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 == 0) {
 103                 obd_init_ea_size(sbi->ll_mdc_exp,
 104                                  obd_size_diskmd(sbi->ll_osc_exp, NULL),
 105                                  desc.ld_tgt_count*sizeof(struct llog_cookie));
 106
 107                 /* declare ourself as real client. not connection
 108                  * from another MDS
 109                  * FIXME: remove fake valsize, mdsize --bzzz */
 110                 rc = obd_set_info(sbi->ll_mdc_exp, strlen("client"),
 111                                   "client", valsize, &desc);
 112         }
 113         RETURN(rc);
 114 }
 115 #if CONFIG_SNAPFS
 116 int lustre_set_clone_info(struct super_block *sb, int clone_index)
 117 {
 118         struct ll_sb_info *sbi = ll_s2sbi(sb);
 119
 120         ENTRY;
 121
 122         CDEBUG(D_INFO, "set clone index %d\n", clone_index);
 123         if (!clone_index)
 124                 RETURN(0);
 125
 126         if (sbi->ll_mdc_exp) {
 127                 struct obd_import *climp = class_exp2cliimp(sbi->ll_mdc_exp);
 128                 struct client_obd *cl_obd = &climp->imp_obd->u.cli;
 129
 130                 OBD_ALLOC(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 131                 if (!cl_obd->cl_clone_info)
 132                         RETURN(-ENOMEM);
 133                 SET_CLONE_INDEX(cl_obd->cl_clone_info, clone_index);
 134                 SET_CLONE_FLAGS(cl_obd->cl_clone_info, SM_CLONE_FS);
 135         }
 136
 137         if (sbi->ll_osc_exp) {
 138                 struct obd_import *climp = class_exp2cliimp(sbi->ll_osc_exp);
 139                 struct client_obd *cl_obd = &climp->imp_obd->u.cli;
 140
 141                 OBD_ALLOC(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 142                 if (!cl_obd->cl_clone_info)
 143                         RETURN(-ENOMEM);
 144                 SET_CLONE_INDEX(cl_obd->cl_clone_info, clone_index);
 145                 SET_CLONE_FLAGS(cl_obd->cl_clone_info, SM_CLONE_FS);
 146         }
 147         RETURN(0);
 148 }
 149
 150 int lustre_cleanup_clone_info(struct super_block *sb)
 151 {
 152         struct ll_sb_info *sbi = ll_s2sbi(sb);
 153
 154         ENTRY;
 155
 156         if (sbi->ll_mdc_exp) {
 157                 struct obd_import *climp = class_exp2cliimp(sbi->ll_mdc_exp);
 158                 struct client_obd *cl_obd = &climp->imp_obd->u.cli;
 159
 160                 if (!cl_obd->cl_clone_info)
 161                         RETURN(0);
 162                 CDEBUG(D_INFO, "clean clone info %p\n", cl_obd->cl_clone_info);
 163                 OBD_FREE(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 164         }
 165
 166         if (sbi->ll_osc_exp) {
 167                 struct obd_import *climp = class_exp2cliimp(sbi->ll_osc_exp);
 168                 struct client_obd *cl_obd = &climp->imp_obd->u.cli;
 169
 170                 if (!cl_obd->cl_clone_info)
 171                         RETURN(0);
 172                 CDEBUG(D_INFO, "clean clone info %p\n", cl_obd->cl_clone_info);
 173                 OBD_FREE(cl_obd->cl_clone_info, sizeof(struct clonefs_info));
 174         }
 175         RETURN(0);
 176 }
 177 #endif
 178
 179
 180 int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
 181 {
 182         struct inode *root = 0;
 183         struct ll_sb_info *sbi = ll_s2sbi(sb);
 184         struct obd_device *obd;
 185         struct ll_fid rootfid;
 186         struct obd_statfs osfs;
 187         struct ptlrpc_request *request = NULL;
 188         struct lustre_handle osc_conn = {0, };
 189         struct lustre_handle mdc_conn = {0, };
 190         struct lustre_md md;
 191         kdev_t devno;
 192         int err;
 193
 194         obd = class_name2obd(mdc);
 195         if (!obd) {
 196                 CERROR("MDC %s: not setup or attached\n", mdc);
 197                 RETURN(-EINVAL);
 198         }
 199
 200         if (proc_lustre_fs_root) {
 201                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 202                                                   osc, mdc);
 203                 if (err < 0)
 204                         CERROR("could not register mount in /proc/lustre");
 205         }
 206
 207         err = obd_connect(&mdc_conn, obd, &sbi->ll_sb_uuid);
 208         if (err == -EBUSY) {
 209                 CERROR("An MDS (mdc %s) is performing recovery, of which this"
 210                        " client is not a part.  Please wait for recovery to "
 211                        "complete, abort, or time out.\n", mdc);
 212                 GOTO(out, err);
 213         } else if (err) {
 214                 CERROR("cannot connect to %s: rc = %d\n", mdc, err);
 215                 GOTO(out, err);
 216         }
 217         sbi->ll_mdc_exp = class_conn2export(&mdc_conn);
 218
 219         err = obd_statfs(obd, &osfs, jiffies - HZ);
 220         if (err)
 221                 GOTO(out_mdc, err);
 222
 223         LASSERT(osfs.os_bsize);
 224         sb->s_blocksize = osfs.os_bsize;
 225         sb->s_blocksize_bits = log2(osfs.os_bsize);
 226         sb->s_magic = LL_SUPER_MAGIC;
 227         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 228
 229         devno = get_uuid2int(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid,
 230                              strlen(sbi->ll_mdc_exp->exp_obd->obd_uuid.uuid));
 231
 232         sb->s_dev = devno;
 233
 234         obd = class_name2obd(osc);
 235         if (!obd) {
 236                 CERROR("OSC %s: not setup or attached\n", osc);
 237                 GOTO(out_mdc, err);
 238         }
 239
 240         err = obd_connect(&osc_conn, obd, &sbi->ll_sb_uuid);
 241         if (err == -EBUSY) {
 242                 CERROR("An OST (osc %s) is performing recovery, of which this"
 243                        " client is not a part.  Please wait for recovery to "
 244                        "complete, abort, or time out.\n", osc);
 245                 GOTO(out, err);
 246         } else if (err) {
 247                 CERROR("cannot connect to %s: rc = %d\n", osc, err);
 248                 GOTO(out_mdc, err);
 249         }
 250         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 251
 252         lustre_init_ea_size(sbi);
 253
 254         err = md_getstatus(sbi->ll_mdc_exp, &rootfid);
 255         if (err) {
 256                 CERROR("cannot mds_connect: rc = %d\n", err);
 257                 GOTO(out_osc, err);
 258         }
 259         CDEBUG(D_SUPER, "rootfid "LPU64"\n", rootfid.id);
 260         sbi->ll_rootino = rootfid.id;
 261
 262         sb->s_op = &lustre_super_operations;
 263
 264         /* make root inode
 265          * XXX: move this to after cbd setup? */
 266         err = md_getattr(sbi->ll_mdc_exp, &rootfid,
 267                          OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS, 0, &request);
 268         if (err) {
 269                 CERROR("md_getattr failed for root: rc = %d\n", err);
 270                 GOTO(out_osc, err);
 271         }
 272
 273         err = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
 274                                 sbi->ll_osc_exp, &md);
 275         if (err) {
 276                 CERROR("failed to understand root inode md: rc = %d\n",err);
 277                 ptlrpc_req_finished (request);
 278                 GOTO(out_osc, err);
 279         }
 280
 281         LASSERT(sbi->ll_rootino != 0);
 282         root = ll_iget(sb, sbi->ll_rootino, &md);
 283
 284         ptlrpc_req_finished(request);
 285
 286         if (root == NULL || is_bad_inode(root)) {
 287                 /* XXX might need iput() for bad inode */
 288                 CERROR("lustre_lite: bad iget4 for root\n");
 289                 GOTO(out_root, err = -EBADF);
 290         }
 291
 292         err = ll_close_thread_start(&sbi->ll_lcq);
 293         if (err) {
 294                 CERROR("cannot start close thread: rc %d\n", err);
 295                 GOTO(out_root, err);
 296         }
 297
 298         ll_gns_add_timer(sbi);
 299
 300         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 301            backing dev info assigned to inode mapping is used for
 302            determining maximal readahead. */
 303 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 304         /* bug 2805 - set VM readahead to zero */
 305         vm_max_readahead = vm_min_readahead = 0;
 306 #endif
 307
 308         sb->s_root = d_alloc_root(root);
 309
 310 #ifdef S_PDIROPS
 311         CWARN("Enabling PDIROPS\n");
 312         sb->s_flags |= S_PDIROPS;
 313 #endif
 314
 315         RETURN(err);
 316
 317 out_root:
 318         if (root)
 319                 iput(root);
 320 out_osc:
 321         obd_disconnect(sbi->ll_osc_exp, 0);
 322 out_mdc:
 323         obd_disconnect(sbi->ll_mdc_exp, 0);
 324 out:
 325         lprocfs_unregister_mountpoint(sbi);
 326         RETURN(err);
 327 }
 328
 329 void lustre_common_put_super(struct super_block *sb)
 330 {
 331         struct ll_sb_info *sbi = ll_s2sbi(sb);
 332         struct hlist_node *tmp, *next;
 333         ENTRY;
 334
 335         ll_gns_del_timer(sbi);
 336
 337         ll_close_thread_shutdown(sbi->ll_lcq);
 338
 339         list_del(&sbi->ll_conn_chain);
 340         obd_disconnect(sbi->ll_osc_exp, 0);
 341
 342         lprocfs_unregister_mountpoint(sbi);
 343         if (sbi->ll_proc_root) {
 344                 lprocfs_remove(sbi->ll_proc_root);
 345                 sbi->ll_proc_root = NULL;
 346         }
 347
 348         obd_disconnect(sbi->ll_mdc_exp, 0);
 349
 350         // We do this to get rid of orphaned dentries. That is not really trw.
 351         spin_lock(&dcache_lock);
 352         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 353                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 354                 CWARN("orphan dentry %*s (%p) at unmount\n",
 355                       dentry->d_name.len, dentry->d_name.name, dentry);
 356                 shrink_dcache_parent(dentry);
 357         }
 358         spin_unlock(&dcache_lock);
 359         EXIT;
 360 }
 361
 362
 363 char *ll_read_opt(const char *opt, char *data)
 364 {
 365         char *value;
 366         char *retval;
 367         ENTRY;
 368
 369         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 370         if (strncmp(opt, data, strlen(opt)))
 371                 RETURN(NULL);
 372         if ((value = strchr(data, '=')) == NULL)
 373                 RETURN(NULL);
 374
 375         value++;
 376         OBD_ALLOC(retval, strlen(value) + 1);
 377         if (!retval) {
 378                 CERROR("out of memory!\n");
 379                 RETURN(NULL);
 380         }
 381
 382         memcpy(retval, value, strlen(value)+1);
 383         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 384         RETURN(retval);
 385 }
 386
 387 int ll_set_opt(const char *opt, char *data, int fl)
 388 {
 389         ENTRY;
 390
 391         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 392         if (strncmp(opt, data, strlen(opt)))
 393                 RETURN(0);
 394         else
 395                 RETURN(fl);
 396 }
 397
 398 void ll_options(char *options, char **ost, char **mdc, int *flags,
 399                 char **clone_opts)
 400 {
 401         char *this_char;
 402 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 403         char *opt_ptr = options;
 404 #endif
 405         ENTRY;
 406
 407         if (!options) {
 408                 EXIT;
 409                 return;
 410         }
 411
 412 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 413         for (this_char = strtok (options, ",");
 414              this_char != NULL;
 415              this_char = strtok (NULL, ",")) {
 416 #else
 417         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 418 #endif
 419                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 420                 if (!*ost && (*ost = ll_read_opt("osc", this_char)))
 421                         continue;
 422                 if (!*mdc && (*mdc = ll_read_opt("mdc", this_char)))
 423                         continue;
 424                 if (!*clone_opts && (*clone_opts = ll_read_opt("clone",
 425                                                                 this_char)))
 426                         continue;
 427                 if (!(*flags & LL_SBI_NOLCK) &&
 428                     ((*flags) = (*flags) |
 429                                 ll_set_opt("nolock", this_char,
 430                                            LL_SBI_NOLCK)))
 431                         continue;
 432         }
 433
 434         EXIT;
 435 }
 436
 437 void ll_lli_init(struct ll_inode_info *lli)
 438 {
 439         sema_init(&lli->lli_open_sem, 1);
 440         lli->lli_flags = 0;
 441         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 442         spin_lock_init(&lli->lli_lock);
 443         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 444 }
 445
 446 int ll_fill_super(struct super_block *sb, void *data, int silent)
 447 {
 448         struct ll_sb_info *sbi;
 449         char *osc = NULL;
 450         char *mdc = NULL;
 451         char *clone_opts = NULL;
 452         int err;
 453         ENTRY;
 454
 455         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 456
 457         sbi = lustre_init_sbi(sb);
 458         if (!sbi)
 459                 RETURN(-ENOMEM);
 460
 461         sbi->ll_flags |= LL_SBI_READAHEAD;
 462         ll_options(data, &osc, &mdc, &sbi->ll_flags, &clone_opts);
 463
 464         if (!osc) {
 465                 CERROR("no osc\n");
 466                 GOTO(out, err = -EINVAL);
 467         }
 468
 469         if (!mdc) {
 470                 CERROR("no mdc\n");
 471                 GOTO(out, err = -EINVAL);
 472         }
 473
 474         err = lustre_common_fill_super(sb, mdc, osc);
 475 #if CONFIG_SNAPFS
 476         if (clone_opts) {
 477                 int clone_index = 0;
 478                 char *endp;
 479
 480                 /*set clone info to the super block*/
 481                 clone_index = simple_strtoul(clone_opts, &endp, 0);
 482                 err = lustre_set_clone_info(sb, clone_index);
 483         }
 484 #endif
 485 out:
 486         if (err)
 487                 lustre_free_sbi(sb);
 488
 489         if (mdc)
 490                 OBD_FREE(mdc, strlen(mdc) + 1);
 491         if (osc)
 492                 OBD_FREE(osc, strlen(osc) + 1);
 493         if (clone_opts)
 494                 OBD_FREE(clone_opts, strlen(clone_opts) + 1);
 495
 496         RETURN(err);
 497 } /* ll_read_super */
 498
 499 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 500                               struct config_llog_instance *cfg, int allow_recov)
 501 {
 502         struct lustre_cfg lcfg;
 503         struct portals_cfg pcfg;
 504         char * peer = "MDS_PEER_UUID";
 505         struct obd_device *obd;
 506         struct lustre_handle mdc_conn = {0, };
 507         struct obd_export *exp;
 508         char * name = "mdc_dev";
 509         class_uuid_t uuid;
 510         struct obd_uuid mdc_uuid;
 511         struct llog_ctxt *ctxt;
 512         int rc = 0;
 513         int err;
 514         ENTRY;
 515
 516         if (lmd_bad_magic(lmd))
 517                 RETURN(-EINVAL);
 518
 519         generate_random_uuid(uuid);
 520         class_uuid_unparse(uuid, &mdc_uuid);
 521
 522         if (lmd->lmd_local_nid) {
 523                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 524                 pcfg.pcfg_nal = lmd->lmd_nal;
 525                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 526                 err = libcfs_nal_cmd(&pcfg);
 527                 if (err <0)
 528                         GOTO(out, err);
 529         }
 530
 531         if (lmd->lmd_nal == SOCKNAL) {
 532                 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
 533                 pcfg.pcfg_nal     = lmd->lmd_nal;
 534                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 535                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 536                 pcfg.pcfg_misc    = lmd->lmd_port;
 537                 pcfg.pcfg_size    = 8388608;
 538                 pcfg.pcfg_flags   = 0x4; /*share*/
 539                 err = libcfs_nal_cmd(&pcfg);
 540                 if (err <0)
 541                         GOTO(out, err);
 542         }
 543
 544         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
 545         lcfg.lcfg_nid = lmd->lmd_server_nid;
 546         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 547         lcfg.lcfg_inlbuf1 = peer;
 548         lcfg.lcfg_nal = lmd->lmd_nal;
 549         err = class_process_config(&lcfg);
 550         if (err < 0)
 551                 GOTO(out_del_conn, err);
 552
 553         LCFG_INIT(lcfg, LCFG_ATTACH, name);
 554         lcfg.lcfg_inlbuf1 = "mdc";
 555         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 556         lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
 557         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 558         err = class_process_config(&lcfg);
 559         if (err < 0)
 560                 GOTO(out_del_uuid, err);
 561
 562         LCFG_INIT(lcfg, LCFG_SETUP, name);
 563         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
 564         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 565         lcfg.lcfg_inlbuf2 = peer;
 566         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 567         err = class_process_config(&lcfg);
 568         if (err < 0)
 569                 GOTO(out_detach, err);
 570
 571         obd = class_name2obd(name);
 572         if (obd == NULL)
 573                 GOTO(out_cleanup, err = -EINVAL);
 574
 575         /* Disable initial recovery on this import */
 576         err = obd_set_info(obd->obd_self_export,
 577                            strlen("initial_recov"), "initial_recov",
 578                            sizeof(allow_recov), &allow_recov);
 579         if (err)
 580                 GOTO(out_cleanup, err);
 581
 582         err = obd_connect(&mdc_conn, obd, &mdc_uuid);
 583         if (err) {
 584                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
 585                 GOTO(out_cleanup, err);
 586         }
 587
 588         exp = class_conn2export(&mdc_conn);
 589
 590         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 591         rc = class_config_process_llog(ctxt, profile, cfg);
 592         if (rc)
 593                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 594
 595         err = obd_disconnect(exp, 0);
 596
 597 out_cleanup:
 598         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
 599         err = class_process_config(&lcfg);
 600         if (err < 0)
 601                 GOTO(out, err);
 602
 603 out_detach:
 604         LCFG_INIT(lcfg, LCFG_DETACH, name);
 605         err = class_process_config(&lcfg);
 606         if (err < 0)
 607                 GOTO(out, err);
 608
 609 out_del_uuid:
 610         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
 611         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 612         lcfg.lcfg_inlbuf1 = peer;
 613         err = class_process_config(&lcfg);
 614
 615 out_del_conn:
 616         if (lmd->lmd_nal == SOCKNAL) {
 617                 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
 618                 pcfg.pcfg_nal     = lmd->lmd_nal;
 619                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 620                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 621                 pcfg.pcfg_flags   = 1; /*share*/
 622                 err = libcfs_nal_cmd(&pcfg);
 623                 if (err <0)
 624                         GOTO(out, err);
 625         }
 626 out:
 627         if (rc == 0)
 628                 rc = err;
 629
 630         RETURN(rc);
 631 }
 632
 633 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 634 {
 635         struct lustre_mount_data * lmd = data;
 636         struct ll_sb_info *sbi;
 637         char *osc = NULL;
 638         char *mdc = NULL;
 639         int err;
 640         ENTRY;
 641
 642         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 643         if (lmd_bad_magic(lmd))
 644                 RETURN(-EINVAL);
 645
 646         sbi = lustre_init_sbi(sb);
 647         if (!sbi)
 648                 RETURN(-ENOMEM);
 649
 650         sbi->ll_flags |= LL_SBI_READAHEAD;
 651
 652         if (lmd->lmd_profile) {
 653                 struct lustre_profile *lprof;
 654                 struct config_llog_instance cfg;
 655                 int len;
 656
 657                 if (lmd->lmd_mds[0] == '\0') {
 658                         CERROR("no mds name\n");
 659                         GOTO(out_free, err = -EINVAL);
 660                 }
 661
 662                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 663                 if (sbi->ll_lmd == NULL)
 664                         GOTO(out_free, err = -ENOMEM);
 665                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 666
 667                 /* generate a string unique to this super, let's try
 668                  the address of the super itself.*/
 669                 len = (sizeof(sb) * 2) + 1;
 670                 OBD_ALLOC(sbi->ll_instance, len);
 671                 if (sbi->ll_instance == NULL)
 672                         GOTO(out_free, err = -ENOMEM);
 673                 sprintf(sbi->ll_instance, "%p", sb);
 674
 675                 cfg.cfg_instance = sbi->ll_instance;
 676                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 677                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 678                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 679                 if (err < 0) {
 680                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 681                         GOTO(out_free, err);
 682                 }
 683
 684                 lprof = class_get_profile(lmd->lmd_profile);
 685                 if (lprof == NULL) {
 686                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 687                         GOTO(out_free, err = -EINVAL);
 688                 }
 689                 if (osc)
 690                         OBD_FREE(osc, strlen(osc) + 1);
 691                 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
 692                           strlen(sbi->ll_instance) + 2);
 693                 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
 694
 695                 if (mdc)
 696                         OBD_FREE(mdc, strlen(mdc) + 1);
 697                 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
 698                           strlen(sbi->ll_instance) + 2);
 699                 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
 700         }
 701
 702         if (!osc) {
 703                 CERROR("no osc\n");
 704                 GOTO(out_free, err = -EINVAL);
 705         }
 706
 707         if (!mdc) {
 708                 CERROR("no mdc\n");
 709                 GOTO(out_free, err = -EINVAL);
 710         }
 711
 712         err = lustre_common_fill_super(sb, mdc, osc);
 713
 714         if (err)
 715                 GOTO(out_free, err);
 716
 717 #if CONFIG_SNAPFS
 718         if (lmd->lmd_clone_index) {
 719                 err = lustre_set_clone_info(sb, lmd->lmd_clone_index);
 720         }
 721
 722         if (err)
 723                 GOTO(out_free, err);
 724 #endif
 725
 726 out_dev:
 727         if (mdc)
 728                 OBD_FREE(mdc, strlen(mdc) + 1);
 729         if (osc)
 730                 OBD_FREE(osc, strlen(osc) + 1);
 731
 732         RETURN(err);
 733
 734 out_free:
 735         if (sbi->ll_lmd) {
 736                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 737                 int err;
 738
 739                 if (sbi->ll_instance != NULL) {
 740                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 741                         char * cln_prof;
 742                         struct config_llog_instance cfg;
 743
 744                         cfg.cfg_instance = sbi->ll_instance;
 745                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 746
 747                         OBD_ALLOC(cln_prof, len);
 748                         sprintf(cln_prof, "%s-clean", lmd->lmd_profile);
 749
 750                         err = lustre_process_log(lmd, cln_prof, &cfg, 0);
 751                         if (err < 0)
 752                                 CERROR("Unable to process log: %s\n", cln_prof);
 753                         OBD_FREE(cln_prof, len);
 754                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
 755                 }
 756                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 757         }
 758 #if CONFIG_SNAPFS
 759         lustre_cleanup_clone_info(sb);
 760 #endif
 761         lustre_free_sbi(sb);
 762
 763         goto out_dev;
 764 } /* lustre_fill_super */
 765
 766 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 767 {
 768         struct lustre_cfg lcfg;
 769         struct obd_device *obd;
 770         int next = 0;
 771
 772         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 773         {
 774                 int err;
 775
 776                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
 777                 err = class_process_config(&lcfg);
 778                 if (err) {
 779                         CERROR("cleanup failed: %s\n", obd->obd_name);
 780                         //continue;
 781                 }
 782
 783                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
 784                 err = class_process_config(&lcfg);
 785                 if (err) {
 786                         CERROR("detach failed: %s\n", obd->obd_name);
 787                         //continue;
 788                 }
 789         }
 790
 791         if (sbi->ll_lmd != NULL)
 792                 class_del_profile(sbi->ll_lmd->lmd_profile);
 793 }
 794
 795 void lustre_put_super(struct super_block *sb)
 796 {
 797         struct obd_device *obd;
 798         struct ll_sb_info *sbi = ll_s2sbi(sb);
 799         int force_umount = 0;
 800         ENTRY;
 801
 802         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 803         obd = class_exp2obd(sbi->ll_mdc_exp);
 804         if (obd)
 805                 force_umount = obd->obd_no_recov;
 806         obd = NULL;
 807
 808 #if CONFIG_SNAPFS
 809         lustre_cleanup_clone_info(sb);
 810 #endif
 811
 812         lustre_common_put_super(sb);
 813         if (sbi->ll_lmd != NULL) {
 814                 char * cln_prof;
 815                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 816                 int err;
 817                 struct config_llog_instance cfg;
 818
 819                 if (force_umount) {
 820                         CERROR("force umount, doing manual cleanup\n");
 821                         lustre_manual_cleanup(sbi);
 822                         GOTO(free_lmd, 0);
 823                 }
 824
 825                 cfg.cfg_instance = sbi->ll_instance;
 826                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 827
 828                 OBD_ALLOC(cln_prof, len);
 829                 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 830
 831                 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
 832                 if (err < 0) {
 833                         CERROR("Unable to process log: %s, doing manual cleanup"
 834                                "\n", cln_prof);
 835                         lustre_manual_cleanup(sbi);
 836                 }
 837
 838                 OBD_FREE(cln_prof, len);
 839         free_lmd:
 840                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 841                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 842         }
 843
 844         lustre_free_sbi(sb);
 845
 846         EXIT;
 847 } /* lustre_put_super */
 848
 849 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 850 {
 851         struct obd_export *mdc_exp = sbi->ll_mdc_exp;
 852         struct lustre_mount_data *lmd = sbi->ll_lmd;
 853         struct llog_ctxt *ctxt;
 854         struct config_llog_instance cfg;
 855         char *profile = lmd->lmd_profile, *name = NULL;
 856         int rc, namelen =  0, version;
 857         ENTRY;
 858
 859         if (profile == NULL)
 860                 RETURN(0);
 861         if (lmd == NULL) {
 862                 CERROR("Client not mounted with zero-conf; cannot process "
 863                        "update log.\n");
 864                 RETURN(0);
 865         }
 866
 867         rc = ldlm_cli_cancel_unused(mdc_exp->exp_obd->obd_namespace, NULL,
 868                                     LDLM_FL_CONFIG_CHANGE, NULL);
 869         if (rc != 0)
 870                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
 871
 872         rc = obd_cancel_unused(sbi->ll_osc_exp, NULL, LDLM_FL_CONFIG_CHANGE,
 873                                NULL);
 874         if (rc != 0)
 875                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 876
 877         cfg.cfg_instance = sbi->ll_instance;
 878         cfg.cfg_uuid = sbi->ll_sb_uuid;
 879         cfg.cfg_local_nid = lmd->lmd_local_nid;
 880
 881         namelen = strlen(profile) + 20; /* -clean-######### */
 882         OBD_ALLOC(name, namelen);
 883         if (name == NULL)
 884                 RETURN(-ENOMEM);
 885
 886         if (clean) {
 887                 version = sbi->ll_config_version - 1;
 888                 sprintf(name, "%s-clean-%d", profile, version);
 889         } else {
 890                 version = sbi->ll_config_version + 1;
 891                 sprintf(name, "%s-%d", profile, version);
 892         }
 893
 894         CWARN("Applying configuration log %s\n", name);
 895
 896         ctxt = llog_get_context(&mdc_exp->exp_obd->obd_llogs,
 897                                 LLOG_CONFIG_REPL_CTXT);
 898         rc = class_config_process_llog(ctxt, name, &cfg);
 899         if (rc == 0)
 900                 sbi->ll_config_version = version;
 901         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 902
 903         if (rc == 0 && clean == 0) {
 904                 struct lov_desc desc;
 905                 int rc, valsize;
 906                 valsize = sizeof(desc);
 907                 rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1,
 908                                   "lovdesc", &valsize, &desc);
 909
 910                 rc = obd_init_ea_size(mdc_exp,
 911                                       obd_size_diskmd(sbi->ll_osc_exp, NULL),
 912                                       (desc.ld_tgt_count *
 913                                        sizeof(struct llog_cookie)));
 914         }
 915         OBD_FREE(name, namelen);
 916         RETURN(rc);
 917 }
 918
 919 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 920 {
 921         struct inode *inode;
 922         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 923         if (lock->l_ast_data)
 924                 inode = igrab(lock->l_ast_data);
 925         else
 926                 inode = NULL;
 927         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 928         return inode;
 929 }
 930
 931 int null_if_equal(struct ldlm_lock *lock, void *data)
 932 {
 933         if (data == lock->l_ast_data)
 934                 lock->l_ast_data = NULL;
 935
 936         if (lock->l_req_mode != lock->l_granted_mode)
 937                 return LDLM_ITER_STOP;
 938
 939         return LDLM_ITER_CONTINUE;
 940 }
 941
 942 void ll_clear_inode(struct inode *inode)
 943 {
 944         struct ll_fid fid;
 945         struct ll_inode_info *lli = ll_i2info(inode);
 946         struct ll_sb_info *sbi = ll_i2sbi(inode);
 947         ENTRY;
 948
 949         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 950                inode->i_generation, inode);
 951
 952         ll_inode2fid(&fid, inode);
 953         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 954         md_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
 955
 956         if (lli->lli_smd)
 957                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
 958                                   null_if_equal, inode);
 959
 960         if (lli->lli_smd) {
 961                 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
 962                 lli->lli_smd = NULL;
 963         }
 964
 965         if (lli->lli_mea) {
 966                 /* FIXME: change cbdata for mea here */
 967                 obd_free_memmd(sbi->ll_mdc_exp,
 968                                (struct lov_stripe_md **) &lli->lli_mea);
 969                 lli->lli_mea = NULL;
 970         }
 971
 972         if (lli->lli_symlink_name) {
 973                 OBD_FREE(lli->lli_symlink_name,
 974                          strlen(lli->lli_symlink_name) + 1);
 975                 lli->lli_symlink_name = NULL;
 976         }
 977
 978         EXIT;
 979 }
 980
 981 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 982  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 983  * keep these values until such a time that objects are allocated for it.
 984  * We do the MDS operations first, as it is checking permissions for us.
 985  * We don't to the MDS RPC if there is nothing that we want to store there,
 986  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 987  * going to do an RPC anyways.
 988  *
 989  * If we are doing a truncate, we will send the mtime and ctime updates
 990  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 991  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 992  * at the same time.
 993  */
 994 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
 995 {
 996         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 997         struct ll_sb_info *sbi = ll_i2sbi(inode);
 998         struct ptlrpc_request *request = NULL;
 999         struct mdc_op_data op_data;
1000         int ia_valid = attr->ia_valid;
1001         int rc = 0;
1002         ENTRY;
1003
1004         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1005         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1006
1007         if (ia_valid & ATTR_SIZE) {
1008                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1009                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1010                                attr->ia_size, ll_file_maxbytes(inode));
1011                         RETURN(-EFBIG);
1012                 }
1013
1014                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1015         }
1016
1017         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1018         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1019                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1020                         RETURN(-EPERM);
1021         }
1022
1023         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1024         if (attr->ia_valid & ATTR_CTIME) {
1025                 attr->ia_ctime = CURRENT_TIME;
1026                 attr->ia_valid |= ATTR_CTIME_SET;
1027         }
1028         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1029                 attr->ia_atime = CURRENT_TIME;
1030                 attr->ia_valid |= ATTR_ATIME_SET;
1031         }
1032         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1033                 attr->ia_mtime = CURRENT_TIME;
1034                 attr->ia_valid |= ATTR_MTIME_SET;
1035         }
1036
1037         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1038                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1039                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1040                        LTIME_S(CURRENT_TIME));
1041         if (lsm)
1042                 attr->ia_valid &= ~ATTR_SIZE;
1043
1044         /* If only OST attributes being set on objects, don't do MDS RPC.
1045          * In that case, we need to check permissions and update the local
1046          * inode ourselves so we can call obdo_from_inode() always. */
1047         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
1048                 struct lustre_md md;
1049                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1050
1051                 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
1052                                 attr, NULL, 0, NULL, 0, &request);
1053                 if (rc) {
1054                         ptlrpc_req_finished(request);
1055                         if (rc != -EPERM && rc != -EACCES)
1056                                 CERROR("md_setattr fails: rc = %d\n", rc);
1057                         RETURN(rc);
1058                 }
1059
1060                 rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
1061                                        sbi->ll_osc_exp, &md);
1062                 if (rc) {
1063                         ptlrpc_req_finished(request);
1064                         RETURN(rc);
1065                 }
1066
1067                 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
1068                  * but needed to set timestamps backwards on utime. */
1069                 inode_setattr(inode, attr);
1070                 ll_update_inode(inode, &md);
1071                 ptlrpc_req_finished(request);
1072
1073                 if (!lsm || !S_ISREG(inode->i_mode)) {
1074                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1075                         RETURN(0);
1076                 }
1077         } else {
1078                 /* The OST doesn't check permissions, but the alternative is
1079                  * a gratuitous RPC to the MDS.  We already rely on the client
1080                  * to do read/write/truncate permission checks, so is mtime OK?
1081                  */
1082                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1083                         /* from sys_utime() */
1084                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1085                                 if (current->fsuid != inode->i_uid &&
1086                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1087                                         RETURN(rc);
1088                         } else {
1089                                 /* from inode_change_ok() */
1090                                 if (current->fsuid != inode->i_uid &&
1091                                     !capable(CAP_FOWNER))
1092                                         RETURN(-EPERM);
1093                         }
1094                 }
1095
1096                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1097                 inode_setattr(inode, attr);
1098         }
1099
1100         /* We really need to get our PW lock before we change inode->i_size.
1101          * If we don't we can race with other i_size updaters on our node, like
1102          * ll_file_read.  We can also race with i_size propogation to other
1103          * nodes through dirtying and writeback of final cached pages.  This
1104          * last one is especially bad for racing o_append users on other
1105          * nodes. */
1106         if (ia_valid & ATTR_SIZE) {
1107                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1108                                                            OBD_OBJECT_EOF } };
1109                 struct lustre_handle lockh = { 0 };
1110                 int err, ast_flags = 0;
1111                 /* XXX when we fix the AST intents to pass the discard-range
1112                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1113                  * XXX here. */
1114                 if (attr->ia_size == 0)
1115                         ast_flags = LDLM_AST_DISCARD_DATA;
1116
1117                 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
1118                 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
1119                 up(&inode->i_sem);
1120                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1121                                     ast_flags);
1122                 down(&inode->i_sem);
1123                 if (rc != 0)
1124                         RETURN(rc);
1125
1126                 rc = vmtruncate(inode, attr->ia_size);
1127
1128                 /* We need to drop the semaphore here, because this unlock may
1129                  * result in a cancellation, which will need the i_sem */
1130                 up(&inode->i_sem);
1131                 /* unlock now as we don't mind others file lockers racing with
1132                  * the mds updates below? */
1133                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1134                 down(&inode->i_sem);
1135                 if (err) {
1136                         CERROR("ll_extent_unlock failed: %d\n", err);
1137                         if (!rc)
1138                                 rc = err;
1139                 }
1140         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1141                 struct obdo oa;
1142
1143                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1144                        inode->i_ino, LTIME_S(attr->ia_mtime));
1145                 oa.o_id = lsm->lsm_object_id;
1146                 oa.o_gr = lsm->lsm_object_gr;
1147                 oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1148                 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1149                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1150                 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1151                 if (rc)
1152                         CERROR("obd_setattr fails: rc=%d\n", rc);
1153         }
1154         RETURN(rc);
1155 }
1156
1157 int ll_setattr(struct dentry *de, struct iattr *attr)
1158 {
1159         LBUG(); /* code is unused, but leave this in case of VFS changes */
1160         RETURN(-ENOSYS);
1161 }
1162
1163 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1164                        unsigned long max_age)
1165 {
1166         struct ll_sb_info *sbi = ll_s2sbi(sb);
1167         struct obd_statfs obd_osfs;
1168         int rc;
1169         ENTRY;
1170
1171         rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1172         if (rc) {
1173                 CERROR("mdc_statfs fails: rc = %d\n", rc);
1174                 RETURN(rc);
1175         }
1176
1177         osfs->os_type = sb->s_magic;
1178
1179         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1180                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1181
1182         rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1183         if (rc) {
1184                 CERROR("obd_statfs fails: rc = %d\n", rc);
1185                 RETURN(rc);
1186         }
1187
1188         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1189                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1190                obd_osfs.os_files);
1191
1192         osfs->os_blocks = obd_osfs.os_blocks;
1193         osfs->os_bfree = obd_osfs.os_bfree;
1194         osfs->os_bavail = obd_osfs.os_bavail;
1195
1196         /* If we don't have as many objects free on the OST as inodes
1197          * on the MDS, we reduce the total number of inodes to
1198          * compensate, so that the "inodes in use" number is correct.
1199          */
1200         if (obd_osfs.os_ffree < osfs->os_ffree) {
1201                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1202                         obd_osfs.os_ffree;
1203                 osfs->os_ffree = obd_osfs.os_ffree;
1204         }
1205
1206         RETURN(rc);
1207 }
1208
1209 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1210 {
1211         struct obd_statfs osfs;
1212         int rc;
1213
1214         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1215         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1216
1217         /* For now we will always get up-to-date statfs values, but in the
1218          * future we may allow some amount of caching on the client (e.g.
1219          * from QOS or lprocfs updates). */
1220         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1221         if (rc)
1222                 return rc;
1223
1224         statfs_unpack(sfs, &osfs);
1225
1226         if (sizeof(sfs->f_blocks) == 4) {
1227                 while (osfs.os_blocks > ~0UL) {
1228                         sfs->f_bsize <<= 1;
1229
1230                         osfs.os_blocks >>= 1;
1231                         osfs.os_bfree >>= 1;
1232                         osfs.os_bavail >>= 1;
1233                 }
1234         }
1235
1236         sfs->f_blocks = osfs.os_blocks;
1237         sfs->f_bfree = osfs.os_bfree;
1238         sfs->f_bavail = osfs.os_bavail;
1239
1240         return 0;
1241 }
1242
1243 void dump_lsm(int level, struct lov_stripe_md *lsm)
1244 {
1245         CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1246                "stripe_size %u, stripe_count %u\n",
1247                lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1248                lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1249 }
1250
1251 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1252 {
1253         struct ll_inode_info *lli = ll_i2info(inode);
1254         struct mds_body *body = md->body;
1255         struct lov_stripe_md *lsm = md->lsm;
1256         struct mea *mea = md->mea;
1257         ENTRY;
1258
1259         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1260         LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1261         if (lsm != NULL) {
1262                 LASSERT(lsm->lsm_object_gr > 0);
1263                 if (lli->lli_smd == NULL) {
1264                         lli->lli_smd = lsm;
1265                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1266                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1267                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1268                 } else {
1269                         int i;
1270                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1271                                 CERROR("lsm mismatch for inode %ld\n",
1272                                        inode->i_ino);
1273                                 CERROR("lli_smd:\n");
1274                                 dump_lsm(D_ERROR, lli->lli_smd);
1275                                 CERROR("lsm:\n");
1276                                 dump_lsm(D_ERROR, lsm);
1277                                 LBUG();
1278                         }
1279                         /* XXX FIXME -- We should decide on a safer (atomic) and
1280                          * more elegant way to update the lsm */
1281                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1282                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1283                                         lsm->lsm_oinfo[i].loi_id;
1284                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1285                                         lsm->lsm_oinfo[i].loi_gr;
1286                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1287                                         lsm->lsm_oinfo[i].loi_ost_idx;
1288                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1289                                         lsm->lsm_oinfo[i].loi_ost_gen;
1290                         }
1291                 }
1292                 /* bug 2844 - limit i_blksize for broken user-space apps */
1293                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1294                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1295                 if (lli->lli_smd != lsm)
1296                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
1297         }
1298
1299         if (mea != NULL) {
1300                 if (lli->lli_mea == NULL) {
1301                         lli->lli_mea = mea;
1302                 } else {
1303                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1304                                 CERROR("mea mismatch for inode %lu\n",
1305                                         inode->i_ino);
1306                                 LBUG();
1307                         }
1308                 }
1309                 if (lli->lli_mea != mea)
1310                         obd_free_memmd(ll_i2mdcexp(inode),
1311                                         (struct lov_stripe_md **) &mea);
1312
1313         }
1314
1315         if (body->valid & OBD_MD_FLID)
1316                 inode->i_ino = body->ino;
1317         if (body->valid & OBD_MD_FLATIME)
1318                 LTIME_S(inode->i_atime) = body->atime;
1319         if (body->valid & OBD_MD_FLMTIME &&
1320             body->mtime > LTIME_S(inode->i_mtime)) {
1321                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1322                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1323                 LTIME_S(inode->i_mtime) = body->mtime;
1324         }
1325         if (body->valid & OBD_MD_FLCTIME &&
1326             body->ctime > LTIME_S(inode->i_ctime))
1327                 LTIME_S(inode->i_ctime) = body->ctime;
1328         if (body->valid & OBD_MD_FLMODE)
1329                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1330         if (body->valid & OBD_MD_FLTYPE)
1331                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1332         if (body->valid & OBD_MD_FLUID)
1333                 inode->i_uid = body->uid;
1334         if (body->valid & OBD_MD_FLGID)
1335                 inode->i_gid = body->gid;
1336         if (body->valid & OBD_MD_FLFLAGS)
1337                 inode->i_flags = body->flags;
1338         if (body->valid & OBD_MD_FLNLINK)
1339                 inode->i_nlink = body->nlink;
1340         if (body->valid & OBD_MD_FLGENER)
1341                 inode->i_generation = body->generation;
1342         if (body->valid & OBD_MD_FLRDEV)
1343 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1344                 inode->i_rdev = body->rdev;
1345 #else
1346                 inode->i_rdev = old_decode_dev(body->rdev);
1347 #endif
1348         if (body->valid & OBD_MD_FLSIZE)
1349                 inode->i_size = body->size;
1350         if (body->valid & OBD_MD_FLBLOCKS)
1351                 inode->i_blocks = body->blocks;
1352
1353         if (body->valid & OBD_MD_FLSIZE)
1354                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1355
1356         lli->lli_mds = body->mds;
1357         inode->i_dev = (kdev_t) body->mds;
1358         LASSERT(body->mds < 1000);
1359 }
1360
1361 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1362 static struct backing_dev_info ll_backing_dev_info = {
1363         .ra_pages       = 0,    /* No readahead */
1364         .memory_backed  = 0,    /* Does contribute to dirty memory */
1365 };
1366 #endif
1367
1368 void ll_read_inode2(struct inode *inode, void *opaque)
1369 {
1370         struct lustre_md *md = opaque;
1371         struct ll_inode_info *lli = ll_i2info(inode);
1372         ENTRY;
1373
1374         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1375                inode->i_generation, inode);
1376
1377         ll_lli_init(lli);
1378
1379         LASSERT(!lli->lli_smd);
1380
1381         /* Core attributes from the MDS first.  This is a new inode, and
1382          * the VFS doesn't zero times in the core inode so we have to do
1383          * it ourselves.  They will be overwritten by either MDS or OST
1384          * attributes - we just need to make sure they aren't newer. */
1385         LTIME_S(inode->i_mtime) = 0;
1386         LTIME_S(inode->i_atime) = 0;
1387         LTIME_S(inode->i_ctime) = 0;
1388
1389         inode->i_rdev = 0;
1390         ll_update_inode(inode, md);
1391
1392         /* OIDEBUG(inode); */
1393
1394         if (S_ISREG(inode->i_mode)) {
1395                 inode->i_op = &ll_file_inode_operations;
1396                 inode->i_fop = &ll_file_operations;
1397                 inode->i_mapping->a_ops = &ll_aops;
1398                 EXIT;
1399         } else if (S_ISDIR(inode->i_mode)) {
1400                 inode->i_op = &ll_dir_inode_operations;
1401                 inode->i_fop = &ll_dir_operations;
1402                 inode->i_mapping->a_ops = &ll_dir_aops;
1403                 EXIT;
1404         } else if (S_ISLNK(inode->i_mode)) {
1405                 inode->i_op = &ll_fast_symlink_inode_operations;
1406                 EXIT;
1407         } else {
1408                 inode->i_op = &ll_special_inode_operations;
1409
1410 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1411                 init_special_inode(inode, inode->i_mode,
1412                                    kdev_t_to_nr(inode->i_rdev));
1413
1414                 /* initializing backing dev info. */
1415                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1416 #else
1417                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1418 #endif
1419                 lli->ll_save_ifop = inode->i_fop;
1420
1421                 if (S_ISCHR(inode->i_mode))
1422                         inode->i_fop = &ll_special_chr_inode_fops;
1423                 else if (S_ISBLK(inode->i_mode))
1424                         inode->i_fop = &ll_special_blk_inode_fops;
1425                 else if (S_ISFIFO(inode->i_mode))
1426                         inode->i_fop = &ll_special_fifo_inode_fops;
1427                 else if (S_ISSOCK(inode->i_mode))
1428                         inode->i_fop = &ll_special_sock_inode_fops;
1429
1430                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1431                       inode->i_fop);
1432
1433                 if (lli->ll_save_ifop->owner) {
1434                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1435                               lli->ll_save_ifop->owner);
1436                 }
1437                 EXIT;
1438         }
1439 }
1440
1441 void ll_delete_inode(struct inode *inode)
1442 {
1443         int rc;
1444         struct ll_fid fid;
1445         struct ll_sb_info *sbi = ll_i2sbi(inode);
1446         ENTRY;
1447
1448         ll_inode2fid(&fid, inode);
1449
1450         rc = md_delete_object(sbi->ll_mdc_exp, &fid);
1451         if (rc) {
1452                 CERROR("md_delete_object() failed, error %d.\n",
1453                        rc);
1454         }
1455
1456         clear_inode(inode);
1457         EXIT;
1458 }
1459
1460 int ll_iocontrol(struct inode *inode, struct file *file,
1461                  unsigned int cmd, unsigned long arg)
1462 {
1463         struct ll_sb_info *sbi = ll_i2sbi(inode);
1464         struct ptlrpc_request *req = NULL;
1465         int rc, flags = 0;
1466         ENTRY;
1467
1468         switch(cmd) {
1469         case EXT3_IOC_GETFLAGS: {
1470                 struct ll_fid fid;
1471                 unsigned long valid = OBD_MD_FLFLAGS;
1472                 struct mds_body *body;
1473
1474                 ll_inode2fid(&fid, inode);
1475                 rc = md_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1476                 if (rc) {
1477                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1478                         RETURN(-abs(rc));
1479                 }
1480
1481                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1482
1483                 if (body->flags & S_APPEND)
1484                         flags |= EXT3_APPEND_FL;
1485                 if (body->flags & S_IMMUTABLE)
1486                         flags |= EXT3_IMMUTABLE_FL;
1487                 if (body->flags & S_NOATIME)
1488                         flags |= EXT3_NOATIME_FL;
1489
1490                 ptlrpc_req_finished (req);
1491
1492                 RETURN(put_user(flags, (int *)arg));
1493         }
1494         case EXT3_IOC_SETFLAGS: {
1495                 struct mdc_op_data op_data;
1496                 struct iattr attr;
1497                 struct obdo *oa;
1498                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1499
1500                 if (get_user(flags, (int *)arg))
1501                         RETURN(-EFAULT);
1502
1503                 oa = obdo_alloc();
1504                 if (!oa)
1505                         RETURN(-ENOMEM);
1506
1507                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1508
1509                 memset(&attr, 0x0, sizeof(attr));
1510                 attr.ia_attr_flags = flags;
1511                 attr.ia_valid |= ATTR_ATTR_FLAG;
1512
1513                 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
1514                                 &attr, NULL, 0, NULL, 0, &req);
1515                 if (rc) {
1516                         ptlrpc_req_finished(req);
1517                         if (rc != -EPERM && rc != -EACCES)
1518                                 CERROR("md_setattr fails: rc = %d\n", rc);
1519                         obdo_free(oa);
1520                         RETURN(rc);
1521                 }
1522                 ptlrpc_req_finished(req);
1523
1524                 oa->o_id = lsm->lsm_object_id;
1525                 oa->o_gr = lsm->lsm_object_gr;
1526                 oa->o_flags = flags;
1527                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1528
1529                 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1530                 obdo_free(oa);
1531                 if (rc) {
1532                         if (rc != -EPERM && rc != -EACCES)
1533                                 CERROR("md_setattr fails: rc = %d\n", rc);
1534                         RETURN(rc);
1535                 }
1536
1537                 if (flags & EXT3_APPEND_FL)
1538                         inode->i_flags |= S_APPEND;
1539                 else
1540                         inode->i_flags &= ~S_APPEND;
1541                 if (flags & EXT3_IMMUTABLE_FL)
1542                         inode->i_flags |= S_IMMUTABLE;
1543                 else
1544                         inode->i_flags &= ~S_IMMUTABLE;
1545                 if (flags & EXT3_NOATIME_FL)
1546                         inode->i_flags |= S_NOATIME;
1547                 else
1548                         inode->i_flags &= ~S_NOATIME;
1549
1550                 RETURN(0);
1551         }
1552         default:
1553                 RETURN(-ENOSYS);
1554         }
1555
1556         RETURN(0);
1557 }
1558
1559 void ll_umount_begin(struct super_block *sb)
1560 {
1561         struct ll_sb_info *sbi = ll_s2sbi(sb);
1562         struct obd_device *obd;
1563         struct obd_ioctl_data ioc_data = { 0 };
1564         ENTRY;
1565         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1566
1567         obd = class_exp2obd(sbi->ll_mdc_exp);
1568         if (obd == NULL) {
1569                 CERROR("Invalid MDC connection handle "LPX64"\n",
1570                        sbi->ll_mdc_exp->exp_handle.h_cookie);
1571                 EXIT;
1572                 return;
1573         }
1574         obd->obd_no_recov = 1;
1575         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1576                       &ioc_data, NULL);
1577
1578         obd = class_exp2obd(sbi->ll_osc_exp);
1579         if (obd == NULL) {
1580                 CERROR("Invalid LOV connection handle "LPX64"\n",
1581                        sbi->ll_osc_exp->exp_handle.h_cookie);
1582                 EXIT;
1583                 return;
1584         }
1585
1586         obd->obd_no_recov = 1;
1587         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1588                       &ioc_data, NULL);
1589
1590         /* Really, we'd like to wait until there are no requests outstanding,
1591          * and then continue.  For now, we just invalidate the requests,
1592          * schedule, and hope.
1593          */
1594         schedule();
1595
1596         EXIT;
1597 }
1598
1599 int ll_prep_inode(struct obd_export *osc_exp, struct obd_export *mdc_exp,
1600                   struct inode **inode, struct ptlrpc_request *req,
1601                   int offset, struct super_block *sb)
1602 {
1603         struct lustre_md md;
1604         int rc = 0;
1605
1606         rc = mdc_req2lustre_md(mdc_exp, req, offset, osc_exp, &md);
1607         if (rc)
1608                 RETURN(rc);
1609
1610         if (*inode) {
1611                 ll_update_inode(*inode, &md);
1612         } else {
1613                 LASSERT(sb);
1614                 *inode = ll_iget(sb, md.body->ino, &md);
1615                 if (*inode == NULL || is_bad_inode(*inode)) {
1616                         /* free the lsm if we allocated one above */
1617                         if (md.lsm != NULL)
1618                                 obd_free_memmd(osc_exp, &md.lsm);
1619                         if (md.mea != NULL)
1620                                 obd_free_memmd(mdc_exp,
1621                                                (struct lov_stripe_md**)&md.mea);
1622                         rc = -ENOMEM;
1623                         CERROR("new_inode -fatal: rc %d\n", rc);
1624                 }
1625         }
1626
1627         RETURN(rc);
1628 }