *inode = NULL;
 973         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 974         if (lock->l_ast_data) {
 975                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 976                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 977                         inode = igrab(lock->l_ast_data);
 978                 } else {
 979                         inode = lock->l_ast_data;
 980                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
 981                                "l_ast_data %p is bogus: magic %0x8\n",
 982                                lock->l_ast_data, lli->lli_inode_magic);
 983                         inode = NULL;
 984                 }
 985         }
 986         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 987         return inode;
 988 }
 989
 990 int null_if_equal(struct ldlm_lock *lock, void *data)
 991 {
 992         if (data == lock->l_ast_data) {
 993                 lock->l_ast_data = NULL;
 994
 995                 if (lock->l_req_mode != lock->l_granted_mode)
 996                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 997         }
 998
 999         return LDLM_ITER_CONTINUE;
1000 }
1001
1002 void ll_clear_inode(struct inode *inode)
1003 {
1004         struct lustre_id id;
1005         struct ll_inode_info *lli = ll_i2info(inode);
1006         struct ll_sb_info *sbi = ll_i2sbi(inode);
1007         ENTRY;
1008
1009         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1010                inode->i_generation, inode);
1011
1012         lli->lli_inode_magic = LLI_INODE_DEAD;
1013         ll_inode2id(&id, inode);
1014
1015         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1016         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1017
1018         LASSERT(!lli->lli_open_fd_write_count);
1019         LASSERT(!lli->lli_open_fd_read_count);
1020         LASSERT(!lli->lli_open_fd_exec_count);
1021         if (lli->lli_mds_write_och)
1022                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1023         if (lli->lli_mds_exec_och)
1024                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1025         if (lli->lli_mds_read_och)
1026                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1027         if (lli->lli_smd)
1028                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1029                                   null_if_equal, inode);
1030
1031         if (lli->lli_smd) {
1032                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1033                 lli->lli_smd = NULL;
1034         }
1035
1036         if (lli->lli_mea) {
1037                 obd_free_memmd(sbi->ll_md_exp,
1038                                (struct lov_stripe_md **) &lli->lli_mea);
1039                 lli->lli_mea = NULL;
1040         }
1041
1042         if (lli->lli_symlink_name) {
1043                 OBD_FREE(lli->lli_symlink_name,
1044                          strlen(lli->lli_symlink_name) + 1);
1045                 lli->lli_symlink_name = NULL;
1046         }
1047         lli->lli_inode_magic = LLI_INODE_DEAD;
1048
1049         EXIT;
1050 }
1051
1052 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1053  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1054  * keep these values until such a time that objects are allocated for it.
1055  * We do the MDS operations first, as it is checking permissions for us.
1056  * We don't to the MDS RPC if there is nothing that we want to store there,
1057  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1058  * going to do an RPC anyways.
1059  *
1060  * If we are doing a truncate, we will send the mtime and ctime updates
1061  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1062  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1063  * at the same time.
1064  */
1065 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1066 {
1067         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1068         struct ll_sb_info *sbi = ll_i2sbi(inode);
1069         struct ptlrpc_request *request = NULL;
1070         struct mdc_op_data *op_data;
1071         int ia_valid = attr->ia_valid;
1072         int err, rc = 0;
1073         ENTRY;
1074
1075         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1076         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1077
1078         if (ia_valid & ATTR_SIZE) {
1079                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1080                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1081                                attr->ia_size, ll_file_maxbytes(inode));
1082                         RETURN(-EFBIG);
1083                 }
1084
1085                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1086         }
1087
1088         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1089         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1090                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1091                         RETURN(-EPERM);
1092         }
1093
1094         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1095         if (attr->ia_valid & ATTR_CTIME) {
1096                 attr->ia_ctime = CURRENT_TIME;
1097                 attr->ia_valid |= ATTR_CTIME_SET;
1098         }
1099         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1100                 attr->ia_atime = CURRENT_TIME;
1101                 attr->ia_valid |= ATTR_ATIME_SET;
1102         }
1103         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1104                 attr->ia_mtime = CURRENT_TIME;
1105                 attr->ia_valid |= ATTR_MTIME_SET;
1106         }
1107
1108         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1109                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1110                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1111                        LTIME_S(CURRENT_TIME));
1112
1113         if (lsm)
1114                 attr->ia_valid &= ~ATTR_SIZE;
1115
1116         /* If only OST attributes being set on objects, don't do MDS RPC.
1117          * In that case, we need to check permissions and update the local
1118          * inode ourselves so we can call obdo_from_inode() always. */
1119         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1120                 struct lustre_md md;
1121
1122                 OBD_ALLOC(op_data, sizeof(*op_data));
1123                 if (op_data == NULL)
1124                         RETURN(-ENOMEM);
1125                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1126
1127                 rc = md_setattr(sbi->ll_md_exp, op_data,
1128                                 attr, NULL, 0, NULL, 0, &request);
1129                 OBD_FREE(op_data, sizeof(*op_data));
1130                 if (rc) {
1131                         ptlrpc_req_finished(request);
1132                         if (rc != -EPERM && rc != -EACCES)
1133                                 CERROR("md_setattr fails: rc = %d\n", rc);
1134                         RETURN(rc);
1135                 }
1136
1137                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1138                                        sbi->ll_dt_exp, &md);
1139                 if (rc) {
1140                         ptlrpc_req_finished(request);
1141                         RETURN(rc);
1142                 }
1143
1144                 /* We call inode_setattr to adjust timestamps, but we first
1145                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1146                  *
1147                  * NB: ATTR_SIZE will only be set at this point if the size
1148                  * resides on the MDS, ie, this file has no objects. */
1149                 attr->ia_valid &= ~ATTR_SIZE;
1150
1151                 /*
1152                  * assigning inode_setattr() to @err to disable warning that
1153                  * function's result should be checked by by caller. error is
1154                  * impossible here, as vmtruncate() control path is disabled.
1155                  */
1156                 err = inode_setattr(inode, attr);
1157                 ll_update_inode(inode, &md);
1158                 ptlrpc_req_finished(request);
1159
1160                 if (!lsm || !S_ISREG(inode->i_mode)) {
1161                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1162                         RETURN(0);
1163                 }
1164         } else {
1165                 /* The OST doesn't check permissions, but the alternative is
1166                  * a gratuitous RPC to the MDS.  We already rely on the client
1167                  * to do read/write/truncate permission checks, so is mtime OK?
1168                  */
1169                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1170                         /* from sys_utime() */
1171                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1172                                 if (current->fsuid != inode->i_uid &&
1173                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1174                                         RETURN(rc);
1175                         } else {
1176                                 /* from inode_change_ok() */
1177                                 if (current->fsuid != inode->i_uid &&
1178                                     !capable(CAP_FOWNER))
1179                                         RETURN(-EPERM);
1180                         }
1181                 }
1182
1183                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1184                 err = inode_setattr(inode, attr);
1185                 /*
1186                  * assigning inode_setattr() to @err to disable warning that
1187                  * function's result should be checked by by caller. error is
1188                  * impossible here, as vmtruncate() control path is disabled.
1189                  */
1190         }
1191
1192         /* We really need to get our PW lock before we change inode->i_size.
1193          * If we don't we can race with other i_size updaters on our node, like
1194          * ll_file_read.  We can also race with i_size propogation to other
1195          * nodes through dirtying and writeback of final cached pages.  This
1196          * last one is especially bad for racing o_append users on other
1197          * nodes. */
1198         if (ia_valid & ATTR_SIZE) {
1199                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1200                                                            OBD_OBJECT_EOF } };
1201                 struct lustre_handle lockh = { 0 };
1202                 struct ll_inode_info *lli = ll_i2info(inode);
1203                 int err, ast_flags = 0;
1204                 /* XXX when we fix the AST intents to pass the discard-range
1205                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1206                  * XXX here. */
1207                 if (attr->ia_size == 0)
1208                         ast_flags = LDLM_AST_DISCARD_DATA;
1209
1210                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1211                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1212
1213                 if (rc != 0)
1214                         RETURN(rc);
1215
1216                 down(&lli->lli_size_sem);
1217                 lli->lli_size_pid = current->pid;
1218                 rc = vmtruncate(inode, attr->ia_size);
1219                 if (rc != 0) {
1220                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1221                         lli->lli_size_pid = 0;
1222                         up(&lli->lli_size_sem);
1223                 }
1224
1225                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1226                 if (err) {
1227                         CERROR("ll_extent_unlock failed: %d\n", err);
1228                         if (!rc)
1229                                 rc = err;
1230                 }
1231         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1232                 struct obdo *oa = NULL;
1233
1234                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1235                        inode->i_ino, LTIME_S(attr->ia_mtime));
1236
1237                 oa = obdo_alloc();
1238                 if (oa == NULL)
1239                         RETURN(-ENOMEM);
1240
1241                 oa->o_id = lsm->lsm_object_id;
1242                 oa->o_gr = lsm->lsm_object_gr;
1243                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1244                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1245                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1246                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1247                 obdo_free(oa);
1248                 if (rc)
1249                         CERROR("obd_setattr fails: rc = %d\n", rc);
1250         }
1251         RETURN(rc);
1252 }
1253
1254 int ll_setattr(struct dentry *de, struct iattr *attr)
1255 {
1256         LASSERT(de->d_inode);
1257         return ll_setattr_raw(de->d_inode, attr);
1258 }
1259
1260 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1261                        unsigned long max_age)
1262 {
1263         struct ll_sb_info *sbi = ll_s2sbi(sb);
1264         struct obd_statfs obd_osfs;
1265         int rc;
1266         ENTRY;
1267
1268         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1269         if (rc) {
1270                 CERROR("obd_statfs fails: rc = %d\n", rc);
1271                 RETURN(rc);
1272         }
1273
1274         osfs->os_type = sb->s_magic;
1275
1276         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1277                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1278
1279         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1280         if (rc) {
1281                 CERROR("obd_statfs fails: rc = %d\n", rc);
1282                 RETURN(rc);
1283         }
1284
1285         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1286                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1287                obd_osfs.os_files);
1288
1289         osfs->os_blocks = obd_osfs.os_blocks;
1290         osfs->os_bfree = obd_osfs.os_bfree;
1291         osfs->os_bavail = obd_osfs.os_bavail;
1292
1293         /* If we don't have as many objects free on the OST as inodes
1294          * on the MDS, we reduce the total number of inodes to
1295          * compensate, so that the "inodes in use" number is correct.
1296          */
1297         if (obd_osfs.os_ffree < osfs->os_ffree) {
1298                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1299                         obd_osfs.os_ffree;
1300                 osfs->os_ffree = obd_osfs.os_ffree;
1301         }
1302
1303         RETURN(rc);
1304 }
1305
1306 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1307 {
1308         struct obd_statfs osfs;
1309         int rc;
1310
1311         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1312         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1313
1314         /* For now we will always get up-to-date statfs values, but in the
1315          * future we may allow some amount of caching on the client (e.g.
1316          * from QOS or lprocfs updates). */
1317         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1318         if (rc)
1319                 return rc;
1320
1321         statfs_unpack(sfs, &osfs);
1322
1323         if (sizeof(sfs->f_blocks) == 4) {
1324                 while (osfs.os_blocks > ~0UL) {
1325                         sfs->f_bsize <<= 1;
1326
1327                         osfs.os_blocks >>= 1;
1328                         osfs.os_bfree >>= 1;
1329                         osfs.os_bavail >>= 1;
1330                 }
1331         }
1332
1333         sfs->f_blocks = osfs.os_blocks;
1334         sfs->f_bfree = osfs.os_bfree;
1335         sfs->f_bavail = osfs.os_bavail;
1336
1337         return 0;
1338 }
1339
1340 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1341 {
1342         struct ll_inode_info *lli = ll_i2info(inode);
1343         struct lov_stripe_md *lsm = md->lsm;
1344         struct mds_body *body = md->body;
1345         struct mea *mea = md->mea;
1346         struct posix_acl *ll_acl_access = md->acl_access;
1347         ENTRY;
1348
1349         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1350
1351         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1352                 /* check for default striping info for dir. */
1353                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1354         }
1355
1356         if (lsm != NULL) {
1357                 LASSERT(lsm->lsm_object_gr > 0);
1358                 if (lli->lli_smd == NULL) {
1359                         lli->lli_smd = lsm;
1360                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1361                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1362                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1363                 } else {
1364                         int i;
1365                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1366                                 CERROR("lsm mismatch for inode %ld\n",
1367                                        inode->i_ino);
1368                                 CERROR("lli_smd:\n");
1369                                 dump_lsm(D_ERROR, lli->lli_smd);
1370                                 CERROR("lsm:\n");
1371                                 dump_lsm(D_ERROR, lsm);
1372                                 LBUG();
1373                         }
1374                         /* XXX FIXME -- We should decide on a safer (atomic) and
1375                          * more elegant way to update the lsm */
1376                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1377                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1378                                         lsm->lsm_oinfo[i].loi_id;
1379                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1380                                         lsm->lsm_oinfo[i].loi_gr;
1381                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1382                                         lsm->lsm_oinfo[i].loi_ost_idx;
1383                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1384                                         lsm->lsm_oinfo[i].loi_ost_gen;
1385                         }
1386                 }
1387                 /* bug 2844 - limit i_blksize for broken user-space apps */
1388                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1389                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1390                 if (lli->lli_smd != lsm)
1391                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1392         }
1393
1394         if (mea != NULL) {
1395                 if (lli->lli_mea == NULL) {
1396                         lli->lli_mea = mea;
1397                 } else {
1398                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1399                                 CERROR("mea mismatch for inode %lu\n",
1400                                         inode->i_ino);
1401                                 LBUG();
1402                         }
1403                 }
1404                 if (lli->lli_mea != mea)
1405                         obd_free_memmd(ll_i2mdexp(inode),
1406                                        (struct lov_stripe_md **) &mea);
1407         }
1408
1409         if (body->valid & OBD_MD_FID)
1410                 id_assign_fid(&lli->lli_id, &body->id1);
1411
1412         if (body->valid & OBD_MD_FLID)
1413                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1414
1415         if (body->valid & OBD_MD_FLGENER)
1416                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1417
1418         spin_lock(&lli->lli_lock);
1419         if (ll_acl_access != NULL) {
1420                 if (lli->lli_acl_access != NULL)
1421                         posix_acl_release(lli->lli_acl_access);
1422                 lli->lli_acl_access = ll_acl_access;
1423         }
1424         spin_unlock(&lli->lli_lock);
1425
1426         if (body->valid & OBD_MD_FLID)
1427                 inode->i_ino = id_ino(&body->id1);
1428         if (body->valid & OBD_MD_FLGENER)
1429                 inode->i_generation = id_gen(&body->id1);
1430         if (body->valid & OBD_MD_FLATIME)
1431                 LTIME_S(inode->i_atime) = body->atime;
1432         if (body->valid & OBD_MD_FLMTIME &&
1433             body->mtime > LTIME_S(inode->i_mtime)) {
1434                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1435                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1436                 LTIME_S(inode->i_mtime) = body->mtime;
1437         }
1438         if (body->valid & OBD_MD_FLCTIME &&
1439             body->ctime > LTIME_S(inode->i_ctime))
1440                 LTIME_S(inode->i_ctime) = body->ctime;
1441         if (body->valid & OBD_MD_FLMODE) {
1442                 inode->i_mode = (inode->i_mode & S_IFMT) |
1443                         (body->mode & ~S_IFMT);
1444         }
1445         if (body->valid & OBD_MD_FLTYPE) {
1446                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1447                         (body->mode & S_IFMT);
1448         }
1449         if (body->valid & OBD_MD_FLUID)
1450                 inode->i_uid = body->uid;
1451         if (body->valid & OBD_MD_FLGID)
1452                 inode->i_gid = body->gid;
1453         if (body->valid & OBD_MD_FLFLAGS)
1454                 inode->i_flags = body->flags;
1455         if (body->valid & OBD_MD_FLNLINK)
1456                 inode->i_nlink = body->nlink;
1457         if (body->valid & OBD_MD_FLRDEV)
1458 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1459                 inode->i_rdev = body->rdev;
1460 #else
1461                 inode->i_rdev = old_decode_dev(body->rdev);
1462 #endif
1463         if (body->valid & OBD_MD_FLSIZE)
1464                 inode->i_size = body->size;
1465         if (body->valid & OBD_MD_FLBLOCKS)
1466                 inode->i_blocks = body->blocks;
1467
1468         if (body->valid & OBD_MD_FLSIZE)
1469                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1470
1471 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1472         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1473 #endif
1474         LASSERT(id_fid(&lli->lli_id) != 0);
1475 }
1476
1477 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1478 static struct backing_dev_info ll_backing_dev_info = {
1479         .ra_pages       = 0,    /* No readahead */
1480         .memory_backed  = 0,    /* Does contribute to dirty memory */
1481 };
1482 #endif
1483
1484 void ll_read_inode2(struct inode *inode, void *opaque)
1485 {
1486         struct lustre_md *md = opaque;
1487         struct ll_inode_info *lli = ll_i2info(inode);
1488         ENTRY;
1489
1490         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1491                inode->i_generation, inode);
1492
1493         ll_lli_init(lli);
1494
1495         LASSERT(!lli->lli_smd);
1496
1497         /* Core attributes from the MDS first.  This is a new inode, and
1498          * the VFS doesn't zero times in the core inode so we have to do
1499          * it ourselves.  They will be overwritten by either MDS or OST
1500          * attributes - we just need to make sure they aren't newer. */
1501         LTIME_S(inode->i_mtime) = 0;
1502         LTIME_S(inode->i_atime) = 0;
1503         LTIME_S(inode->i_ctime) = 0;
1504
1505         inode->i_rdev = 0;
1506         ll_update_inode(inode, md);
1507
1508         /* OIDEBUG(inode); */
1509
1510         if (S_ISREG(inode->i_mode)) {
1511                 inode->i_op = &ll_file_inode_operations;
1512                 inode->i_fop = &ll_file_operations;
1513                 inode->i_mapping->a_ops = &ll_aops;
1514                 EXIT;
1515         } else if (S_ISDIR(inode->i_mode)) {
1516                 inode->i_op = &ll_dir_inode_operations;
1517                 inode->i_fop = &ll_dir_operations;
1518                 inode->i_mapping->a_ops = &ll_dir_aops;
1519                 EXIT;
1520         } else if (S_ISLNK(inode->i_mode)) {
1521                 inode->i_op = &ll_fast_symlink_inode_operations;
1522                 EXIT;
1523         } else {
1524                 inode->i_op = &ll_special_inode_operations;
1525
1526 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1527                 init_special_inode(inode, inode->i_mode,
1528                                    kdev_t_to_nr(inode->i_rdev));
1529
1530                 /* initializing backing dev info. */
1531                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1532 #else
1533                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1534 #endif
1535                 lli->ll_save_ifop = inode->i_fop;
1536
1537                 if (S_ISCHR(inode->i_mode))
1538                         inode->i_fop = &ll_special_chr_inode_fops;
1539                 else if (S_ISBLK(inode->i_mode))
1540                         inode->i_fop = &ll_special_blk_inode_fops;
1541                 else if (S_ISFIFO(inode->i_mode))
1542                         inode->i_fop = &ll_special_fifo_inode_fops;
1543                 else if (S_ISSOCK(inode->i_mode))
1544                         inode->i_fop = &ll_special_sock_inode_fops;
1545
1546                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1547                       inode->i_fop);
1548
1549                 if (lli->ll_save_ifop->owner) {
1550                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1551                               lli->ll_save_ifop->owner);
1552                 }
1553                 EXIT;
1554         }
1555 }
1556
1557 void ll_delete_inode(struct inode *inode)
1558 {
1559         struct ll_sb_info *sbi = ll_i2sbi(inode);
1560         struct lustre_id id;
1561         int rc;
1562         ENTRY;
1563
1564         ll_inode2id(&id, inode);
1565
1566         rc = md_delete_inode(sbi->ll_md_exp, &id);
1567         if (rc) {
1568                 CERROR("md_delete_inode() failed, error %d\n",
1569                        rc);
1570         }
1571
1572         clear_inode(inode);
1573         EXIT;
1574 }
1575
1576 int ll_iocontrol(struct inode *inode, struct file *file,
1577                  unsigned int cmd, unsigned long arg)
1578 {
1579         struct ll_sb_info *sbi = ll_i2sbi(inode);
1580         struct ptlrpc_request *req = NULL;
1581         int rc, flags = 0;
1582         ENTRY;
1583
1584         switch(cmd) {
1585         case EXT3_IOC_GETFLAGS: {
1586                 struct lustre_id id;
1587                 __u64 valid = OBD_MD_FLFLAGS;
1588                 struct mds_body *body;
1589
1590                 ll_inode2id(&id, inode);
1591                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1592                 if (rc) {
1593                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1594                         RETURN(-abs(rc));
1595                 }
1596
1597                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1598
1599                 if (body->flags & S_APPEND)
1600                         flags |= EXT3_APPEND_FL;
1601                 if (body->flags & S_IMMUTABLE)
1602                         flags |= EXT3_IMMUTABLE_FL;
1603                 if (body->flags & S_NOATIME)
1604                         flags |= EXT3_NOATIME_FL;
1605
1606                 ptlrpc_req_finished (req);
1607
1608                 RETURN(put_user(flags, (int *)arg));
1609         }
1610         case EXT3_IOC_SETFLAGS: {
1611                 struct mdc_op_data *op_data;
1612                 struct iattr attr;
1613                 struct obdo *oa;
1614                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1615
1616                 if (get_user(flags, (int *)arg))
1617                         RETURN(-EFAULT);
1618
1619                 oa = obdo_alloc();
1620                 if (!oa)
1621                         RETURN(-ENOMEM);
1622
1623                 OBD_ALLOC(op_data, sizeof(*op_data));
1624                 if (op_data == NULL) {
1625                         obdo_free(oa);
1626                         RETURN(-ENOMEM);
1627                 }
1628                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1629
1630                 memset(&attr, 0x0, sizeof(attr));
1631                 attr.ia_attr_flags = flags;
1632                 attr.ia_valid |= ATTR_ATTR_FLAG;
1633
1634                 rc = md_setattr(sbi->ll_md_exp, op_data,
1635                                 &attr, NULL, 0, NULL, 0, &req);
1636                 OBD_FREE(op_data, sizeof(*op_data));
1637                 if (rc) {
1638                         ptlrpc_req_finished(req);
1639                         if (rc != -EPERM && rc != -EACCES)
1640                                 CERROR("md_setattr fails: rc = %d\n", rc);
1641                         obdo_free(oa);
1642                         RETURN(rc);
1643                 }
1644                 ptlrpc_req_finished(req);
1645
1646                 oa->o_id = lsm->lsm_object_id;
1647                 oa->o_gr = lsm->lsm_object_gr;
1648                 oa->o_flags = flags;
1649                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1650
1651                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1652                 obdo_free(oa);
1653                 if (rc) {
1654                         if (rc != -EPERM && rc != -EACCES)
1655                                 CERROR("md_setattr fails: rc = %d\n", rc);
1656                         RETURN(rc);
1657                 }
1658
1659                 if (flags & EXT3_APPEND_FL)
1660                         inode->i_flags |= S_APPEND;
1661                 else
1662                         inode->i_flags &= ~S_APPEND;
1663                 if (flags & EXT3_IMMUTABLE_FL)
1664                         inode->i_flags |= S_IMMUTABLE;
1665                 else
1666                         inode->i_flags &= ~S_IMMUTABLE;
1667                 if (flags & EXT3_NOATIME_FL)
1668                         inode->i_flags |= S_NOATIME;
1669                 else
1670                         inode->i_flags &= ~S_NOATIME;
1671
1672                 RETURN(0);
1673         }
1674         default:
1675                 RETURN(-ENOSYS);
1676         }
1677
1678         RETURN(0);
1679 }
1680
1681 /* this is only called in the case of forced umount. */
1682 void ll_umount_begin(struct super_block *sb)
1683 {
1684         struct ll_sb_info *sbi = ll_s2sbi(sb);
1685         struct obd_ioctl_data ioc_data = { 0 };
1686         struct obd_device *obd;
1687         ENTRY;
1688
1689         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1690                sb->s_count, atomic_read(&sb->s_active));
1691
1692         obd = class_exp2obd(sbi->ll_md_exp);
1693         if (obd == NULL) {
1694                 CERROR("Invalid MDC connection handle "LPX64"\n",
1695                        sbi->ll_md_exp->exp_handle.h_cookie);
1696                 EXIT;
1697                 return;
1698         }
1699         obd->obd_no_recov = 1;
1700         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1701                       sizeof(ioc_data), &ioc_data, NULL);
1702
1703         obd = class_exp2obd(sbi->ll_dt_exp);
1704         if (obd == NULL) {
1705                 CERROR("Invalid LOV connection handle "LPX64"\n",
1706                        sbi->ll_dt_exp->exp_handle.h_cookie);
1707                 EXIT;
1708                 return;
1709         }
1710
1711         obd->obd_no_recov = 1;
1712         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1713                       sizeof(ioc_data), &ioc_data, NULL);
1714
1715         /*
1716          * really, we'd like to wait until there are no requests outstanding,
1717          * and then continue.  For now, we just invalidate the requests,
1718          * schedule, and hope.
1719          */
1720         schedule();
1721
1722         EXIT;
1723 }
1724
1725 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1726                   struct inode **inode, struct ptlrpc_request *req,
1727                   int offset, struct super_block *sb)
1728 {
1729         struct lustre_md md;
1730         int rc = 0;
1731
1732         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1733         if (rc)
1734                 RETURN(rc);
1735
1736         if (*inode) {
1737                 ll_update_inode(*inode, &md);
1738         } else {
1739                 LASSERT(sb);
1740                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1741                 if (*inode == NULL || is_bad_inode(*inode)) {
1742                         /* free the lsm if we allocated one above */
1743                         if (md.lsm != NULL)
1744                                 obd_free_memmd(dt_exp, &md.lsm);
1745                         if (md.mea != NULL)
1746                                 obd_free_memmd(md_exp,
1747                                                (struct lov_stripe_md**)&md.mea);
1748                         rc = -ENOMEM;
1749                         CERROR("new_inode -fatal: rc %d\n", rc);
1750                 }
1751         }
1752
1753         RETURN(rc);
1754 }
1755
1756 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1757                char *filename, struct lustre_id *ret)
1758 {
1759         struct ptlrpc_request *request = NULL;
1760         struct mds_body *body;
1761         int rc;
1762
1763         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1764                              OBD_MD_FID, 0, &request);
1765         if (rc < 0) {
1766                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1767                        filename, rc);
1768                 return rc;
1769         }
1770
1771         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1772         LASSERT(body != NULL);
1773         LASSERT_REPSWABBED(request, 0);
1774
1775         *ret = body->id1;
1776         ptlrpc_req_finished(request);
1777
1778         return rc;
1779 }