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         lli->lli_inode_magic = LLI_INODE_MAGIC;
 445 }
 446
 447 int ll_fill_super(struct super_block *sb, void *data, int silent)
 448 {
 449         struct ll_sb_info *sbi;
 450         char *osc = NULL;
 451         char *mdc = NULL;
 452         char *clone_opts = NULL;
 453         int err;
 454         ENTRY;
 455
 456         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 457
 458         sbi = lustre_init_sbi(sb);
 459         if (!sbi)
 460                 RETURN(-ENOMEM);
 461
 462         sbi->ll_flags |= LL_SBI_READAHEAD;
 463         ll_options(data, &osc, &mdc, &sbi->ll_flags, &clone_opts);
 464
 465         if (!osc) {
 466                 CERROR("no osc\n");
 467                 GOTO(out, err = -EINVAL);
 468         }
 469
 470         if (!mdc) {
 471                 CERROR("no mdc\n");
 472                 GOTO(out, err = -EINVAL);
 473         }
 474
 475         err = lustre_common_fill_super(sb, mdc, osc);
 476 #if CONFIG_SNAPFS
 477         if (clone_opts) {
 478                 int clone_index = 0;
 479                 char *endp;
 480
 481                 /*set clone info to the super block*/
 482                 clone_index = simple_strtoul(clone_opts, &endp, 0);
 483                 err = lustre_set_clone_info(sb, clone_index);
 484         }
 485 #endif
 486 out:
 487         if (err)
 488                 lustre_free_sbi(sb);
 489
 490         if (mdc)
 491                 OBD_FREE(mdc, strlen(mdc) + 1);
 492         if (osc)
 493                 OBD_FREE(osc, strlen(osc) + 1);
 494         if (clone_opts)
 495                 OBD_FREE(clone_opts, strlen(clone_opts) + 1);
 496
 497         RETURN(err);
 498 } /* ll_read_super */
 499
 500 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 501                               struct config_llog_instance *cfg, int allow_recov)
 502 {
 503         struct lustre_cfg lcfg;
 504         struct portals_cfg pcfg;
 505         char * peer = "MDS_PEER_UUID";
 506         struct obd_device *obd;
 507         struct lustre_handle mdc_conn = {0, };
 508         struct obd_export *exp;
 509         char * name = "mdc_dev";
 510         class_uuid_t uuid;
 511         struct obd_uuid mdc_uuid;
 512         struct llog_ctxt *ctxt;
 513         int rc = 0;
 514         int err;
 515         ENTRY;
 516
 517         if (lmd_bad_magic(lmd))
 518                 RETURN(-EINVAL);
 519
 520         generate_random_uuid(uuid);
 521         class_uuid_unparse(uuid, &mdc_uuid);
 522
 523         if (lmd->lmd_local_nid) {
 524                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 525                 pcfg.pcfg_nal = lmd->lmd_nal;
 526                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 527                 err = libcfs_nal_cmd(&pcfg);
 528                 if (err <0)
 529                         GOTO(out, err);
 530         }
 531
 532         if (lmd->lmd_nal == SOCKNAL) {
 533                 PCFG_INIT(pcfg, NAL_CMD_ADD_AUTOCONN);
 534                 pcfg.pcfg_nal     = lmd->lmd_nal;
 535                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 536                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 537                 pcfg.pcfg_misc    = lmd->lmd_port;
 538                 pcfg.pcfg_size    = 8388608;
 539                 pcfg.pcfg_flags   = 0x4; /*share*/
 540                 err = libcfs_nal_cmd(&pcfg);
 541                 if (err <0)
 542                         GOTO(out, err);
 543         }
 544
 545         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
 546         lcfg.lcfg_nid = lmd->lmd_server_nid;
 547         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 548         lcfg.lcfg_inlbuf1 = peer;
 549         lcfg.lcfg_nal = lmd->lmd_nal;
 550         err = class_process_config(&lcfg);
 551         if (err < 0)
 552                 GOTO(out_del_conn, err);
 553
 554         LCFG_INIT(lcfg, LCFG_ATTACH, name);
 555         lcfg.lcfg_inlbuf1 = "mdc";
 556         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 557         lcfg.lcfg_inlbuf2 = mdc_uuid.uuid;
 558         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 559         err = class_process_config(&lcfg);
 560         if (err < 0)
 561                 GOTO(out_del_uuid, err);
 562
 563         LCFG_INIT(lcfg, LCFG_SETUP, name);
 564         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
 565         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 566         lcfg.lcfg_inlbuf2 = peer;
 567         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 568         err = class_process_config(&lcfg);
 569         if (err < 0)
 570                 GOTO(out_detach, err);
 571
 572         obd = class_name2obd(name);
 573         if (obd == NULL)
 574                 GOTO(out_cleanup, err = -EINVAL);
 575
 576         /* Disable initial recovery on this import */
 577         err = obd_set_info(obd->obd_self_export,
 578                            strlen("initial_recov"), "initial_recov",
 579                            sizeof(allow_recov), &allow_recov);
 580         if (err)
 581                 GOTO(out_cleanup, err);
 582
 583         err = obd_connect(&mdc_conn, obd, &mdc_uuid);
 584         if (err) {
 585                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, err);
 586                 GOTO(out_cleanup, err);
 587         }
 588
 589         exp = class_conn2export(&mdc_conn);
 590
 591         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 592         rc = class_config_process_llog(ctxt, profile, cfg);
 593         if (rc)
 594                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 595
 596         err = obd_disconnect(exp, 0);
 597
 598 out_cleanup:
 599         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
 600         err = class_process_config(&lcfg);
 601         if (err < 0)
 602                 GOTO(out, err);
 603
 604 out_detach:
 605         LCFG_INIT(lcfg, LCFG_DETACH, name);
 606         err = class_process_config(&lcfg);
 607         if (err < 0)
 608                 GOTO(out, err);
 609
 610 out_del_uuid:
 611         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
 612         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 613         lcfg.lcfg_inlbuf1 = peer;
 614         err = class_process_config(&lcfg);
 615
 616 out_del_conn:
 617         if (lmd->lmd_nal == SOCKNAL) {
 618                 PCFG_INIT(pcfg, NAL_CMD_DEL_AUTOCONN);
 619                 pcfg.pcfg_nal     = lmd->lmd_nal;
 620                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 621                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 622                 pcfg.pcfg_flags   = 1; /*share*/
 623                 err = libcfs_nal_cmd(&pcfg);
 624                 if (err <0)
 625                         GOTO(out, err);
 626         }
 627 out:
 628         if (rc == 0)
 629                 rc = err;
 630
 631         RETURN(rc);
 632 }
 633
 634 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 635 {
 636         struct lustre_mount_data * lmd = data;
 637         struct ll_sb_info *sbi;
 638         char *osc = NULL;
 639         char *mdc = NULL;
 640         int err;
 641         ENTRY;
 642
 643         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 644         if (lmd_bad_magic(lmd))
 645                 RETURN(-EINVAL);
 646
 647         sbi = lustre_init_sbi(sb);
 648         if (!sbi)
 649                 RETURN(-ENOMEM);
 650
 651         sbi->ll_flags |= LL_SBI_READAHEAD;
 652
 653         if (lmd->lmd_profile) {
 654                 struct lustre_profile *lprof;
 655                 struct config_llog_instance cfg;
 656                 int len;
 657
 658                 if (lmd->lmd_mds[0] == '\0') {
 659                         CERROR("no mds name\n");
 660                         GOTO(out_free, err = -EINVAL);
 661                 }
 662
 663                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 664                 if (sbi->ll_lmd == NULL)
 665                         GOTO(out_free, err = -ENOMEM);
 666                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 667
 668                 /* generate a string unique to this super, let's try
 669                  the address of the super itself.*/
 670                 len = (sizeof(sb) * 2) + 1;
 671                 OBD_ALLOC(sbi->ll_instance, len);
 672                 if (sbi->ll_instance == NULL)
 673                         GOTO(out_free, err = -ENOMEM);
 674                 sprintf(sbi->ll_instance, "%p", sb);
 675
 676                 cfg.cfg_instance = sbi->ll_instance;
 677                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 678                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 679                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 680                 if (err < 0) {
 681                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 682                         GOTO(out_free, err);
 683                 }
 684
 685                 lprof = class_get_profile(lmd->lmd_profile);
 686                 if (lprof == NULL) {
 687                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 688                         GOTO(out_free, err = -EINVAL);
 689                 }
 690                 if (osc)
 691                         OBD_FREE(osc, strlen(osc) + 1);
 692                 OBD_ALLOC(osc, strlen(lprof->lp_osc) +
 693                           strlen(sbi->ll_instance) + 2);
 694                 sprintf(osc, "%s-%s", lprof->lp_osc, sbi->ll_instance);
 695
 696                 if (mdc)
 697                         OBD_FREE(mdc, strlen(mdc) + 1);
 698                 OBD_ALLOC(mdc, strlen(lprof->lp_mdc) +
 699                           strlen(sbi->ll_instance) + 2);
 700                 sprintf(mdc, "%s-%s", lprof->lp_mdc, sbi->ll_instance);
 701         }
 702
 703         if (!osc) {
 704                 CERROR("no osc\n");
 705                 GOTO(out_free, err = -EINVAL);
 706         }
 707
 708         if (!mdc) {
 709                 CERROR("no mdc\n");
 710                 GOTO(out_free, err = -EINVAL);
 711         }
 712
 713         err = lustre_common_fill_super(sb, mdc, osc);
 714
 715         if (err)
 716                 GOTO(out_free, err);
 717
 718 #if CONFIG_SNAPFS
 719         if (lmd->lmd_clone_index) {
 720                 err = lustre_set_clone_info(sb, lmd->lmd_clone_index);
 721         }
 722
 723         if (err)
 724                 GOTO(out_free, err);
 725 #endif
 726
 727 out_dev:
 728         if (mdc)
 729                 OBD_FREE(mdc, strlen(mdc) + 1);
 730         if (osc)
 731                 OBD_FREE(osc, strlen(osc) + 1);
 732
 733         RETURN(err);
 734
 735 out_free:
 736         if (sbi->ll_lmd) {
 737                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 738                 int err;
 739
 740                 if (sbi->ll_instance != NULL) {
 741                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 742                         char * cln_prof;
 743                         struct config_llog_instance cfg;
 744
 745                         cfg.cfg_instance = sbi->ll_instance;
 746                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 747
 748                         OBD_ALLOC(cln_prof, len);
 749                         sprintf(cln_prof, "%s-clean", lmd->lmd_profile);
 750
 751                         err = lustre_process_log(lmd, cln_prof, &cfg, 0);
 752                         if (err < 0)
 753                                 CERROR("Unable to process log: %s\n", cln_prof);
 754                         OBD_FREE(cln_prof, len);
 755                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance)+ 1);
 756                 }
 757                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 758         }
 759 #if CONFIG_SNAPFS
 760         lustre_cleanup_clone_info(sb);
 761 #endif
 762         lustre_free_sbi(sb);
 763
 764         goto out_dev;
 765 } /* lustre_fill_super */
 766
 767 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 768 {
 769         struct lustre_cfg lcfg;
 770         struct obd_device *obd;
 771         int next = 0;
 772
 773         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 774         {
 775                 int err;
 776
 777                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
 778                 err = class_process_config(&lcfg);
 779                 if (err) {
 780                         CERROR("cleanup failed: %s\n", obd->obd_name);
 781                         //continue;
 782                 }
 783
 784                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
 785                 err = class_process_config(&lcfg);
 786                 if (err) {
 787                         CERROR("detach failed: %s\n", obd->obd_name);
 788                         //continue;
 789                 }
 790         }
 791
 792         if (sbi->ll_lmd != NULL)
 793                 class_del_profile(sbi->ll_lmd->lmd_profile);
 794 }
 795
 796 void lustre_put_super(struct super_block *sb)
 797 {
 798         struct obd_device *obd;
 799         struct ll_sb_info *sbi = ll_s2sbi(sb);
 800         int force_umount = 0;
 801         ENTRY;
 802
 803         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 804         obd = class_exp2obd(sbi->ll_mdc_exp);
 805         if (obd)
 806                 force_umount = obd->obd_no_recov;
 807         obd = NULL;
 808
 809 #if CONFIG_SNAPFS
 810         lustre_cleanup_clone_info(sb);
 811 #endif
 812
 813         lustre_common_put_super(sb);
 814         if (sbi->ll_lmd != NULL) {
 815                 char * cln_prof;
 816                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 817                 int err;
 818                 struct config_llog_instance cfg;
 819
 820                 if (force_umount) {
 821                         CERROR("force umount, doing manual cleanup\n");
 822                         lustre_manual_cleanup(sbi);
 823                         GOTO(free_lmd, 0);
 824                 }
 825
 826                 cfg.cfg_instance = sbi->ll_instance;
 827                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 828
 829                 OBD_ALLOC(cln_prof, len);
 830                 sprintf(cln_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 831
 832                 err = lustre_process_log(sbi->ll_lmd, cln_prof, &cfg, 0);
 833                 if (err < 0) {
 834                         CERROR("Unable to process log: %s, doing manual cleanup"
 835                                "\n", cln_prof);
 836                         lustre_manual_cleanup(sbi);
 837                 }
 838
 839                 OBD_FREE(cln_prof, len);
 840         free_lmd:
 841                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 842                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 843         }
 844
 845         lustre_free_sbi(sb);
 846
 847         EXIT;
 848 } /* lustre_put_super */
 849
 850 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 851 {
 852         struct obd_export *mdc_exp = sbi->ll_mdc_exp;
 853         struct lustre_mount_data *lmd = sbi->ll_lmd;
 854         struct llog_ctxt *ctxt;
 855         struct config_llog_instance cfg;
 856         char *profile = lmd->lmd_profile, *name = NULL;
 857         int rc, namelen =  0, version;
 858         ENTRY;
 859
 860         if (profile == NULL)
 861                 RETURN(0);
 862         if (lmd == NULL) {
 863                 CERROR("Client not mounted with zero-conf; cannot process "
 864                        "update log.\n");
 865                 RETURN(0);
 866         }
 867
 868         rc = ldlm_cli_cancel_unused(mdc_exp->exp_obd->obd_namespace, NULL,
 869                                     LDLM_FL_CONFIG_CHANGE, NULL);
 870         if (rc != 0)
 871                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
 872
 873         rc = obd_cancel_unused(sbi->ll_osc_exp, NULL, LDLM_FL_CONFIG_CHANGE,
 874                                NULL);
 875         if (rc != 0)
 876                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 877
 878         cfg.cfg_instance = sbi->ll_instance;
 879         cfg.cfg_uuid = sbi->ll_sb_uuid;
 880         cfg.cfg_local_nid = lmd->lmd_local_nid;
 881
 882         namelen = strlen(profile) + 20; /* -clean-######### */
 883         OBD_ALLOC(name, namelen);
 884         if (name == NULL)
 885                 RETURN(-ENOMEM);
 886
 887         if (clean) {
 888                 version = sbi->ll_config_version - 1;
 889                 sprintf(name, "%s-clean-%d", profile, version);
 890         } else {
 891                 version = sbi->ll_config_version + 1;
 892                 sprintf(name, "%s-%d", profile, version);
 893         }
 894
 895         CWARN("Applying configuration log %s\n", name);
 896
 897         ctxt = llog_get_context(&mdc_exp->exp_obd->obd_llogs,
 898                                 LLOG_CONFIG_REPL_CTXT);
 899         rc = class_config_process_llog(ctxt, name, &cfg);
 900         if (rc == 0)
 901                 sbi->ll_config_version = version;
 902         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 903
 904         if (rc == 0 && clean == 0) {
 905                 struct lov_desc desc;
 906                 int rc, valsize;
 907                 valsize = sizeof(desc);
 908                 rc = obd_get_info(sbi->ll_osc_exp, strlen("lovdesc") + 1,
 909                                   "lovdesc", &valsize, &desc);
 910
 911                 rc = obd_init_ea_size(mdc_exp,
 912                                       obd_size_diskmd(sbi->ll_osc_exp, NULL),
 913                                       (desc.ld_tgt_count *
 914                                        sizeof(struct llog_cookie)));
 915         }
 916         OBD_FREE(name, namelen);
 917         RETURN(rc);
 918 }
 919
 920 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 921 {
 922         struct inode *inode = NULL;
 923         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 924         if (lock->l_ast_data) {
 925                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 926                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 927                         inode = igrab(lock->l_ast_data);
 928                 } else {
 929                         CERROR("DEBUG: l_ast_data %p is bogus: magic %x\n",
 930                                lock->l_ast_data, lli->lli_inode_magic);
 931                 }
 932         }
 933         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 934         return inode;
 935 }
 936
 937 int null_if_equal(struct ldlm_lock *lock, void *data)
 938 {
 939         if (data == lock->l_ast_data)
 940                 lock->l_ast_data = NULL;
 941
 942         if (lock->l_req_mode != lock->l_granted_mode)
 943                 return LDLM_ITER_STOP;
 944
 945         return LDLM_ITER_CONTINUE;
 946 }
 947
 948 void ll_clear_inode(struct inode *inode)
 949 {
 950         struct ll_fid fid;
 951         struct ll_inode_info *lli = ll_i2info(inode);
 952         struct ll_sb_info *sbi = ll_i2sbi(inode);
 953         ENTRY;
 954
 955         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 956                inode->i_generation, inode);
 957
 958         lli->lli_inode_magic = LLI_INODE_DEAD;
 959         ll_inode2fid(&fid, inode);
 960         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 961         md_change_cbdata(sbi->ll_mdc_exp, &fid, null_if_equal, inode);
 962
 963         if (lli->lli_smd)
 964                 obd_change_cbdata(sbi->ll_osc_exp, lli->lli_smd,
 965                                   null_if_equal, inode);
 966
 967         if (lli->lli_smd) {
 968                 obd_free_memmd(sbi->ll_osc_exp, &lli->lli_smd);
 969                 lli->lli_smd = NULL;
 970         }
 971
 972         if (lli->lli_mea) {
 973                 /* FIXME: change cbdata for mea here */
 974                 obd_free_memmd(sbi->ll_mdc_exp,
 975                                (struct lov_stripe_md **) &lli->lli_mea);
 976                 lli->lli_mea = NULL;
 977         }
 978
 979         if (lli->lli_symlink_name) {
 980                 OBD_FREE(lli->lli_symlink_name,
 981                          strlen(lli->lli_symlink_name) + 1);
 982                 lli->lli_symlink_name = NULL;
 983         }
 984
 985         EXIT;
 986 }
 987
 988 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 989  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 990  * keep these values until such a time that objects are allocated for it.
 991  * We do the MDS operations first, as it is checking permissions for us.
 992  * We don't to the MDS RPC if there is nothing that we want to store there,
 993  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 994  * going to do an RPC anyways.
 995  *
 996  * If we are doing a truncate, we will send the mtime and ctime updates
 997  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 998  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 999  * at the same time.
1000  */
1001 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1002 {
1003         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1004         struct ll_sb_info *sbi = ll_i2sbi(inode);
1005         struct ptlrpc_request *request = NULL;
1006         struct mdc_op_data op_data;
1007         int ia_valid = attr->ia_valid;
1008         int rc = 0;
1009         ENTRY;
1010
1011         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1012         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1013
1014         if (ia_valid & ATTR_SIZE) {
1015                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1016                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1017                                attr->ia_size, ll_file_maxbytes(inode));
1018                         RETURN(-EFBIG);
1019                 }
1020
1021                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1022         }
1023
1024         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1025         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1026                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1027                         RETURN(-EPERM);
1028         }
1029
1030         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1031         if (attr->ia_valid & ATTR_CTIME) {
1032                 attr->ia_ctime = CURRENT_TIME;
1033                 attr->ia_valid |= ATTR_CTIME_SET;
1034         }
1035         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1036                 attr->ia_atime = CURRENT_TIME;
1037                 attr->ia_valid |= ATTR_ATIME_SET;
1038         }
1039         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1040                 attr->ia_mtime = CURRENT_TIME;
1041                 attr->ia_valid |= ATTR_MTIME_SET;
1042         }
1043
1044         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1045                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1046                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1047                        LTIME_S(CURRENT_TIME));
1048         if (lsm)
1049                 attr->ia_valid &= ~ATTR_SIZE;
1050
1051         /* If only OST attributes being set on objects, don't do MDS RPC.
1052          * In that case, we need to check permissions and update the local
1053          * inode ourselves so we can call obdo_from_inode() always. */
1054         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN | ATTR_RAW) : ~0)) {
1055                 struct lustre_md md;
1056                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1057
1058                 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
1059                                 attr, NULL, 0, NULL, 0, &request);
1060                 if (rc) {
1061                         ptlrpc_req_finished(request);
1062                         if (rc != -EPERM && rc != -EACCES)
1063                                 CERROR("md_setattr fails: rc = %d\n", rc);
1064                         RETURN(rc);
1065                 }
1066
1067                 rc = mdc_req2lustre_md(sbi->ll_mdc_exp, request, 0,
1068                                        sbi->ll_osc_exp, &md);
1069                 if (rc) {
1070                         ptlrpc_req_finished(request);
1071                         RETURN(rc);
1072                 }
1073
1074                 /* Won't invoke vmtruncate as we already cleared ATTR_SIZE,
1075                  * but needed to set timestamps backwards on utime. */
1076                 inode_setattr(inode, attr);
1077                 ll_update_inode(inode, &md);
1078                 ptlrpc_req_finished(request);
1079
1080                 if (!lsm || !S_ISREG(inode->i_mode)) {
1081                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1082                         RETURN(0);
1083                 }
1084         } else {
1085                 /* The OST doesn't check permissions, but the alternative is
1086                  * a gratuitous RPC to the MDS.  We already rely on the client
1087                  * to do read/write/truncate permission checks, so is mtime OK?
1088                  */
1089                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1090                         /* from sys_utime() */
1091                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1092                                 if (current->fsuid != inode->i_uid &&
1093                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1094                                         RETURN(rc);
1095                         } else {
1096                                 /* from inode_change_ok() */
1097                                 if (current->fsuid != inode->i_uid &&
1098                                     !capable(CAP_FOWNER))
1099                                         RETURN(-EPERM);
1100                         }
1101                 }
1102
1103                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1104                 inode_setattr(inode, attr);
1105         }
1106
1107         /* We really need to get our PW lock before we change inode->i_size.
1108          * If we don't we can race with other i_size updaters on our node, like
1109          * ll_file_read.  We can also race with i_size propogation to other
1110          * nodes through dirtying and writeback of final cached pages.  This
1111          * last one is especially bad for racing o_append users on other
1112          * nodes. */
1113         if (ia_valid & ATTR_SIZE) {
1114                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1115                                                            OBD_OBJECT_EOF } };
1116                 struct lustre_handle lockh = { 0 };
1117                 int err, ast_flags = 0;
1118                 /* XXX when we fix the AST intents to pass the discard-range
1119                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1120                  * XXX here. */
1121                 if (attr->ia_size == 0)
1122                         ast_flags = LDLM_AST_DISCARD_DATA;
1123
1124                 /* bug 1639: avoid write/truncate i_sem/DLM deadlock */
1125                 LASSERT(atomic_read(&inode->i_sem.count) <= 0);
1126                 up(&inode->i_sem);
1127                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1128                                     ast_flags);
1129                 down(&inode->i_sem);
1130                 if (rc != 0)
1131                         RETURN(rc);
1132
1133                 rc = vmtruncate(inode, attr->ia_size);
1134
1135                 /* We need to drop the semaphore here, because this unlock may
1136                  * result in a cancellation, which will need the i_sem */
1137                 up(&inode->i_sem);
1138                 /* unlock now as we don't mind others file lockers racing with
1139                  * the mds updates below? */
1140                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1141                 down(&inode->i_sem);
1142                 if (err) {
1143                         CERROR("ll_extent_unlock failed: %d\n", err);
1144                         if (!rc)
1145                                 rc = err;
1146                 }
1147         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1148                 struct obdo oa;
1149
1150                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1151                        inode->i_ino, LTIME_S(attr->ia_mtime));
1152                 oa.o_id = lsm->lsm_object_id;
1153                 oa.o_gr = lsm->lsm_object_gr;
1154                 oa.o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1155                 obdo_from_inode(&oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1156                                             OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1157                 rc = obd_setattr(sbi->ll_osc_exp, &oa, lsm, NULL);
1158                 if (rc)
1159                         CERROR("obd_setattr fails: rc=%d\n", rc);
1160         }
1161         RETURN(rc);
1162 }
1163
1164 int ll_setattr(struct dentry *de, struct iattr *attr)
1165 {
1166         LBUG(); /* code is unused, but leave this in case of VFS changes */
1167         RETURN(-ENOSYS);
1168 }
1169
1170 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1171                        unsigned long max_age)
1172 {
1173         struct ll_sb_info *sbi = ll_s2sbi(sb);
1174         struct obd_statfs obd_osfs;
1175         int rc;
1176         ENTRY;
1177
1178         rc = obd_statfs(class_exp2obd(sbi->ll_mdc_exp), osfs, max_age);
1179         if (rc) {
1180                 CERROR("mdc_statfs fails: rc = %d\n", rc);
1181                 RETURN(rc);
1182         }
1183
1184         osfs->os_type = sb->s_magic;
1185
1186         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1187                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1188
1189         rc = obd_statfs(class_exp2obd(sbi->ll_osc_exp), &obd_osfs, max_age);
1190         if (rc) {
1191                 CERROR("obd_statfs fails: rc = %d\n", rc);
1192                 RETURN(rc);
1193         }
1194
1195         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1196                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1197                obd_osfs.os_files);
1198
1199         osfs->os_blocks = obd_osfs.os_blocks;
1200         osfs->os_bfree = obd_osfs.os_bfree;
1201         osfs->os_bavail = obd_osfs.os_bavail;
1202
1203         /* If we don't have as many objects free on the OST as inodes
1204          * on the MDS, we reduce the total number of inodes to
1205          * compensate, so that the "inodes in use" number is correct.
1206          */
1207         if (obd_osfs.os_ffree < osfs->os_ffree) {
1208                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1209                         obd_osfs.os_ffree;
1210                 osfs->os_ffree = obd_osfs.os_ffree;
1211         }
1212
1213         RETURN(rc);
1214 }
1215
1216 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1217 {
1218         struct obd_statfs osfs;
1219         int rc;
1220
1221         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1222         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1223
1224         /* For now we will always get up-to-date statfs values, but in the
1225          * future we may allow some amount of caching on the client (e.g.
1226          * from QOS or lprocfs updates). */
1227         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1228         if (rc)
1229                 return rc;
1230
1231         statfs_unpack(sfs, &osfs);
1232
1233         if (sizeof(sfs->f_blocks) == 4) {
1234                 while (osfs.os_blocks > ~0UL) {
1235                         sfs->f_bsize <<= 1;
1236
1237                         osfs.os_blocks >>= 1;
1238                         osfs.os_bfree >>= 1;
1239                         osfs.os_bavail >>= 1;
1240                 }
1241         }
1242
1243         sfs->f_blocks = osfs.os_blocks;
1244         sfs->f_bfree = osfs.os_bfree;
1245         sfs->f_bavail = osfs.os_bavail;
1246
1247         return 0;
1248 }
1249
1250 void dump_lsm(int level, struct lov_stripe_md *lsm)
1251 {
1252         CDEBUG(level, "objid "LPX64", maxbytes "LPX64", magic 0x%08X, "
1253                "stripe_size %u, stripe_count %u\n",
1254                lsm->lsm_object_id, lsm->lsm_maxbytes, lsm->lsm_magic,
1255                lsm->lsm_stripe_size, lsm->lsm_stripe_count);
1256 }
1257
1258 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1259 {
1260         struct ll_inode_info *lli = ll_i2info(inode);
1261         struct mds_body *body = md->body;
1262         struct lov_stripe_md *lsm = md->lsm;
1263         struct mea *mea = md->mea;
1264         ENTRY;
1265
1266         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1267         LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1268         if (lsm != NULL) {
1269                 LASSERT(lsm->lsm_object_gr > 0);
1270                 if (lli->lli_smd == NULL) {
1271                         lli->lli_smd = lsm;
1272                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1273                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1274                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1275                 } else {
1276                         int i;
1277                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1278                                 CERROR("lsm mismatch for inode %ld\n",
1279                                        inode->i_ino);
1280                                 CERROR("lli_smd:\n");
1281                                 dump_lsm(D_ERROR, lli->lli_smd);
1282                                 CERROR("lsm:\n");
1283                                 dump_lsm(D_ERROR, lsm);
1284                                 LBUG();
1285                         }
1286                         /* XXX FIXME -- We should decide on a safer (atomic) and
1287                          * more elegant way to update the lsm */
1288                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1289                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1290                                         lsm->lsm_oinfo[i].loi_id;
1291                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1292                                         lsm->lsm_oinfo[i].loi_gr;
1293                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1294                                         lsm->lsm_oinfo[i].loi_ost_idx;
1295                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1296                                         lsm->lsm_oinfo[i].loi_ost_gen;
1297                         }
1298                 }
1299                 /* bug 2844 - limit i_blksize for broken user-space apps */
1300                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1301                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1302                 if (lli->lli_smd != lsm)
1303                         obd_free_memmd(ll_i2obdexp(inode), &lsm);
1304         }
1305
1306         if (mea != NULL) {
1307                 if (lli->lli_mea == NULL) {
1308                         lli->lli_mea = mea;
1309                 } else {
1310                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1311                                 CERROR("mea mismatch for inode %lu\n",
1312                                         inode->i_ino);
1313                                 LBUG();
1314                         }
1315                 }
1316                 if (lli->lli_mea != mea)
1317                         obd_free_memmd(ll_i2mdcexp(inode),
1318                                         (struct lov_stripe_md **) &mea);
1319
1320         }
1321
1322         if (body->valid & OBD_MD_FLID)
1323                 inode->i_ino = body->ino;
1324         if (body->valid & OBD_MD_FLATIME)
1325                 LTIME_S(inode->i_atime) = body->atime;
1326         if (body->valid & OBD_MD_FLMTIME &&
1327             body->mtime > LTIME_S(inode->i_mtime)) {
1328                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1329                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1330                 LTIME_S(inode->i_mtime) = body->mtime;
1331         }
1332         if (body->valid & OBD_MD_FLCTIME &&
1333             body->ctime > LTIME_S(inode->i_ctime))
1334                 LTIME_S(inode->i_ctime) = body->ctime;
1335         if (body->valid & OBD_MD_FLMODE)
1336                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1337         if (body->valid & OBD_MD_FLTYPE)
1338                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1339         if (body->valid & OBD_MD_FLUID)
1340                 inode->i_uid = body->uid;
1341         if (body->valid & OBD_MD_FLGID)
1342                 inode->i_gid = body->gid;
1343         if (body->valid & OBD_MD_FLFLAGS)
1344                 inode->i_flags = body->flags;
1345         if (body->valid & OBD_MD_FLNLINK)
1346                 inode->i_nlink = body->nlink;
1347         if (body->valid & OBD_MD_FLGENER)
1348                 inode->i_generation = body->generation;
1349         if (body->valid & OBD_MD_FLRDEV)
1350 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1351                 inode->i_rdev = body->rdev;
1352 #else
1353                 inode->i_rdev = old_decode_dev(body->rdev);
1354 #endif
1355         if (body->valid & OBD_MD_FLSIZE)
1356                 inode->i_size = body->size;
1357         if (body->valid & OBD_MD_FLBLOCKS)
1358                 inode->i_blocks = body->blocks;
1359
1360         if (body->valid & OBD_MD_FLSIZE)
1361                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1362
1363         lli->lli_mds = body->mds;
1364         inode->i_dev = (kdev_t) body->mds;
1365         LASSERT(body->mds < 1000);
1366 }
1367
1368 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1369 static struct backing_dev_info ll_backing_dev_info = {
1370         .ra_pages       = 0,    /* No readahead */
1371         .memory_backed  = 0,    /* Does contribute to dirty memory */
1372 };
1373 #endif
1374
1375 void ll_read_inode2(struct inode *inode, void *opaque)
1376 {
1377         struct lustre_md *md = opaque;
1378         struct ll_inode_info *lli = ll_i2info(inode);
1379         ENTRY;
1380
1381         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1382                inode->i_generation, inode);
1383
1384         ll_lli_init(lli);
1385
1386         LASSERT(!lli->lli_smd);
1387
1388         /* Core attributes from the MDS first.  This is a new inode, and
1389          * the VFS doesn't zero times in the core inode so we have to do
1390          * it ourselves.  They will be overwritten by either MDS or OST
1391          * attributes - we just need to make sure they aren't newer. */
1392         LTIME_S(inode->i_mtime) = 0;
1393         LTIME_S(inode->i_atime) = 0;
1394         LTIME_S(inode->i_ctime) = 0;
1395
1396         inode->i_rdev = 0;
1397         ll_update_inode(inode, md);
1398
1399         /* OIDEBUG(inode); */
1400
1401         if (S_ISREG(inode->i_mode)) {
1402                 inode->i_op = &ll_file_inode_operations;
1403                 inode->i_fop = &ll_file_operations;
1404                 inode->i_mapping->a_ops = &ll_aops;
1405                 EXIT;
1406         } else if (S_ISDIR(inode->i_mode)) {
1407                 inode->i_op = &ll_dir_inode_operations;
1408                 inode->i_fop = &ll_dir_operations;
1409                 inode->i_mapping->a_ops = &ll_dir_aops;
1410                 EXIT;
1411         } else if (S_ISLNK(inode->i_mode)) {
1412                 inode->i_op = &ll_fast_symlink_inode_operations;
1413                 EXIT;
1414         } else {
1415                 inode->i_op = &ll_special_inode_operations;
1416
1417 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1418                 init_special_inode(inode, inode->i_mode,
1419                                    kdev_t_to_nr(inode->i_rdev));
1420
1421                 /* initializing backing dev info. */
1422                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1423 #else
1424                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1425 #endif
1426                 lli->ll_save_ifop = inode->i_fop;
1427
1428                 if (S_ISCHR(inode->i_mode))
1429                         inode->i_fop = &ll_special_chr_inode_fops;
1430                 else if (S_ISBLK(inode->i_mode))
1431                         inode->i_fop = &ll_special_blk_inode_fops;
1432                 else if (S_ISFIFO(inode->i_mode))
1433                         inode->i_fop = &ll_special_fifo_inode_fops;
1434                 else if (S_ISSOCK(inode->i_mode))
1435                         inode->i_fop = &ll_special_sock_inode_fops;
1436
1437                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1438                       inode->i_fop);
1439
1440                 if (lli->ll_save_ifop->owner) {
1441                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1442                               lli->ll_save_ifop->owner);
1443                 }
1444                 EXIT;
1445         }
1446 }
1447
1448 void ll_delete_inode(struct inode *inode)
1449 {
1450         int rc;
1451         struct ll_fid fid;
1452         struct ll_sb_info *sbi = ll_i2sbi(inode);
1453         ENTRY;
1454
1455         ll_inode2fid(&fid, inode);
1456
1457         rc = md_delete_object(sbi->ll_mdc_exp, &fid);
1458         if (rc) {
1459                 CERROR("md_delete_object() failed, error %d.\n",
1460                        rc);
1461         }
1462
1463         clear_inode(inode);
1464         EXIT;
1465 }
1466
1467 int ll_iocontrol(struct inode *inode, struct file *file,
1468                  unsigned int cmd, unsigned long arg)
1469 {
1470         struct ll_sb_info *sbi = ll_i2sbi(inode);
1471         struct ptlrpc_request *req = NULL;
1472         int rc, flags = 0;
1473         ENTRY;
1474
1475         switch(cmd) {
1476         case EXT3_IOC_GETFLAGS: {
1477                 struct ll_fid fid;
1478                 unsigned long valid = OBD_MD_FLFLAGS;
1479                 struct mds_body *body;
1480
1481                 ll_inode2fid(&fid, inode);
1482                 rc = md_getattr(sbi->ll_mdc_exp, &fid, valid, 0, &req);
1483                 if (rc) {
1484                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1485                         RETURN(-abs(rc));
1486                 }
1487
1488                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1489
1490                 if (body->flags & S_APPEND)
1491                         flags |= EXT3_APPEND_FL;
1492                 if (body->flags & S_IMMUTABLE)
1493                         flags |= EXT3_IMMUTABLE_FL;
1494                 if (body->flags & S_NOATIME)
1495                         flags |= EXT3_NOATIME_FL;
1496
1497                 ptlrpc_req_finished (req);
1498
1499                 RETURN(put_user(flags, (int *)arg));
1500         }
1501         case EXT3_IOC_SETFLAGS: {
1502                 struct mdc_op_data op_data;
1503                 struct iattr attr;
1504                 struct obdo *oa;
1505                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1506
1507                 if (get_user(flags, (int *)arg))
1508                         RETURN(-EFAULT);
1509
1510                 oa = obdo_alloc();
1511                 if (!oa)
1512                         RETURN(-ENOMEM);
1513
1514                 ll_prepare_mdc_op_data(&op_data, inode, NULL, NULL, 0, 0);
1515
1516                 memset(&attr, 0x0, sizeof(attr));
1517                 attr.ia_attr_flags = flags;
1518                 attr.ia_valid |= ATTR_ATTR_FLAG;
1519
1520                 rc = md_setattr(sbi->ll_mdc_exp, &op_data,
1521                                 &attr, NULL, 0, NULL, 0, &req);
1522                 if (rc) {
1523                         ptlrpc_req_finished(req);
1524                         if (rc != -EPERM && rc != -EACCES)
1525                                 CERROR("md_setattr fails: rc = %d\n", rc);
1526                         obdo_free(oa);
1527                         RETURN(rc);
1528                 }
1529                 ptlrpc_req_finished(req);
1530
1531                 oa->o_id = lsm->lsm_object_id;
1532                 oa->o_gr = lsm->lsm_object_gr;
1533                 oa->o_flags = flags;
1534                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1535
1536                 rc = obd_setattr(sbi->ll_osc_exp, oa, lsm, NULL);
1537                 obdo_free(oa);
1538                 if (rc) {
1539                         if (rc != -EPERM && rc != -EACCES)
1540                                 CERROR("md_setattr fails: rc = %d\n", rc);
1541                         RETURN(rc);
1542                 }
1543
1544                 if (flags & EXT3_APPEND_FL)
1545                         inode->i_flags |= S_APPEND;
1546                 else
1547                         inode->i_flags &= ~S_APPEND;
1548                 if (flags & EXT3_IMMUTABLE_FL)
1549                         inode->i_flags |= S_IMMUTABLE;
1550                 else
1551                         inode->i_flags &= ~S_IMMUTABLE;
1552                 if (flags & EXT3_NOATIME_FL)
1553                         inode->i_flags |= S_NOATIME;
1554                 else
1555                         inode->i_flags &= ~S_NOATIME;
1556
1557                 RETURN(0);
1558         }
1559         default:
1560                 RETURN(-ENOSYS);
1561         }
1562
1563         RETURN(0);
1564 }
1565
1566 void ll_umount_begin(struct super_block *sb)
1567 {
1568         struct ll_sb_info *sbi = ll_s2sbi(sb);
1569         struct obd_device *obd;
1570         struct obd_ioctl_data ioc_data = { 0 };
1571         ENTRY;
1572         CDEBUG(D_VFSTRACE, "VFS Op:\n");
1573
1574         obd = class_exp2obd(sbi->ll_mdc_exp);
1575         if (obd == NULL) {
1576                 CERROR("Invalid MDC connection handle "LPX64"\n",
1577                        sbi->ll_mdc_exp->exp_handle.h_cookie);
1578                 EXIT;
1579                 return;
1580         }
1581         obd->obd_no_recov = 1;
1582         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
1583                       &ioc_data, NULL);
1584
1585         obd = class_exp2obd(sbi->ll_osc_exp);
1586         if (obd == NULL) {
1587                 CERROR("Invalid LOV connection handle "LPX64"\n",
1588                        sbi->ll_osc_exp->exp_handle.h_cookie);
1589                 EXIT;
1590                 return;
1591         }
1592
1593         obd->obd_no_recov = 1;
1594         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
1595                       &ioc_data, NULL);
1596
1597         /* Really, we'd like to wait until there are no requests outstanding,
1598          * and then continue.  For now, we just invalidate the requests,
1599          * schedule, and hope.
1600          */
1601         schedule();
1602
1603         EXIT;
1604 }
1605
1606 int ll_prep_inode(struct obd_export *osc_exp, struct obd_export *mdc_exp,
1607                   struct inode **inode, struct ptlrpc_request *req,
1608                   int offset, struct super_block *sb)
1609 {
1610         struct lustre_md md;
1611         int rc = 0;
1612
1613         rc = mdc_req2lustre_md(mdc_exp, req, offset, osc_exp, &md);
1614         if (rc)
1615                 RETURN(rc);
1616
1617         if (*inode) {
1618                 ll_update_inode(*inode, &md);
1619         } else {
1620                 LASSERT(sb);
1621                 *inode = ll_iget(sb, md.body->ino, &md);
1622                 if (*inode == NULL || is_bad_inode(*inode)) {
1623                         /* free the lsm if we allocated one above */
1624                         if (md.lsm != NULL)
1625                                 obd_free_memmd(osc_exp, &md.lsm);
1626                         if (md.mea != NULL)
1627                                 obd_free_memmd(mdc_exp,
1628                                                (struct lov_stripe_md**)&md.mea);
1629                         rc = -ENOMEM;
1630                         CERROR("new_inode -fatal: rc %d\n", rc);
1631                 }
1632         }
1633
1634         RETURN(rc);
1635 }