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_snap.h>
  36 #include "llite_internal.h"
  37
  38 kmem_cache_t *ll_file_data_slab;
  39
  40 extern struct address_space_operations ll_aops;
  41 extern struct address_space_operations ll_dir_aops;
  42
  43 #ifndef log2
  44 #define log2(n) ffz(~(n))
  45 #endif
  46
  47 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
  48 {
  49         struct ll_sb_info *sbi = NULL;
  50         class_uuid_t uuid;
  51         ENTRY;
  52
  53         OBD_ALLOC(sbi, sizeof(*sbi));
  54         if (!sbi)
  55                 RETURN(NULL);
  56
  57         spin_lock_init(&sbi->ll_lock);
  58         INIT_LIST_HEAD(&sbi->ll_pglist);
  59         sbi->ll_pglist_gen = 0;
  60         sbi->ll_max_read_ahead_pages = SBI_DEFAULT_RA_MAX;
  61         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  62         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  63         INIT_LIST_HEAD(&sbi->ll_mnt_list);
  64         sema_init(&sbi->ll_gns_sem, 1);
  65         init_completion(&sbi->ll_gns_completion);
  66         sbi->ll_gns_state = LL_GNS_STATE_IDLE;
  67         sbi->ll_gns_timer.data = (unsigned long)sbi;
  68         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  69         init_timer(&sbi->ll_gns_timer);
  70         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
  71
  72         ll_set_sbi(sb, sbi);
  73
  74         generate_random_uuid(uuid);
  75         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  76         RETURN(sbi);
  77 }
  78
  79 void lustre_free_sbi(struct super_block *sb)
  80 {
  81         struct ll_sb_info *sbi = ll_s2sbi(sb);
  82         ENTRY;
  83
  84         if (sbi != NULL) {
  85                 list_del(&sbi->ll_gns_sbi_head);
  86                 del_timer(&sbi->ll_gns_timer);
  87                 OBD_FREE(sbi, sizeof(*sbi));
  88         }
  89         ll_set_sbi(sb, NULL);
  90         EXIT;
  91 }
  92
  93 int lustre_init_ea_size(struct ll_sb_info *sbi)
  94 {
  95         struct lov_desc desc;
  96         int rc, valsize;
  97
  98         /* OSC may have larger ea size */
  99         valsize = sizeof(desc);
 100         rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1, "lovdesc",
 101                           &valsize, &desc);
 102         if (rc)
 103                 RETURN(rc);
 104         obd_init_ea_size(sbi->ll_mdc_exp, obd_size_diskmd(sbi->ll_osc_exp, NULL),
 105                          desc.ld_tgt_count*sizeof(struct llog_cookie));
 106
 107         RETURN(rc);
 108 }
 109 #if CONFIG_SNAPFS
 110 int lustre_set_clone_info(struct super_block *sb, int clone_index)
 111 {
 112         struct ll_sb_info *sbi = ll_s2sbi(sb);
 113
 114         ENTRY;
 115
 116         CDEBUG(D_INFO, "set clone index %d\n", clone_index);
 117         if (!clone_index)
 118                 RETURN(0);
 119
 120         if (sbi->ll_mdc_exp) {
 121                 struct obd_device *obd = class_exp2obd(sbi->ll_mdc_exp);
 122                 struct client_obd *cl_obd = &obd->u.cli;
 123
 124                 OBD_ALLOC(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 125                 if (!cl_obd->cl_clone_info)
 126                         RETURN(-ENOMEM);
 127                 SET_CLONE_INDEX(cl_obd->cl_clone_info, clone_index);
 128                 SET_CLONE_FLAGS(cl_obd->cl_clone_info, SM_CLONE_FS);
 129                 cl_obd->cl_clone_info->cl_clone_info->clone_magic =
 130                                                   cpu_to_le32(CLONE_INFO_MAGI);
 131         }
 132
 133         if (sbi->ll_osc_exp) {
 134                 struct obd_device *obd = class_exp2obd(sbi->ll_osc_exp);
 135                 struct client_obd *cl_obd = &obd->u.cli;
 136
 137                 OBD_ALLOC(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 138                 if (!cl_obd->cl_clone_info)
 139                         RETURN(-ENOMEM);
 140                 SET_CLONE_INDEX(cl_obd->cl_clone_info, clone_index);
 141                 SET_CLONE_FLAGS(cl_obd->cl_clone_info, SM_CLONE_FS);
 142                 cl_obd->cl_clone_info->cl_clone_info->clone_magic =
 143                                                   cpu_to_le32(CLONE_INFO_MAGI);
 144         }
 145         RETURN(0);
 146 }
 147
 148 int lustre_cleanup_clone_info(struct super_block *sb)
 149 {
 150         struct ll_sb_info *sbi = ll_s2sbi(sb);
 151
 152         ENTRY;
 153
 154         if (sbi->ll_mdc_exp) {
 155                 struct obd_device *obd = class_exp2obd(sbi->ll_mdc_exp);
 156                 struct client_obd *cl_obd = &obd->u.cli;
 157
 158                 if (!cl_obd->cl_clone_info)
 159                         RETURN(0);
 160                 CDEBUG(D_INFO, "clean clone info %p\n", cl_obd->cl_clone_info);
 161                 OBD_FREE(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 162         }
 163
 164         if (sbi->ll_osc_exp) {
 165                 struct obd_device *obd = class_exp2obd(sbi->ll_osc_exp);
 166                 struct client_obd *cl_obd = &obd->u.cli;
 167
 168                 if (!cl_obd->cl_clone_info)
 169                         RETURN(0);
 170                 CDEBUG(D_INFO, "clean clone info %p\n", cl_obd->cl_clone_info);
 171                 OBD_FREE(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 172         }
 173         RETURN(0);
 174 }
 175 #endif
 176
 177
 178 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 179 {
 180         struct inode *root = 0;
 181         struct ll_sb_info *sbi = ll_s2sbi(sb);
 182         struct obd_device *obd;
 183         struct ll_fid rootfid;
 184         struct obd_statfs osfs;
 185         struct ptlrpc_request *request = NULL;
 186         struct lustre_handle osc_conn = {0, };
 187         struct lustre_handle mdc_conn = {0, };
 188         struct lustre_md md;
 189         kdev_t devno;
 190         int err;
 191
 192         obd = class_name2obd(mdc);
 193         if (!obd) {
 194                 CERROR("MDC %s: not setup or attached\n", mdc);
 195                 RETURN(-EINVAL);
 196         }
 197
 198         if (proc_lustre_fs_root) {
 199                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 200                                                   osc, mdc);
 201                 if (err < 0)
 202                         CERROR("could not register mount in /proc/lustre");
 203         }
 204
 205         err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid, 0);
 206         if (err == -EBUSY) {
 207                 CERROR("An MDS (mdc %s) is performing recovery, of which this"
 208                        " client is not a part.  Please wait for recovery to "
 209                        "complete, abort, or time out.\n", mdc);
 210                 GOTO(out, err);
 211         } else if (err) {
 212                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 213                 GOTO(out, err);
 214         }
 215         sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
 216
 217         err = obd_statfs(obd, &osfs, jiffies - HZ);
 218         if (err)
 219                 GOTO(out_mdc, err);
 220
 221         LASSERT(osfs.os_bsize);
 222         sb->s_blocksize = osfs.os_bsize;
 223         sb->s_blocksize_bits = log2(osfs.os_bsize);
 224         sb->s_magic = LL_SUPER_MAGIC;
 225         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 226
 227         devno = get_uuid2int(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid,
 228                              strlen(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid));
 229
 230         sb->s_dev = devno;
 231
 232         obd = class_name2obd(osc);
 233         if (!obd) {
 234                 CERROR("OSC %s: not setup or attached\n", osc);
 235                 GOTO(out_mdc, err);
 236         }
 237
 238         err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid, 0);
 239         if (err == -EBUSY) {
 240                 CERROR("An OST (osc %s) is performing recovery, of which this"
 241                        " client is not a part.  Please wait for recovery to "
 242                        "complete, abort, or time out.\n", osc);
 243                 GOTO(out, err);
 244         } else if (err) {
 245                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 246                 GOTO(out_mdc, err);
 247         }
 248         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 249
 250         lustre_init_ea_size(sbi);
 251
 252         err = md_getstatus(sbi->ll_mdc_exp, &rootfid);
 253         if (err) {
 254                 CERROR("cannot mds_connect: rc = %d\n", err);
 255                 GOTO(out_osc, err);
 256         }
 257         CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
 258         sbi->ll_rootino = rootfid.id;
 259
 260         sb->s_op = &lustre_super_operations;
 261
 262         /* make root inode
 263          * XXX: move this to after cbd setup? */
 264         err = md_getattr(sbi->ll_mdc_exp, &rootfid,
 265                          OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS, 0, &request);
 266         if (err) {
 267                 CERROR("md_getattr failed for root: rc = %d\n", err);
 268                 GOTO(out_osc, err);
 269         }
 270
 271         err = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
 272                                 sbi->ll_osc_exp, &md);
 273         if (err) {
 274                 CERROR("failed to understand root inode md: rc = %d\n",err);
 275                 ptlrpc_req_finished (request);
 276                 GOTO(out_osc, err);
 277         }
 278
 279         LASSERT(sbi->ll_rootino != 0);
 280         root = ll_iget(sb, sbi->ll_rootino, &md);
 281
 282         ptlrpc_req_finished(request);
 283
 284         if (root == NULL || is_bad_inode(root)) {
 285                 /* XXX might need iput() for bad inode */
 286                 CERROR("lustre_lite: bad iget4 for root\n");
 287                 GOTO(out_root, err = -EBADF);
 288         }
 289
 290         err = ll_close_thread_start(&sbi->ll_lcq);
 291         if (err) {
 292                 CERROR("cannot start close thread: rc %d\n", err);
 293                 GOTO(out_root, err);
 294         }
 295
 296         ll_gns_add_timer(sbi);
 297
 298         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 299            backing dev info assigned to inode mapping is used for
 300            determining maximal readahead. */
 301 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 302         /* bug 2805 - set VM readahead to zero */
 303         vm_max_readahead = vm_min_readahead = 0;
 304 #endif
 305
 306         sb->s_root = d_alloc_root(root);
 307
 308 #ifdef S_PDIROPS
 309         CWARN("Enabling PDIROPS\n");
 310         sb->s_flags |= S_PDIROPS;
 311 #endif
 312
 313         RETURN(err);
 314
 315 out_root:
 316         if (root)
 317                 iput(root);
 318 out_osc:
 319         obd_disconnect(sbi->ll_osc_exp, 0);
 320 out_mdc:
 321         obd_disconnect(sbi->ll_mdc_exp, 0);
 322 out:
 323         lprocfs_unregister_mountpoint(sbi);
 324         RETURN(err);
 325 }
 326
 327 void lustre_common_put_super(struct super_block *sb)
 328 {
 329         struct ll_sb_info *sbi = ll_s2sbi(sb);
 330         struct hlist_node *tmp, *next;
 331         ENTRY;
 332
 333         ll_gns_del_timer(sbi);
 334
 335         ll_close_thread_shutdown(sbi->ll_lcq);
 336
 337         list_del(&sbi->ll_conn_chain);
 338         obd_disconnect(sbi->ll_osc_exp, 0);
 339
 340         lprocfs_unregister_mountpoint(sbi);
 341         if (sbi->ll_proc_root) {
 342                 lprocfs_remove(sbi->ll_proc_root);
 343                 sbi->ll_proc_root = NULL;
 344         }
 345
 346         obd_disconnect(sbi->ll_mdc_exp, 0);
 347
 348         // We do this to get rid of orphaned dentries. That is not really trw.
 349         spin_lock(&dcache_lock);
 350         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 351                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 352                 CWARN("orphan dentry %*s (%p) at unmount\n",
 353                       dentry->d_name.len, dentry->d_name.name, dentry);
 354                 shrink_dcache_parent(dentry);
 355         }
 356         spin_unlock(&dcache_lock);
 357         EXIT;
 358 }
 359
 360
 361 char *ll_read_opt(const char *opt, char *data)
 362 {
 363         char *value;
 364         char *retval;
 365         ENTRY;
 366
 367         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 368         if (strncmp(opt, data, strlen(opt)))
 369                 RETURN(NULL);
 370         if ((value = strchr(data, '=')) == NULL)
 371                 RETURN(NULL);
 372
 373         value++;
 374         OBD_ALLOC(retval, strlen(value) + 1);
 375         if (!retval) {
 376                 CERROR("out of memory!\n");
 377                 RETURN(NULL);
 378         }
 379
 380         memcpy(retval, value, strlen(value)+1);
 381         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 382         RETURN(retval);
 383 }
 384
 385 int ll_set_opt(const char *opt, char *data, int fl)
 386 {
 387         ENTRY;
 388
 389         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 390         if (strncmp(opt, data, strlen(opt)))
 391                 RETURN(0);
 392         else
 393                 RETURN(fl);
 394 }
 395
 396 void ll_options(char *options, char **ost, char **mdc, int *flags,
 397                 char **clone_opts)
 398 {
 399         char *this_char;
 400 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 401         char *opt_ptr = options;
 402 #endif
 403         ENTRY;
 404
 405         if (!options) {
 406                 EXIT;
 407                 return;
 408         }
 409
 410 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 411         for (this_char = strtok (options, ",");
 412              this_char != NULL;
 413              this_char = strtok (NULL, ",")) {
 414 #else
 415         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 416 #endif
 417                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 418                 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
 419                         continue;
 420                 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
 421                         continue;
 422                 if (!*clone_opts && (*clone_opts = ll_read_opt("clone",
 423                                                                 this_char)))
 424                         continue;
 425                 if (!(*flags & LL_SBI_NOLCK) &&
 426                     ((*flags) = (*flags) |
 427                                 ll_set_opt("nolock", this_char,
 428                                            LL_SBI_NOLCK)))
 429                         continue;
 430         }
 431
 432         EXIT;
 433 }
 434
 435 void ll_lli_init(struct ll_inode_info *lli)
 436 {
 437         sema_init(&lli->lli_open_sem, 1);
 438         lli->lli_flags = 0;
 439         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 440         spin_lock_init(&lli->lli_lock);
 441         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 442         lli->lli_inode_magic = LLI_INODE_MAGIC;
 443 }
 444
 445 int ll_fill_super(struct super_block *sb, void *data, int silent)
 446 {
 447         struct ll_sb_info *sbi;
 448         char *osc = NULL;
 449         char *mdc = NULL;
 450         char *clone_opts = NULL;
 451         int err;
 452         ENTRY;
 453
 454         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 455
 456         sbi = lustre_init_sbi(sb);
 457         if (!sbi)
 458                 RETURN(-ENOMEM);
 459
 460         sbi->ll_flags |= LL_SBI_READAHEAD;
 461         ll_options(data, &osc, &mdc, &sbi->ll_flags, &clone_opts);
 462
 463         if (!osc) {
 464                 CERROR("no osc\n");
 465                 GOTO(out, err = -EINVAL);
 466         }
 467
 468         if (!mdc) {
 469                 CERROR("no mdc\n");
 470                 GOTO(out, err = -EINVAL);
 471         }
 472
 473         err = lustre_common_fill_super(sb, mdc, osc);
 474 #if CONFIG_SNAPFS
 475         if (clone_opts) {
 476                 int clone_index = 0;
 477                 char *endp;
 478
 479                 /*set clone info to the super block*/
 480                 clone_index = simple_strtoul(clone_opts, &endp, 0);
 481                 err = lustre_set_clone_info(sb, clone_index);
 482         }
 483 #endif
 484 out:
 485         if (err)
 486                 lustre_free_sbi(sb);
 487
 488         if (mdc)
 489                 OBD_FREE(mdc, strlen(mdc) + 1);
 490         if (osc)
 491                 OBD_FREE(osc, strlen(osc) + 1);
 492         if (clone_opts)
 493                 OBD_FREE(clone_opts, strlen(clone_opts) + 1);
 494
 495         RETURN(err);
 496 } /* ll_read_super */
 497
 498 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 499                               struct config_llog_instance *cfg, int allow_recov)
 500 {
 501         struct lustre_cfg lcfg;
 502         struct portals_cfg pcfg;
 503         char * peer = "MDS_PEER_UUID";
 504         struct obd_device *obd;
 505         struct lustre_handle mdc_conn = {0, };
 506         struct obd_export *exp;
 507         char * name = "mdc_dev";
 508         class_uuid_t uuid;
 509         struct obd_uuid mdc_uuid;
 510         struct llog_ctxt *ctxt;
 511         int rc = 0;
 512         int err;
 513         ENTRY;
 514
 515         if (lmd_bad_magic(lmd))
 516                 RETURN(-EINVAL);
 517
 518         generate_random_uuid(uuid);
 519         class_uuid_unparse(uuid, &mdc_uuid);
 520
 521         if (lmd->lmd_local_nid) {
 522                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 523                 pcfg.pcfg_nal = lmd->lmd_nal;
 524                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 525                 err = libcfs_nal_cmd(&pcfg);
 526                 if (err <0)
 527                         GOTO(out, err);
 528         }
 529
 530         if (lmd->lmd_nal == SOCKNAL) {
 531                 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
 532                 pcfg.pcfg_nal     = lmd->lmd_nal;
 533                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 534                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 535                 pcfg.pcfg_misc    = lmd->lmd_port;
 536                 pcfg.pcfg_size    = 8388608;
 537                 pcfg.pcfg_flags   = 0x4; /*share*/
 538                 err = libcfs_nal_cmd(&pcfg);
 539                 if (err <0)
 540                         GOTO(out, err);
 541         }
 542
 543         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
 544         lcfg.lcfg_nid = lmd->lmd_server_nid;
 545         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 546         lcfg.lcfg_inlbuf1 = peer;
 547         lcfg.lcfg_nal = lmd->lmd_nal;
 548         err = class_process_config(&lcfg);
 549         if (err < 0)
 550                 GOTO(out_del_conn, err);
 551
 552         LCFG_INIT(lcfg, LCFG_ATTACH, name);
 553         lcfg.lcfg_inlbuf1 = "mdc";
 554         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 555         lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
 556         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 557         err = class_process_config(&lcfg);
 558         if (err < 0)
 559                 GOTO(out_del_uuid, err);
 560
 561         LCFG_INIT(lcfg, LCFG_SETUP, name);
 562         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
 563         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 564         lcfg.lcfg_inlbuf2 = peer;
 565         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 566         err = class_process_config(&lcfg);
 567         if (err < 0)
 568                 GOTO(out_detach, err);
 569
 570         obd = class_name2obd(name);
 571         if (obd == NULL)
 572                 GOTO(out_cleanup, err = -EINVAL);
 573
 574         /* Disable initial recovery on this import */
 575         err = obd_set_info(obd->obd_self_export,
 576                            strlen("initial_recov"), "initial_recov",
 577                            sizeof(allow_recov), &allow_recov);
 578         if (err)
 579                 GOTO(out_cleanup, err);
 580
 581         err = obd_connect(&mdc_conn, obd, &mdc_uuid, 0);
 582         if (err) {
 583                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
 584                 GOTO(out_cleanup, err);
 585         }
 586
 587         exp = class_conn2export(&mdc_conn);
 588
 589         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 590         rc = class_config_process_llog(ctxt, profile, cfg);
 591         if (rc)
 592                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 593
 594         err = obd_disconnect(exp, 0);
 595
 596 out_cleanup:
 597         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
 598         err = class_process_config(&lcfg);
 599         if (err < 0)
 600                 GOTO(out, err);
 601
 602 out_detach:
 603         LCFG_INIT(lcfg, LCFG_DETACH, name);
 604         err = class_process_config(&lcfg);
 605         if (err < 0)
 606                 GOTO(out, err);
 607
 608 out_del_uuid:
 609         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
 610         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 611         lcfg.lcfg_inlbuf1 = peer;
 612         err = class_process_config(&lcfg);
 613
 614 out_del_conn:
 615         if (lmd->lmd_nal == SOCKNAL) {
 616                 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
 617                 pcfg.pcfg_nal     = lmd->lmd_nal;
 618                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 619                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 620                 pcfg.pcfg_flags   = 1; /*share*/
 621                 err = libcfs_nal_cmd(&pcfg);
 622                 if (err <0)
 623                         GOTO(out, err);
 624         }
 625 out:
 626         if (rc == 0)
 627                 rc = err;
 628
 629         RETURN(rc);
 630 }
 631
 632 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 633 {
 634         struct lustre_mount_data * lmd = data;
 635         struct ll_sb_info *sbi;
 636         char *osc = NULL;
 637         char *mdc = NULL;
 638         int err;
 639         ENTRY;
 640
 641         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 642         if (lmd_bad_magic(lmd))
 643                 RETURN(-EINVAL);
 644
 645         sbi = lustre_init_sbi(sb);
 646         if (!sbi)
 647                 RETURN(-ENOMEM);
 648
 649         sbi->ll_flags |= LL_SBI_READAHEAD;
 650
 651         if (lmd->lmd_profile) {
 652                 struct lustre_profile *lprof;
 653                 struct config_llog_instance cfg;
 654                 int len;
 655
 656                 if (lmd->lmd_mds[0] == '\0') {
 657                         CERROR("no mds name\n");
 658                         GOTO(out_free, err = -EINVAL);
 659                 }
 660
 661                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 662                 if (sbi->ll_lmd == NULL)
 663                         GOTO(out_free, err = -ENOMEM);
 664                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 665
 666                 /* generate a string unique to this super, let's try
 667                  the address of the super itself.*/
 668                 len = (sizeof(sb) * 2) + 1;
 669                 OBD_ALLOC(sbi->ll_instance, len);
 670                 if (sbi->ll_instance == NULL)
 671                         GOTO(out_free, err = -ENOMEM);
 672                 sprintf(sbi->ll_instance, "%p", sb);
 673
 674                 cfg.cfg_instance = sbi->ll_instance;
 675                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 676                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 677                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 678                 if (err < 0) {
 679                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 680                         GOTO(out_free, err);
 681                 }
 682
 683                 lprof = class_get_profile(lmd->lmd_profile);
 684                 if (lprof == NULL) {
 685                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 686                         GOTO(out_free, err = -EINVAL);
 687                 }
 688                 if (osc)
 689                         OBD_FREE(osc, strlen(osc) + 1);
 690                 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
 691                           strlen(sbi->ll_instance) + 2);
 692                 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
 693
 694                 if (mdc)
 695                         OBD_FREE(mdc, strlen(mdc) + 1);
 696                 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
 697                           strlen(sbi->ll_instance) + 2);
 698                 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
 699         }
 700
 701         if (!osc) {
 702                 CERROR("no osc\n");
 703                 GOTO(out_free, err = -EINVAL);
 704         }
 705
 706         if (!mdc) {
 707                 CERROR("no mdc\n");
 708                 GOTO(out_free, err = -EINVAL);
 709         }
 710
 711         err = lustre_common_fill_super(sb, mdc, osc);
 712
 713         if (err)
 714                 GOTO(out_free, err);
 715
 716 #if CONFIG_SNAPFS
 717         if (lmd->lmd_clone_index) {
 718                 err = lustre_set_clone_info(sb, lmd->lmd_clone_index);
 719         }
 720
 721         if (err)
 722                 GOTO(out_free, err);
 723 #endif
 724
 725 out_dev:
 726         if (mdc)
 727                 OBD_FREE(mdc, strlen(mdc) + 1);
 728         if (osc)
 729                 OBD_FREE(osc, strlen(osc) + 1);
 730
 731         RETURN(err);
 732
 733 out_free:
 734         if (sbi->ll_lmd) {
 735                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 736                 int err;
 737
 738                 if (sbi->ll_instance != NULL) {
 739                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 740                         char * cln_prof;
 741                         struct config_llog_instance cfg;
 742
 743                         cfg.cfg_instance = sbi->ll_instance;
 744                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 745
 746                         OBD_ALLOC(cln_prof, len);
 747                         sprintf(cln_prof, "%s-clean", lmd->lmd_profile);
 748
 749                         err = lustre_process_log(lmd, cln_prof, &cfg, 0);
 750                         if (err < 0)
 751                                 CERROR("Unable to process log: %s\n", cln_prof);
 752                         OBD_FREE(cln_prof, len);
 753                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
 754                 }
 755                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 756         }
 757 #if CONFIG_SNAPFS
 758         lustre_cleanup_clone_info(sb);
 759 #endif
 760         lustre_free_sbi(sb);
 761
 762         goto out_dev;
 763 } /* lustre_fill_super */
 764
 765 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 766 {
 767         struct lustre_cfg lcfg;
 768         struct obd_device *obd;
 769         int next = 0;
 770
 771         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 772         {
 773                 int err;
 774
 775                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
 776                 err = class_process_config(&lcfg);
 777                 if (err) {
 778                         CERROR("cleanup failed: %s\n", obd->obd_name);
 779                         //continue;
 780                 }
 781
 782                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
 783                 err = class_process_config(&lcfg);
 784                 if (err) {
 785                         CERROR("detach failed: %s\n", obd->obd_name);
 786                         //continue;
 787                 }
 788         }
 789
 790         if (sbi->ll_lmd != NULL)
 791                 class_del_profile(sbi->ll_lmd->lmd_profile);
 792 }
 793
 794 void lustre_put_super(struct super_block *sb)
 795 {
 796         struct obd_device *obd;
 797         struct ll_sb_info *sbi = ll_s2sbi(sb);
 798         int force_umount = 0;
 799         ENTRY;
 800
 801         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 802         obd = class_exp2obd(sbi->ll_mdc_exp);
 803         if (obd)
 804                 force_umount = obd->obd_no_recov;
 805         obd = NULL;
 806
 807 #if CONFIG_SNAPFS
 808         lustre_cleanup_clone_info(sb);
 809 #endif
 810
 811         lustre_common_put_super(sb);
 812         if (sbi->ll_lmd != NULL) {
 813                 char * cln_prof;
 814                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 815                 int err;
 816                 struct config_llog_instance cfg;
 817
 818                 if (force_umount) {
 819                         CERROR("force umount, doing manual cleanup\n");
 820                         lustre_manual_cleanup(sbi);
 821                         GOTO(free_lmd, 0);
 822                 }
 823
 824                 cfg.cfg_instance = sbi->ll_instance;
 825                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 826
 827                 OBD_ALLOC(cln_prof, len);
 828                 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 829
 830                 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
 831                 if (err < 0) {
 832                         CERROR("Unable to process log: %s, doing manual cleanup"
 833                                "\n", cln_prof);
 834                         lustre_manual_cleanup(sbi);
 835                 }
 836
 837                 OBD_FREE(cln_prof, len);
 838         free_lmd:
 839                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 840                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 841         }
 842
 843         lustre_free_sbi(sb);
 844
 845         EXIT;
 846 } /* lustre_put_super */
 847
 848 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 849 {
 850         struct obd_export *mdc_exp = sbi->ll_mdc_exp;
 851         struct lustre_mount_data *lmd = sbi->ll_lmd;
 852         struct llog_ctxt *ctxt;
 853         struct config_llog_instance cfg;
 854         char *profile = lmd->lmd_profile, *name = NULL;
 855         int rc, namelen =  0, version;
 856         ENTRY;
 857
 858         if (profile == NULL)
 859                 RETURN(0);
 860         if (lmd == NULL) {
 861                 CERROR("Client not mounted with zero-conf; cannot process "
 862                        "update log.\n");
 863                 RETURN(0);
 864         }
 865
 866         rc = ldlm_cli_cancel_unused(mdc_exp->exp_obd->obd_namespace, NULL,
 867                                     LDLM_FL_CONFIG_CHANGE, NULL);
 868         if (rc != 0)
 869                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
 870
 871         rc = obd_cancel_unused(sbi->ll_osc_exp, NULL, LDLM_FL_CONFIG_CHANGE,
 872                                NULL);
 873         if (rc != 0)
 874                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 875
 876         cfg.cfg_instance = sbi->ll_instance;
 877         cfg.cfg_uuid = sbi->ll_sb_uuid;
 878         cfg.cfg_local_nid = lmd->lmd_local_nid;
 879
 880         namelen = strlen(profile) + 20; /* -clean-######### */
 881         OBD_ALLOC(name, namelen);
 882         if (name == NULL)
 883                 RETURN(-ENOMEM);
 884
 885         if (clean) {
 886                 version = sbi->ll_config_version - 1;
 887                 sprintf(name, "%s-clean-%d", profile, version);
 888         } else {
 889                 version = sbi->ll_config_version + 1;
 890                 sprintf(name, "%s-%d", profile, version);
 891         }
 892
 893         CWARN("Applying configuration log %s\n", name);
 894
 895         ctxt = llog_get_context(&mdc_exp->exp_obd->obd_llogs,
 896                                 LLOG_CONFIG_REPL_CTXT);
 897         rc = class_config_process_llog(ctxt, name, &cfg);
 898         if (rc == 0)
 899                 sbi->ll_config_version = version;
 900         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 901
 902         if (rc == 0 && clean == 0) {
 903                 struct lov_desc desc;
 904                 int rc, valsize;
 905                 valsize = sizeof(desc);
 906                 rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1,
 907                                   "lovdesc", &valsize, &desc);
 908
 909                 rc = obd_init_ea_size(mdc_exp,
 910                                       obd_size_diskmd(sbi->ll_osc_exp, NULL),
 911                                       (desc.ld_tgt_count *
 912                                        sizeof(struct llog_cookie)));
 913         }
 914         OBD_FREE(name, namelen);
 915         RETURN(rc);
 916 }
 917
 918 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 919 {
 920         struct inode *inode = NULL;
 921         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 922         if (lock->l_ast_data) {
 923                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 924                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 925                         inode = igrab(lock->l_ast_data);
 926                 } else {
 927                         CERROR("DEBUG: l_ast_data %p is bogus: magic %x\n",
 928                                lock->l_ast_data, lli->lli_inode_magic);
 929                 }
 930         }
 931         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 932         return inode;
 933 }
 934
 935 int null_if_equal(struct ldlm_lock *lock, void *data)
 936 {
 937         if (data == lock->l_ast_data)
 938                 lock->l_ast_data = NULL;
 939
 940         if (lock->l_req_mode != lock->l_granted_mode)
 941                 return LDLM_ITER_STOP;
 942
 943         return LDLM_ITER_CONTINUE;
 944 }
 945
 946 void ll_clear_inode(struct inode *inode)
 947 {
 948         struct ll_fid fid;
 949         struct ll_inode_info *lli = ll_i2info(inode);
 950         struct ll_sb_info *sbi = ll_i2sbi(inode);
 951         ENTRY;
 952
 953         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 954                inode->i_generation, inode);
 955
 956         lli->lli_inode_magic = LLI_INODE_DEAD;
 957         ll_inode2fid(&fid, inode);
 958         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 959         md_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
 960
 961         if (lli->lli_smd)
 962                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
 963                                   null_if_equal, inode);
 964
 965         if (lli->lli_smd) {
 966                 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
 967                 lli->lli_smd = NULL;
 968         }
 969
 970         if (lli->lli_mea) {
 971                 /* FIXME: change cbdata for mea here */
 972                 obd_free_memmd(sbi->ll_mdc_exp,
 973                                (struct lov_stripe_md **) &lli->lli_mea);
 974                 lli->lli_mea = NULL;
 975         }
 976
 977         if (lli->lli_symlink_name) {
 978                 OBD_FREE(lli->lli_symlink_name,
 979                          strlen(lli->lli_symlink_name) + 1);
 980                 lli->lli_symlink_name = NULL;
 981         }
 982
 983         EXIT;
 984 }
 985
 986 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 987  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 988  * keep these values until such a time that objects are allocated for it.
 989  * We do the MDS operations first, as it is checking permissions for us.
 990  * We don't to the MDS RPC if there is nothing that we want to store there,
 991  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 992  * going to do an RPC anyways.
 993  *
 994  * If we are doing a truncate, we will send the mtime and ctime updates
 995  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 996  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 997  * at the same time.
 998  */
 999 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1000 {
1001         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1002         struct ll_sb_info *sbi = ll_i2sbi(inode);
1003         struct ptlrpc_request *request = NULL;
1004         struct mdc_op_data op_data;
1005         int ia_valid = attr->ia_valid;
1006         int rc = 0;
1007         ENTRY;
1008
1009         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1010         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1011
1012         if (ia_valid & ATTR_SIZE) {
1013                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1014                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1015                                attr->ia_size, ll_file_maxbytes(inode));
1016                         RETURN(-EFBIG);
1017                 }
1018
1019                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1020         }
1021
1022         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1023         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1024                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1025                         RETURN(-EPERM);
1026         }
1027
1028         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1029         if (attr->ia_valid & ATTR_CTIME) {
1030                 attr->ia_ctime = CURRENT_TIME;
1031                 attr->ia_valid |= ATTR_CTIME_SET;
1032         }
1033         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1034                 attr->ia_atime = CURRENT_TIME;
1035                 attr->ia_valid |= ATTR_ATIME_SET;
1036         }
1037         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1038                 attr->ia_mtime = CURRENT_TIME;
1039                 attr->ia_valid |= ATTR_MTIME_SET;
1040         }
1041
1042         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1043                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1044                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1045                        LTIME_S(CURRENT_TIME));
1046         if (lsm)
1047                 attr->ia_valid &= ~ATTR_SIZE;
1048
1049         /* If only OST attributes being set on objects, don't do MDS RPC.
1050          * In that case, we need to check permissions and update the local
1051          * inode ourselves so we can call obdo_from_inode() always. */
1052         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
1053                 struct lustre_md md;
1054                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1055
1056                 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
1057                                 attr, NULL, 0, NULL, 0, &request);
1058                 if (rc) {
1059                         ptlrpc_req_finished(request);
1060                         if (rc != -EPERM && rc != -EACCES)
1061                                 CERROR("md_setattr fails: rc = %d\n", rc);
1062                         RETURN(rc);
1063                 }
1064
1065                 rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
1066                                        sbi->ll_osc_exp, &md);
1067                 if (rc) {
1068                         ptlrpc_req_finished(request);
1069                         RETURN(rc);
1070                 }
1071
1072                 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
1073                  * but needed to set timestamps backwards on utime. */
1074                 inode_setattr(inode, attr);
1075                 ll_update_inode(inode, &md);
1076                 ptlrpc_req_finished(request);
1077
1078                 if (!lsm || !S_ISREG(inode->i_mode)) {
1079                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1080                         RETURN(0);
1081                 }
1082         } else {
1083                 /* The OST doesn't check permissions, but the alternative is
1084                  * a gratuitous RPC to the MDS.  We already rely on the client
1085                  * to do read/write/truncate permission checks, so is mtime OK?
1086                  */
1087                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1088                         /* from sys_utime() */
1089                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1090                                 if (current->fsuid != inode->i_uid &&
1091                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1092                                         RETURN(rc);
1093                         } else {
1094                                 /* from inode_change_ok() */
1095                                 if (current->fsuid != inode->i_uid &&
1096                                     !capable(CAP_FOWNER))
1097                                         RETURN(-EPERM);
1098                         }
1099                 }
1100
1101                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1102                 inode_setattr(inode, attr);
1103         }
1104
1105         /* We really need to get our PW lock before we change inode->i_size.
1106          * If we don't we can race with other i_size updaters on our node, like
1107          * ll_file_read.  We can also race with i_size propogation to other
1108          * nodes through dirtying and writeback of final cached pages.  This
1109          * last one is especially bad for racing o_append users on other
1110          * nodes. */
1111         if (ia_valid & ATTR_SIZE) {
1112                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1113                                                            OBD_OBJECT_EOF } };
1114                 struct lustre_handle lockh = { 0 };
1115                 int err, ast_flags = 0;
1116                 /* XXX when we fix the AST intents to pass the discard-range
1117                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1118                  * XXX here. */
1119                 if (attr->ia_size == 0)
1120                         ast_flags = LDLM_AST_DISCARD_DATA;
1121
1122                 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
1123                 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
1124                 up(&inode->i_sem);
1125                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1126                                     ast_flags);
1127                 down(&inode->i_sem);
1128                 if (rc != 0)
1129                         RETURN(rc);
1130
1131                 rc = vmtruncate(inode, attr->ia_size);
1132
1133                 /* We need to drop the semaphore here, because this unlock may
1134                  * result in a cancellation, which will need the i_sem */
1135                 up(&inode->i_sem);
1136                 /* unlock now as we don't mind others file lockers racing with
1137                  * the mds updates below? */
1138                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1139                 down(&inode->i_sem);
1140                 if (err) {
1141                         CERROR("ll_extent_unlock failed: %d\n", err);
1142                         if (!rc)
1143                                 rc = err;
1144                 }
1145         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1146                 struct obdo oa;
1147
1148                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1149                        inode->i_ino, LTIME_S(attr->ia_mtime));
1150                 oa.o_id = lsm->lsm_object_id;
1151                 oa.o_gr = lsm->lsm_object_gr;
1152                 oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1153                 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1154                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1155                 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1156                 if (rc)
1157                         CERROR("obd_setattr fails: rc=%d\n", rc);
1158         }
1159         RETURN(rc);
1160 }
1161
1162 int ll_setattr(struct dentry *de, struct iattr *attr)
1163 {
1164         LBUG(); /* code is unused, but leave this in case of VFS changes */
1165         RETURN(-ENOSYS);
1166 }
1167
1168 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1169                        unsigned long max_age)
1170 {
1171         struct ll_sb_info *sbi = ll_s2sbi(sb);
1172         struct obd_statfs obd_osfs;
1173         int rc;
1174         ENTRY;
1175
1176         rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1177         if (rc) {
1178                 CERROR("mdc_statfs fails: rc = %d\n", rc);
1179                 RETURN(rc);
1180         }
1181
1182         osfs->os_type = sb->s_magic;
1183
1184         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1185                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1186
1187         rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1188         if (rc) {
1189                 CERROR("obd_statfs fails: rc = %d\n", rc);
1190                 RETURN(rc);
1191         }
1192
1193         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1194                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1195                obd_osfs.os_files);
1196
1197         osfs->os_blocks = obd_osfs.os_blocks;
1198         osfs->os_bfree = obd_osfs.os_bfree;
1199         osfs->os_bavail = obd_osfs.os_bavail;
1200
1201         /* If we don't have as many objects free on the OST as inodes
1202          * on the MDS, we reduce the total number of inodes to
1203          * compensate, so that the "inodes in use" number is correct.
1204          */
1205         if (obd_osfs.os_ffree < osfs->os_ffree) {
1206                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1207                         obd_osfs.os_ffree;
1208                 osfs->os_ffree = obd_osfs.os_ffree;
1209         }
1210
1211         RETURN(rc);
1212 }
1213
1214 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1215 {
1216         struct obd_statfs osfs;
1217         int rc;
1218
1219         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1220         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1221
1222         /* For now we will always get up-to-date statfs values, but in the
1223          * future we may allow some amount of caching on the client (e.g.
1224          * from QOS or lprocfs updates). */
1225         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1226         if (rc)
1227                 return rc;
1228
1229         statfs_unpack(sfs, &osfs);
1230
1231         if (sizeof(sfs->f_blocks) == 4) {
1232                 while (osfs.os_blocks > ~0UL) {
1233                         sfs->f_bsize <<= 1;
1234
1235                         osfs.os_blocks >>= 1;
1236                         osfs.os_bfree >>= 1;
1237                         osfs.os_bavail >>= 1;
1238                 }
1239         }
1240
1241         sfs->f_blocks = osfs.os_blocks;
1242         sfs->f_bfree = osfs.os_bfree;
1243         sfs->f_bavail = osfs.os_bavail;
1244
1245         return 0;
1246 }
1247
1248 void dump_lsm(int level, struct lov_stripe_md *lsm)
1249 {
1250         CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1251                "stripe_size %u, stripe_count %u\n",
1252                lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1253                lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1254 }
1255
1256 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1257 {
1258         struct ll_inode_info *lli = ll_i2info(inode);
1259         struct mds_body *body = md->body;
1260         struct lov_stripe_md *lsm = md->lsm;
1261         struct mea *mea = md->mea;
1262         ENTRY;
1263
1264         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1265         LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1266         if (lsm != NULL) {
1267                 LASSERT(lsm->lsm_object_gr > 0);
1268                 if (lli->lli_smd == NULL) {
1269                         lli->lli_smd = lsm;
1270                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1271                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1272                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1273                 } else {
1274                         int i;
1275                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1276                                 CERROR("lsm mismatch for inode %ld\n",
1277                                        inode->i_ino);
1278                                 CERROR("lli_smd:\n");
1279                                 dump_lsm(D_ERROR, lli->lli_smd);
1280                                 CERROR("lsm:\n");
1281                                 dump_lsm(D_ERROR, lsm);
1282                                 LBUG();
1283                         }
1284                         /* XXX FIXME -- We should decide on a safer (atomic) and
1285                          * more elegant way to update the lsm */
1286                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1287                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1288                                         lsm->lsm_oinfo[i].loi_id;
1289                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1290                                         lsm->lsm_oinfo[i].loi_gr;
1291                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1292                                         lsm->lsm_oinfo[i].loi_ost_idx;
1293                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1294                                         lsm->lsm_oinfo[i].loi_ost_gen;
1295                         }
1296                 }
1297                 /* bug 2844 - limit i_blksize for broken user-space apps */
1298                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1299                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1300                 if (lli->lli_smd != lsm)
1301                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
1302         }
1303
1304         if (mea != NULL) {
1305                 if (lli->lli_mea == NULL) {
1306                         lli->lli_mea = mea;
1307                 } else {
1308                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1309                                 CERROR("mea mismatch for inode %lu\n",
1310                                         inode->i_ino);
1311                                 LBUG();
1312                         }
1313                 }
1314                 if (lli->lli_mea != mea)
1315                         obd_free_memmd(ll_i2mdcexp(inode),
1316                                         (struct lov_stripe_md **) &mea);
1317
1318         }
1319
1320         if (body->valid & OBD_MD_FLID)
1321                 inode->i_ino = body->ino;
1322         if (body->valid & OBD_MD_FLATIME)
1323                 LTIME_S(inode->i_atime) = body->atime;
1324         if (body->valid & OBD_MD_FLMTIME &&
1325             body->mtime > LTIME_S(inode->i_mtime)) {
1326                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1327                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1328                 LTIME_S(inode->i_mtime) = body->mtime;
1329         }
1330         if (body->valid & OBD_MD_FLCTIME &&
1331             body->ctime > LTIME_S(inode->i_ctime))
1332                 LTIME_S(inode->i_ctime) = body->ctime;
1333         if (body->valid & OBD_MD_FLMODE)
1334                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1335         if (body->valid & OBD_MD_FLTYPE)
1336                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1337         if (body->valid & OBD_MD_FLUID)
1338                 inode->i_uid = body->uid;
1339         if (body->valid & OBD_MD_FLGID)
1340                 inode->i_gid = body->gid;
1341         if (body->valid & OBD_MD_FLFLAGS)
1342                 inode->i_flags = body->flags;
1343         if (body->valid & OBD_MD_FLNLINK)
1344                 inode->i_nlink = body->nlink;
1345         if (body->valid & OBD_MD_FLGENER)
1346                 inode->i_generation = body->generation;
1347         if (body->valid & OBD_MD_FLRDEV)
1348 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1349                 inode->i_rdev = body->rdev;
1350 #else
1351                 inode->i_rdev = old_decode_dev(body->rdev);
1352 #endif
1353         if (body->valid & OBD_MD_FLSIZE)
1354                 inode->i_size = body->size;
1355         if (body->valid & OBD_MD_FLBLOCKS)
1356                 inode->i_blocks = body->blocks;
1357
1358         if (body->valid & OBD_MD_FLSIZE)
1359                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1360
1361         lli->lli_mds = body->mds;
1362         inode->i_dev = (kdev_t) body->mds;
1363         LASSERT(body->mds < 1000);
1364 }
1365
1366 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1367 static struct backing_dev_info ll_backing_dev_info = {
1368         .ra_pages       = 0,    /* No readahead */
1369         .memory_backed  = 0,    /* Does contribute to dirty memory */
1370 };
1371 #endif
1372
1373 void ll_read_inode2(struct inode *inode, void *opaque)
1374 {
1375         struct lustre_md *md = opaque;
1376         struct ll_inode_info *lli = ll_i2info(inode);
1377         ENTRY;
1378
1379         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1380                inode->i_generation, inode);
1381
1382         ll_lli_init(lli);
1383
1384         LASSERT(!lli->lli_smd);
1385
1386         /* Core attributes from the MDS first.  This is a new inode, and
1387          * the VFS doesn't zero times in the core inode so we have to do
1388          * it ourselves.  They will be overwritten by either MDS or OST
1389          * attributes - we just need to make sure they aren't newer. */
1390         LTIME_S(inode->i_mtime) = 0;
1391         LTIME_S(inode->i_atime) = 0;
1392         LTIME_S(inode->i_ctime) = 0;
1393
1394         inode->i_rdev = 0;
1395         ll_update_inode(inode, md);
1396
1397         /* OIDEBUG(inode); */
1398
1399         if (S_ISREG(inode->i_mode)) {
1400                 inode->i_op = &ll_file_inode_operations;
1401                 inode->i_fop = &ll_file_operations;
1402                 inode->i_mapping->a_ops = &ll_aops;
1403                 EXIT;
1404         } else if (S_ISDIR(inode->i_mode)) {
1405                 inode->i_op = &ll_dir_inode_operations;
1406                 inode->i_fop = &ll_dir_operations;
1407                 inode->i_mapping->a_ops = &ll_dir_aops;
1408                 EXIT;
1409         } else if (S_ISLNK(inode->i_mode)) {
1410                 inode->i_op = &ll_fast_symlink_inode_operations;
1411                 EXIT;
1412         } else {
1413                 inode->i_op = &ll_special_inode_operations;
1414
1415 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1416                 init_special_inode(inode, inode->i_mode,
1417                                    kdev_t_to_nr(inode->i_rdev));
1418
1419                 /* initializing backing dev info. */
1420                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1421 #else
1422                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1423 #endif
1424                 lli->ll_save_ifop = inode->i_fop;
1425
1426                 if (S_ISCHR(inode->i_mode))
1427                         inode->i_fop = &ll_special_chr_inode_fops;
1428                 else if (S_ISBLK(inode->i_mode))
1429                         inode->i_fop = &ll_special_blk_inode_fops;
1430                 else if (S_ISFIFO(inode->i_mode))
1431                         inode->i_fop = &ll_special_fifo_inode_fops;
1432                 else if (S_ISSOCK(inode->i_mode))
1433                         inode->i_fop = &ll_special_sock_inode_fops;
1434
1435                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1436                       inode->i_fop);
1437
1438                 if (lli->ll_save_ifop->owner) {
1439                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1440                               lli->ll_save_ifop->owner);
1441                 }
1442                 EXIT;
1443         }
1444 }
1445
1446 void ll_delete_inode(struct inode *inode)
1447 {
1448         int rc;
1449         struct ll_fid fid;
1450         struct ll_sb_info *sbi = ll_i2sbi(inode);
1451         ENTRY;
1452
1453         ll_inode2fid(&fid, inode);
1454
1455         rc = md_delete_object(sbi->ll_mdc_exp, &fid);
1456         if (rc) {
1457                 CERROR("md_delete_object() failed, error %d.\n",
1458                        rc);
1459         }
1460
1461         clear_inode(inode);
1462         EXIT;
1463 }
1464
1465 int ll_iocontrol(struct inode *inode, struct file *file,
1466                  unsigned int cmd, unsigned long arg)
1467 {
1468         struct ll_sb_info *sbi = ll_i2sbi(inode);
1469         struct ptlrpc_request *req = NULL;
1470         int rc, flags = 0;
1471         ENTRY;
1472
1473         switch(cmd) {
1474         case EXT3_IOC_GETFLAGS: {
1475                 struct ll_fid fid;
1476                 unsigned long valid = OBD_MD_FLFLAGS;
1477                 struct mds_body *body;
1478
1479                 ll_inode2fid(&fid, inode);
1480                 rc = md_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1481                 if (rc) {
1482                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1483                         RETURN(-abs(rc));
1484                 }
1485
1486                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1487
1488                 if (body->flags & S_APPEND)
1489                         flags |= EXT3_APPEND_FL;
1490                 if (body->flags & S_IMMUTABLE)
1491                         flags |= EXT3_IMMUTABLE_FL;
1492                 if (body->flags & S_NOATIME)
1493                         flags |= EXT3_NOATIME_FL;
1494
1495                 ptlrpc_req_finished (req);
1496
1497                 RETURN(put_user(flags, (int *)arg));
1498         }
1499         case EXT3_IOC_SETFLAGS: {
1500                 struct mdc_op_data op_data;
1501                 struct iattr attr;
1502                 struct obdo *oa;
1503                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1504
1505                 if (get_user(flags, (int *)arg))
1506                         RETURN(-EFAULT);
1507
1508                 oa = obdo_alloc();
1509                 if (!oa)
1510                         RETURN(-ENOMEM);
1511
1512                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1513
1514                 memset(&attr, 0x0, sizeof(attr));
1515                 attr.ia_attr_flags = flags;
1516                 attr.ia_valid |= ATTR_ATTR_FLAG;
1517
1518                 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
1519                                 &attr, NULL, 0, NULL, 0, &req);
1520                 if (rc) {
1521                         ptlrpc_req_finished(req);
1522                         if (rc != -EPERM && rc != -EACCES)
1523                                 CERROR("md_setattr fails: rc = %d\n", rc);
1524                         obdo_free(oa);
1525                         RETURN(rc);
1526                 }
1527                 ptlrpc_req_finished(req);
1528
1529                 oa->o_id = lsm->lsm_object_id;
1530                 oa->o_gr = lsm->lsm_object_gr;
1531                 oa->o_flags = flags;
1532                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1533
1534                 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1535                 obdo_free(oa);
1536                 if (rc) {
1537                         if (rc != -EPERM && rc != -EACCES)
1538                                 CERROR("md_setattr fails: rc = %d\n", rc);
1539                         RETURN(rc);
1540                 }
1541
1542                 if (flags & EXT3_APPEND_FL)
1543                         inode->i_flags |= S_APPEND;
1544                 else
1545                         inode->i_flags &= ~S_APPEND;
1546                 if (flags & EXT3_IMMUTABLE_FL)
1547                         inode->i_flags |= S_IMMUTABLE;
1548                 else
1549                         inode->i_flags &= ~S_IMMUTABLE;
1550                 if (flags & EXT3_NOATIME_FL)
1551                         inode->i_flags |= S_NOATIME;
1552                 else
1553                         inode->i_flags &= ~S_NOATIME;
1554
1555                 RETURN(0);
1556         }
1557         default:
1558                 RETURN(-ENOSYS);
1559         }
1560
1561         RETURN(0);
1562 }
1563
1564 void ll_umount_begin(struct super_block *sb)
1565 {
1566         struct ll_sb_info *sbi = ll_s2sbi(sb);
1567         struct obd_device *obd;
1568         struct obd_ioctl_data ioc_data = { 0 };
1569         ENTRY;
1570         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1571
1572         obd = class_exp2obd(sbi->ll_mdc_exp);
1573         if (obd == NULL) {
1574                 CERROR("Invalid MDC connection handle "LPX64"\n",
1575                        sbi->ll_mdc_exp->exp_handle.h_cookie);
1576                 EXIT;
1577                 return;
1578         }
1579         obd->obd_no_recov = 1;
1580         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1581                       &ioc_data, NULL);
1582
1583         obd = class_exp2obd(sbi->ll_osc_exp);
1584         if (obd == NULL) {
1585                 CERROR("Invalid LOV connection handle "LPX64"\n",
1586                        sbi->ll_osc_exp->exp_handle.h_cookie);
1587                 EXIT;
1588                 return;
1589         }
1590
1591         obd->obd_no_recov = 1;
1592         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1593                       &ioc_data, NULL);
1594
1595         /* Really, we'd like to wait until there are no requests outstanding,
1596          * and then continue.  For now, we just invalidate the requests,
1597          * schedule, and hope.
1598          */
1599         schedule();
1600
1601         EXIT;
1602 }
1603
1604 int ll_prep_inode(struct obd_export *osc_exp, struct obd_export *mdc_exp,
1605                   struct inode **inode, struct ptlrpc_request *req,
1606                   int offset, struct super_block *sb)
1607 {
1608         struct lustre_md md;
1609         int rc = 0;
1610
1611         rc = mdc_req2lustre_md(mdc_exp, req, offset, osc_exp, &md);
1612         if (rc)
1613                 RETURN(rc);
1614
1615         if (*inode) {
1616                 ll_update_inode(*inode, &md);
1617         } else {
1618                 LASSERT(sb);
1619                 *inode = ll_iget(sb, md.body->ino, &md);
1620                 if (*inode == NULL || is_bad_inode(*inode)) {
1621                         /* free the lsm if we allocated one above */
1622                         if (md.lsm != NULL)
1623                                 obd_free_memmd(osc_exp, &md.lsm);
1624                         if (md.mea != NULL)
1625                                 obd_free_memmd(mdc_exp,
1626                                                (struct lov_stripe_md**)&md.mea);
1627                         rc = -ENOMEM;
1628                         CERROR("new_inode -fatal: rc %d\n", rc);
1629                 }
1630         }
1631
1632         RETURN(rc);
1633 }