lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Lustre Light Super operations
   5  *
   6  *  Copyright (c) 2002, 2003 Cluster File Systems, Inc.
   7  *
   8  *   This file is part of Lustre, http://www.lustre.org.
   9  *
  10  *   Lustre is free software; you can redistribute it and/or
  11  *   modify it under the terms of version 2 of the GNU General Public
  12  *   License as published by the Free Software Foundation.
  13  *
  14  *   Lustre is distributed in the hope that it will be useful,
  15  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  *   GNU General Public License for more details.
  18  *
  19  *   You should have received a copy of the GNU General Public License
  20  *   along with Lustre; if not, write to the Free Software
  21  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  22  */
  23
  24 #define DEBUG_SUBSYSTEM S_LLITE
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/random.h>
  29 #include <linux/version.h>
  30
  31 #include <linux/lustre_lite.h>
  32 #include <linux/lustre_ha.h>
  33 #include <linux/lustre_dlm.h>
  34 #include <linux/lprocfs_status.h>
  35 #include <linux/lustre_acl.h>
  36 #include "llite_internal.h"
  37
  38 kmem_cache_t *ll_file_data_slab;
  39 kmem_cache_t *ll_intent_slab;
  40
  41 extern struct address_space_operations ll_aops;
  42 extern struct address_space_operations ll_dir_aops;
  43
  44 #ifndef log2
  45 #define log2(n) ffz(~(n))
  46 #endif
  47
  48 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
  49 {
  50         struct ll_sb_info *sbi = NULL;
  51         class_uuid_t uuid;
  52         ENTRY;
  53
  54         OBD_ALLOC(sbi, sizeof(*sbi));
  55         if (!sbi)
  56                 RETURN(NULL);
  57
  58         spin_lock_init(&sbi->ll_lock);
  59         INIT_LIST_HEAD(&sbi->ll_pglist);
  60         sbi->ll_pglist_gen = 0;
  61         if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
  62                 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
  63         else
  64                 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
  65         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  66         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  67         INIT_LIST_HEAD(&sbi->ll_mnt_list);
  68
  69         sema_init(&sbi->ll_gns_sem, 1);
  70         spin_lock_init(&sbi->ll_gns_lock);
  71         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
  72         init_waitqueue_head(&sbi->ll_gns_waitq);
  73         init_completion(&sbi->ll_gns_mount_finished);
  74
  75         /* this later may be reset via /proc/fs/... */
  76         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
  77         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname) - 1] = '\0';
  78
  79         /* this later may be reset via /proc/fs/... */
  80         memset(sbi->ll_gns_upcall, 0, sizeof(sbi->ll_gns_upcall));
  81
  82         /* default values, may be changed via /proc/fs/... */
  83         sbi->ll_gns_state = LL_GNS_IDLE;
  84         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  85         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  86
  87         sbi->ll_gns_timer.data = (unsigned long)sbi;
  88         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  89         init_timer(&sbi->ll_gns_timer);
  90
  91         ll_set_sbi(sb, sbi);
  92
  93         generate_random_uuid(uuid);
  94         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  95         RETURN(sbi);
  96 }
  97
  98 void lustre_free_sbi(struct super_block *sb)
  99 {
 100         struct ll_sb_info *sbi = ll_s2sbi(sb);
 101         ENTRY;
 102
 103         if (sbi != NULL) {
 104                 list_del(&sbi->ll_gns_sbi_head);
 105                 del_timer(&sbi->ll_gns_timer);
 106                 OBD_FREE(sbi, sizeof(*sbi));
 107         }
 108         ll_set_sbi(sb, NULL);
 109         EXIT;
 110 }
 111
 112 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 113 {
 114         int valsize, rc;
 115         ENTRY;
 116
 117         valsize = sizeof(sbi->ll_dt_desc);
 118         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 119         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 120                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 121         RETURN(rc);
 122 }
 123
 124 extern struct dentry_operations ll_d_ops;
 125
 126 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 127                              char *security, __u32 *nllu, int async)
 128 {
 129         struct ll_sb_info *sbi = ll_s2sbi(sb);
 130         struct ptlrpc_request *request = NULL;
 131         struct lustre_handle dt_conn = {0, };
 132         struct lustre_handle md_conn = {0, };
 133         struct inode *root = NULL;
 134         struct obd_device *obd;
 135         struct obd_statfs osfs;
 136         struct lustre_md md;
 137         kdev_t devno;
 138         int err;
 139         ENTRY;
 140
 141         obd = class_name2obd(lmv);
 142         if (!obd) {
 143                 CERROR("MDC %s: not setup or attached\n", lmv);
 144                 RETURN(-EINVAL);
 145         }
 146         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 147                      sizeof(async), &async);
 148
 149         if (security == NULL)
 150                 security = "null";
 151
 152         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 153                            strlen(security), security);
 154         if (err) {
 155                 CERROR("LMV %s: failed to set security %s, err %d\n",
 156                         lmv, security, err);
 157                 RETURN(err);
 158         }
 159
 160         err = obd_set_info(obd->obd_self_export, strlen("nllu"), "nllu",
 161                            sizeof(__u32) * 2, nllu);
 162         if (err) {
 163                 CERROR("LMV %s: failed to set NLLU, err %d\n",
 164                         lmv, err);
 165                 RETURN(err);
 166         }
 167
 168         if (proc_lustre_fs_root) {
 169                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 170                                                   lov, lmv);
 171                 if (err < 0)
 172                         CERROR("could not register mount in /proc/lustre");
 173         }
 174
 175         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
 176         if (err == -EBUSY) {
 177                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 178                        " client is not a part.  Please wait for recovery to "
 179                        "complete, abort, or time out.\n", lmv);
 180                 GOTO(out, err);
 181         } else if (err) {
 182                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 183                 GOTO(out, err);
 184         }
 185         sbi->ll_md_exp = class_conn2export(&md_conn);
 186         err = obd_statfs(obd, &osfs, jiffies - HZ);
 187         if (err)
 188                 GOTO(out_lmv, err);
 189
 190         if (!osfs.os_bsize) {
 191                 CERROR("Invalid block size is detected.");
 192                 GOTO(out_lmv, err);
 193         }
 194
 195         sb->s_magic = LL_SUPER_MAGIC;
 196         sb->s_blocksize = osfs.os_bsize;
 197         sb->s_blocksize_bits = log2(osfs.os_bsize);
 198         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 199
 200         devno = get_uuid2int(sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 201                              strlen(sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 202
 203         sb->s_dev = devno;
 204
 205         obd = class_name2obd(lov);
 206         if (!obd) {
 207                 CERROR("OSC %s: not setup or attached\n", lov);
 208                 GOTO(out_lmv, err);
 209         }
 210         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 211                      sizeof(async), &async);
 212
 213         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, OBD_OPT_REAL_CLIENT);
 214         if (err == -EBUSY) {
 215                 CERROR("An OST (lov %s) is performing recovery, of which this"
 216                        " client is not a part.  Please wait for recovery to "
 217                        "complete, abort, or time out.\n", lov);
 218                 GOTO(out, err);
 219         } else if (err) {
 220                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 221                 GOTO(out_lmv, err);
 222         }
 223         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 224
 225         err = lustre_init_dt_desc(sbi);
 226         if (err == 0) {
 227                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 228                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
 229                                  sbi->ll_dt_desc.ld_tgt_count *
 230                                  sizeof(struct llog_cookie));
 231         }
 232
 233         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 234         if (err) {
 235                 CERROR("cannot mds_connect: rc = %d\n", err);
 236                 GOTO(out_lov, err);
 237         }
 238         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 239
 240         sb->s_op = &lustre_super_operations;
 241
 242         /* make root inode
 243          * XXX: move this to after cbd setup? */
 244         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 245                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID), NULL, 0,
 246                          0, &request);
 247         if (err) {
 248                 CERROR("md_getattr failed for root: rc = %d\n", err);
 249                 GOTO(out_lov, err);
 250         }
 251
 252         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 253                                 sbi->ll_dt_exp, &md);
 254         if (err) {
 255                 CERROR("failed to understand root inode md: rc = %d\n", err);
 256                 ptlrpc_req_finished(request);
 257                 GOTO(out_lov, err);
 258         }
 259
 260         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 261         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 262
 263         ptlrpc_req_finished(request);
 264
 265         if (root == NULL || is_bad_inode(root)) {
 266                 CERROR("lustre_lite: bad iget4 for root\n");
 267                 GOTO(out_root, err = -EBADF);
 268         }
 269
 270         err = ll_close_thread_start(&sbi->ll_lcq);
 271         if (err) {
 272                 CERROR("cannot start close thread: rc %d\n", err);
 273                 GOTO(out_root, err);
 274         }
 275
 276         ll_gns_add_timer(sbi);
 277
 278         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 279            backing dev info assigned to inode mapping is used for
 280            determining maximal readahead. */
 281 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 282     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 283         /* bug 2805 - set VM readahead to zero */
 284         vm_max_readahead = vm_min_readahead = 0;
 285 #endif
 286
 287         sb->s_root = d_alloc_root(root);
 288         sb->s_root->d_op = &ll_d_ops;
 289
 290 #ifdef S_PDIROPS
 291         CWARN("Enabling PDIROPS\n");
 292         sb->s_flags |= S_PDIROPS;
 293 #endif
 294
 295         RETURN(err);
 296 out_root:
 297         if (root)
 298                 iput(root);
 299 out_lov:
 300         obd_disconnect(sbi->ll_dt_exp, 0);
 301 out_lmv:
 302         obd_disconnect(sbi->ll_md_exp, 0);
 303 out:
 304         lprocfs_unregister_mountpoint(sbi);
 305         return err;
 306 }
 307
 308 void lustre_common_put_super(struct super_block *sb)
 309 {
 310         struct ll_sb_info *sbi = ll_s2sbi(sb);
 311         struct hlist_node *tmp, *next;
 312         ENTRY;
 313
 314         ll_gns_del_timer(sbi);
 315         ll_close_thread_shutdown(sbi->ll_lcq);
 316
 317         list_del(&sbi->ll_conn_chain);
 318         obd_disconnect(sbi->ll_dt_exp, 0);
 319
 320         lprocfs_unregister_mountpoint(sbi);
 321         if (sbi->ll_proc_root) {
 322                 lprocfs_remove(sbi->ll_proc_root);
 323                 sbi->ll_proc_root = NULL;
 324         }
 325
 326         obd_disconnect(sbi->ll_md_exp, 0);
 327
 328         // We do this to get rid of orphaned dentries. That is not really trw.
 329         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 330                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 331                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 332                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 333                 shrink_dcache_parent(dentry);
 334         }
 335         EXIT;
 336 }
 337
 338 char *ll_read_opt(const char *opt, char *data)
 339 {
 340         char *value;
 341         char *retval;
 342         ENTRY;
 343
 344         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 345         if (strncmp(opt, data, strlen(opt)))
 346                 RETURN(NULL);
 347         if ((value = strchr(data, '=')) == NULL)
 348                 RETURN(NULL);
 349
 350         value++;
 351         OBD_ALLOC(retval, strlen(value) + 1);
 352         if (!retval) {
 353                 CERROR("out of memory!\n");
 354                 RETURN(NULL);
 355         }
 356
 357         memcpy(retval, value, strlen(value)+1);
 358         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 359         RETURN(retval);
 360 }
 361
 362 int ll_set_opt(const char *opt, char *data, int fl)
 363 {
 364         ENTRY;
 365
 366         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 367         if (strncmp(opt, data, strlen(opt)))
 368                 RETURN(0);
 369         else
 370                 RETURN(fl);
 371 }
 372
 373 void ll_options(char *options, char **lov, char **lmv, char **sec,
 374                 int *async, int *flags)
 375 {
 376         char *this_char;
 377 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 378         char *opt_ptr = options;
 379 #endif
 380         ENTRY;
 381
 382         if (!options) {
 383                 EXIT;
 384                 return;
 385         }
 386
 387         *async = 0;
 388 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 389         for (this_char = strtok (options, ",");
 390              this_char != NULL;
 391              this_char = strtok (NULL, ",")) {
 392 #else
 393         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 394 #endif
 395                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 396                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 397                         continue;
 398                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 399                         continue;
 400                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 401                         *async = 1;
 402                         continue;
 403                 }
 404                 if (!*sec && (*sec = ll_read_opt("sec", this_char)))
 405                         continue;
 406                 if (!(*flags & LL_SBI_NOLCK) &&
 407                     ((*flags) = (*flags) |
 408                                 ll_set_opt("nolock", this_char,
 409                                            LL_SBI_NOLCK)))
 410                         continue;
 411         }
 412
 413         EXIT;
 414 }
 415
 416 void ll_lli_init(struct ll_inode_info *lli)
 417 {
 418         sema_init(&lli->lli_open_sem, 1);
 419         sema_init(&lli->lli_size_sem, 1);
 420         lli->lli_flags = 0;
 421         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 422         spin_lock_init(&lli->lli_lock);
 423         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 424         lli->lli_inode_magic = LLI_INODE_MAGIC;
 425         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 426         sema_init(&lli->lli_och_sem, 1);
 427         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 428         lli->lli_mds_exec_och = NULL;
 429         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 430         lli->lli_open_fd_exec_count = 0;
 431 }
 432
 433 int ll_fill_super(struct super_block *sb, void *data, int silent)
 434 {
 435         struct ll_sb_info *sbi;
 436         char *lov = NULL;
 437         char *lmv = NULL;
 438         int async, err;
 439         char *sec = NULL;
 440         __u32 nllu[2] = { 99, 99 };
 441         ENTRY;
 442
 443         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 444
 445         sbi = lustre_init_sbi(sb);
 446         if (!sbi)
 447                 RETURN(-ENOMEM);
 448
 449         sbi->ll_flags |= LL_SBI_READAHEAD;
 450         ll_options(data, &lov, &lmv, &sec, &async, &sbi->ll_flags);
 451
 452         if (!lov) {
 453                 CERROR("no osc\n");
 454                 GOTO(out, err = -EINVAL);
 455         }
 456
 457         if (!lmv) {
 458                 CERROR("no mdc\n");
 459                 GOTO(out, err = -EINVAL);
 460         }
 461
 462         err = lustre_common_fill_super(sb, lmv, lov, sec, nllu, async);
 463         EXIT;
 464 out:
 465         if (err)
 466                 lustre_free_sbi(sb);
 467
 468         if (sec)
 469                 OBD_FREE(sec, strlen(sec) + 1);
 470         if (lmv)
 471                 OBD_FREE(lmv, strlen(lmv) + 1);
 472         if (lov)
 473                 OBD_FREE(lov, strlen(lov) + 1);
 474         return err;
 475 } /* ll_read_super */
 476
 477 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 478                               struct config_llog_instance *cfg, int allow_recov)
 479 {
 480         struct lustre_cfg lcfg;
 481         struct portals_cfg pcfg;
 482         char *peer = "MDS_PEER_UUID";
 483         struct obd_device *obd;
 484         struct lustre_handle md_conn = {0, };
 485         struct obd_export *exp;
 486         char *name = "mdc_dev";
 487         class_uuid_t uuid;
 488         struct obd_uuid lmv_uuid;
 489         struct llog_ctxt *ctxt;
 490         int rc, err = 0;
 491         ENTRY;
 492
 493         if (lmd_bad_magic(lmd))
 494                 RETURN(-EINVAL);
 495
 496         generate_random_uuid(uuid);
 497         class_uuid_unparse(uuid, &lmv_uuid);
 498
 499         if (lmd->lmd_local_nid) {
 500                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 501                 pcfg.pcfg_nal = lmd->lmd_nal;
 502                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 503                 rc = libcfs_nal_cmd(&pcfg);
 504                 if (rc < 0)
 505                         GOTO(out, rc);
 506         }
 507
 508         if (lmd->lmd_nal == SOCKNAL ||
 509             lmd->lmd_nal == OPENIBNAL ||
 510             lmd->lmd_nal == IIBNAL ||
 511             lmd->lmd_nal == VIBNAL ||
 512             lmd->lmd_nal == RANAL) {
 513                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 514                 pcfg.pcfg_nal     = lmd->lmd_nal;
 515                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 516                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 517                 pcfg.pcfg_misc    = lmd->lmd_port;
 518                 rc = libcfs_nal_cmd(&pcfg);
 519                 if (rc < 0)
 520                         GOTO(out, rc);
 521         }
 522
 523         LCFG_INIT(lcfg, LCFG_ADD_UUID, name);
 524         lcfg.lcfg_nid = lmd->lmd_server_nid;
 525         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 526         lcfg.lcfg_inlbuf1 = peer;
 527         lcfg.lcfg_nal = lmd->lmd_nal;
 528         rc = class_process_config(&lcfg);
 529         if (rc < 0)
 530                 GOTO(out_del_conn, rc);
 531
 532         LCFG_INIT(lcfg, LCFG_ATTACH, name);
 533         lcfg.lcfg_inlbuf1 = "mdc";
 534         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 535         lcfg.lcfg_inlbuf2 = lmv_uuid.uuid;
 536         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 537         err = class_process_config(&lcfg);
 538         if (rc < 0)
 539                 GOTO(out_del_uuid, rc);
 540
 541         LCFG_INIT(lcfg, LCFG_SETUP, name);
 542         lcfg.lcfg_inlbuf1 = lmd->lmd_mds;
 543         lcfg.lcfg_inllen1 = strlen(lcfg.lcfg_inlbuf1) + 1;
 544         lcfg.lcfg_inlbuf2 = peer;
 545         lcfg.lcfg_inllen2 = strlen(lcfg.lcfg_inlbuf2) + 1;
 546         rc = class_process_config(&lcfg);
 547         if (rc < 0)
 548                 GOTO(out_detach, rc);
 549
 550         obd = class_name2obd(name);
 551         if (obd == NULL)
 552                 GOTO(out_cleanup, rc = -EINVAL);
 553
 554         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 555                           strlen(lmd->lmd_security), lmd->lmd_security);
 556         if (rc)
 557                 GOTO(out_cleanup, rc);
 558
 559         /* Disable initial recovery on this import */
 560         rc = obd_set_info(obd->obd_self_export,
 561                           strlen("initial_recov"), "initial_recov",
 562                           sizeof(allow_recov), &allow_recov);
 563         if (rc)
 564                 GOTO(out_cleanup, rc);
 565
 566         rc = obd_connect(&md_conn, obd, &lmv_uuid, 0);
 567         if (rc) {
 568                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 569                 GOTO(out_cleanup, rc);
 570         }
 571
 572         exp = class_conn2export(&md_conn);
 573
 574         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 575         rc = class_config_process_llog(ctxt, profile, cfg);
 576         if (rc)
 577                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 578
 579         rc = obd_disconnect(exp, 0);
 580
 581         EXIT;
 582 out_cleanup:
 583         LCFG_INIT(lcfg, LCFG_CLEANUP, name);
 584         err = class_process_config(&lcfg);
 585         if (err < 0)
 586                 GOTO(out, err);
 587
 588 out_detach:
 589         LCFG_INIT(lcfg, LCFG_DETACH, name);
 590         err = class_process_config(&lcfg);
 591         if (err < 0)
 592                 GOTO(out, err);
 593
 594 out_del_uuid:
 595         LCFG_INIT(lcfg, LCFG_DEL_UUID, name);
 596         lcfg.lcfg_inllen1 = strlen(peer) + 1;
 597         lcfg.lcfg_inlbuf1 = peer;
 598         err = class_process_config(&lcfg);
 599
 600 out_del_conn:
 601         if (lmd->lmd_nal == SOCKNAL ||
 602             lmd->lmd_nal == OPENIBNAL ||
 603             lmd->lmd_nal == IIBNAL ||
 604             lmd->lmd_nal == VIBNAL ||
 605             lmd->lmd_nal == RANAL) {
 606                 int err2;
 607
 608                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 609                 pcfg.pcfg_nal     = lmd->lmd_nal;
 610                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 611                 pcfg.pcfg_flags   = 1;          /* single_share */
 612                 err2 = libcfs_nal_cmd(&pcfg);
 613                 if (err2 && !err)
 614                         err = err2;
 615                 if (err < 0)
 616                         GOTO(out, err);
 617         }
 618 out:
 619         if (rc == 0)
 620                 rc = err;
 621
 622         return rc;
 623 }
 624
 625 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 626 {
 627         struct lustre_cfg lcfg;
 628         struct obd_device *obd;
 629         int next = 0;
 630
 631         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 632         {
 633                 int err;
 634
 635                 LCFG_INIT(lcfg, LCFG_CLEANUP, obd->obd_name);
 636                 err = class_process_config(&lcfg);
 637                 if (err) {
 638                         CERROR("cleanup failed: %s\n", obd->obd_name);
 639                         //continue;
 640                 }
 641
 642                 LCFG_INIT(lcfg, LCFG_DETACH, obd->obd_name);
 643                 err = class_process_config(&lcfg);
 644                 if (err) {
 645                         CERROR("detach failed: %s\n", obd->obd_name);
 646                         //continue;
 647                 }
 648         }
 649
 650         if (sbi->ll_lmd != NULL)
 651                 class_del_profile(sbi->ll_lmd->lmd_profile);
 652 }
 653
 654 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 655 {
 656         struct lustre_mount_data * lmd = data;
 657         char *lov = NULL, *lmv = NULL;
 658         struct ll_sb_info *sbi;
 659         int err;
 660         ENTRY;
 661
 662         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 663         if (lmd_bad_magic(lmd))
 664                 RETURN(-EINVAL);
 665
 666         sbi = lustre_init_sbi(sb);
 667         if (!sbi)
 668                 RETURN(-ENOMEM);
 669
 670         sbi->ll_flags |= LL_SBI_READAHEAD;
 671
 672         if (lmd->lmd_profile) {
 673                 struct lustre_profile *lprof;
 674                 struct config_llog_instance cfg;
 675                 int len;
 676
 677                 if (lmd->lmd_mds[0] == '\0') {
 678                         CERROR("no mds name\n");
 679                         GOTO(out_free, err = -EINVAL);
 680                 }
 681                 lmd->lmd_security[sizeof(lmd->lmd_security) - 1] = 0;
 682
 683                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 684                 if (sbi->ll_lmd == NULL)
 685                         GOTO(out_free, err = -ENOMEM);
 686                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 687
 688                 /* generate a string unique to this super, let's try
 689                  the address of the super itself.*/
 690                 len = (sizeof(sb) * 2) + 1;
 691                 OBD_ALLOC(sbi->ll_instance, len);
 692                 if (sbi->ll_instance == NULL)
 693                         GOTO(out_free, err = -ENOMEM);
 694                 sprintf(sbi->ll_instance, "%p", sb);
 695
 696                 cfg.cfg_instance = sbi->ll_instance;
 697                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 698                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 699                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 700                 if (err < 0) {
 701                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 702                         GOTO(out_free, err);
 703                 }
 704
 705                 lprof = class_get_profile(lmd->lmd_profile);
 706                 if (lprof == NULL) {
 707                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 708                         GOTO(out_free, err = -EINVAL);
 709                 }
 710                 if (lov)
 711                         OBD_FREE(lov, strlen(lov) + 1);
 712                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
 713                           strlen(sbi->ll_instance) + 2);
 714                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 715
 716                 if (lmv)
 717                         OBD_FREE(lmv, strlen(lmv) + 1);
 718                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
 719                           strlen(sbi->ll_instance) + 2);
 720                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 721         }
 722
 723         if (!lov) {
 724                 CERROR("no osc\n");
 725                 GOTO(out_free, err = -EINVAL);
 726         }
 727
 728         if (!lmv) {
 729                 CERROR("no mdc\n");
 730                 GOTO(out_free, err = -EINVAL);
 731         }
 732
 733         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_security,
 734                                        &lmd->lmd_nllu, lmd->lmd_async);
 735
 736         if (err)
 737                 GOTO(out_free, err);
 738
 739 out_dev:
 740         if (lmv)
 741                 OBD_FREE(lmv, strlen(lmv) + 1);
 742         if (lov)
 743                 OBD_FREE(lov, strlen(lov) + 1);
 744
 745         RETURN(err);
 746
 747 out_free:
 748         if (sbi->ll_lmd) {
 749                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 750                 int err;
 751
 752                 if (sbi->ll_instance != NULL) {
 753                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 754                         struct config_llog_instance cfg;
 755                         char *cl_prof;
 756
 757                         cfg.cfg_instance = sbi->ll_instance;
 758                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 759
 760                         OBD_ALLOC(cl_prof, len);
 761                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 762                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 763                         if (err < 0) {
 764                                 CERROR("Unable to process log: %s\n", cl_prof);
 765                                 lustre_manual_cleanup(sbi);
 766                         }
 767                         OBD_FREE(cl_prof, len);
 768                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 769                 }
 770                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 771         }
 772         lustre_free_sbi(sb);
 773         goto out_dev;
 774 } /* lustre_fill_super */
 775
 776 void lustre_put_super(struct super_block *sb)
 777 {
 778         struct obd_device *obd;
 779         struct ll_sb_info *sbi = ll_s2sbi(sb);
 780         int force_umount = 0;
 781         ENTRY;
 782
 783         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 784         obd = class_exp2obd(sbi->ll_md_exp);
 785         if (obd)
 786                 force_umount = obd->obd_no_recov;
 787         obd = NULL;
 788
 789         lustre_common_put_super(sb);
 790         if (sbi->ll_lmd != NULL) {
 791                 char *cl_prof;
 792                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 793                 int err;
 794                 struct config_llog_instance cfg;
 795
 796                 if (force_umount) {
 797                         CERROR("force umount, doing manual cleanup\n");
 798                         lustre_manual_cleanup(sbi);
 799                         GOTO(free_lmd, 0);
 800                 }
 801
 802                 cfg.cfg_instance = sbi->ll_instance;
 803                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 804
 805                 OBD_ALLOC(cl_prof, len);
 806                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 807                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
 808                 if (err < 0) {
 809                         CERROR("Unable to process log: %s, doing manual cleanup"
 810                                "\n", cl_prof);
 811                         lustre_manual_cleanup(sbi);
 812                 }
 813
 814                 OBD_FREE(cl_prof, len);
 815         free_lmd:
 816                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 817                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 818         }
 819
 820         lustre_free_sbi(sb);
 821
 822         EXIT;
 823 } /* lustre_put_super */
 824
 825 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 826 {
 827         struct obd_export *md_exp = sbi->ll_md_exp;
 828         struct lustre_mount_data *lmd = sbi->ll_lmd;
 829         struct llog_ctxt *ctxt;
 830         struct config_llog_instance cfg;
 831         char *profile = lmd->lmd_profile, *name = NULL;
 832         int rc, namelen =  0, version;
 833         ENTRY;
 834
 835         if (profile == NULL)
 836                 RETURN(0);
 837         if (lmd == NULL) {
 838                 CERROR("Client not mounted with zero-conf; cannot process "
 839                        "update log.\n");
 840                 RETURN(0);
 841         }
 842
 843         rc = ldlm_cli_cancel_unused(md_exp->exp_obd->obd_namespace, NULL,
 844                                     LDLM_FL_CONFIG_CHANGE, NULL);
 845         if (rc != 0)
 846                 CWARN("ldlm_cli_cancel_unused(mdc): %d\n", rc);
 847
 848         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL, LDLM_FL_CONFIG_CHANGE,
 849                                NULL);
 850         if (rc != 0)
 851                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 852
 853         cfg.cfg_instance = sbi->ll_instance;
 854         cfg.cfg_uuid = sbi->ll_sb_uuid;
 855         cfg.cfg_local_nid = lmd->lmd_local_nid;
 856
 857         namelen = strlen(profile) + 20; /* -clean-######### */
 858         OBD_ALLOC(name, namelen);
 859         if (name == NULL)
 860                 RETURN(-ENOMEM);
 861
 862         if (clean) {
 863                 version = sbi->ll_config_version - 1;
 864                 sprintf(name, "%s-clean-%d", profile, version);
 865         } else {
 866                 version = sbi->ll_config_version + 1;
 867                 sprintf(name, "%s-%d", profile, version);
 868         }
 869
 870         CWARN("Applying configuration log %s\n", name);
 871
 872         ctxt = llog_get_context(&md_exp->exp_obd->obd_llogs,
 873                                 LLOG_CONFIG_REPL_CTXT);
 874         rc = class_config_process_llog(ctxt, name, &cfg);
 875         if (rc == 0)
 876                 sbi->ll_config_version = version;
 877         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 878
 879         if (rc == 0 && clean == 0) {
 880                 struct lov_desc desc;
 881                 int rc, valsize;
 882                 valsize = sizeof(desc);
 883                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 884                                   "lovdesc", &valsize, &desc);
 885
 886                 rc = obd_init_ea_size(md_exp,
 887                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
 888                                       (desc.ld_tgt_count *
 889                                        sizeof(struct llog_cookie)));
 890         }
 891         OBD_FREE(name, namelen);
 892         RETURN(rc);
 893 }
 894
 895 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 896 {
 897         struct inode *inode = NULL;
 898         l_lock(&lock->l_resource->lr_namespace->ns_lock);
 899         if (lock->l_ast_data) {
 900                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 901                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 902                         inode = igrab(lock->l_ast_data);
 903                 } else {
 904                         inode = lock->l_ast_data;
 905                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
 906                                "l_ast_data %p is bogus: magic %0x8\n",
 907                                lock->l_ast_data, lli->lli_inode_magic);
 908                         inode = NULL;
 909                 }
 910         }
 911         l_unlock(&lock->l_resource->lr_namespace->ns_lock);
 912         return inode;
 913 }
 914
 915 int null_if_equal(struct ldlm_lock *lock, void *data)
 916 {
 917         if (data == lock->l_ast_data) {
 918                 lock->l_ast_data = NULL;
 919
 920                 if (lock->l_req_mode != lock->l_granted_mode)
 921                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
 922         }
 923
 924         return LDLM_ITER_CONTINUE;
 925 }
 926
 927 void ll_clear_inode(struct inode *inode)
 928 {
 929         struct lustre_id id;
 930         struct ll_inode_info *lli = ll_i2info(inode);
 931         struct ll_sb_info *sbi = ll_i2sbi(inode);
 932         ENTRY;
 933
 934         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
 935                inode->i_generation, inode);
 936
 937         lli->lli_inode_magic = LLI_INODE_DEAD;
 938         ll_inode2id(&id, inode);
 939
 940         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
 941         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
 942
 943         LASSERT(!lli->lli_open_fd_write_count);
 944         LASSERT(!lli->lli_open_fd_read_count);
 945         LASSERT(!lli->lli_open_fd_exec_count);
 946
 947         if (lli->lli_mds_write_och)
 948                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
 949         if (lli->lli_mds_exec_och)
 950                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
 951         if (lli->lli_mds_read_och)
 952                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
 953
 954         if (lli->lli_smd)
 955                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
 956                                   null_if_equal, inode);
 957
 958         if (lli->lli_smd) {
 959                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
 960                 lli->lli_smd = NULL;
 961         }
 962
 963         if (lli->lli_mea) {
 964                 obd_free_memmd(sbi->ll_md_exp,
 965                                (struct lov_stripe_md **) &lli->lli_mea);
 966                 lli->lli_mea = NULL;
 967         }
 968
 969         if (lli->lli_symlink_name) {
 970                 OBD_FREE(lli->lli_symlink_name,
 971                          strlen(lli->lli_symlink_name) + 1);
 972                 lli->lli_symlink_name = NULL;
 973         }
 974         lli->lli_inode_magic = LLI_INODE_DEAD;
 975
 976         EXIT;
 977 }
 978
 979 /* If this inode has objects allocated to it (lsm != NULL), then the OST
 980  * object(s) determine the file size and mtime.  Otherwise, the MDS will
 981  * keep these values until such a time that objects are allocated for it.
 982  * We do the MDS operations first, as it is checking permissions for us.
 983  * We don't to the MDS RPC if there is nothing that we want to store there,
 984  * otherwise there is no harm in updating mtime/atime on the MDS if we are
 985  * going to do an RPC anyways.
 986  *
 987  * If we are doing a truncate, we will send the mtime and ctime updates
 988  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
 989  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
 990  * at the same time.
 991  */
 992 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
 993 {
 994         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
 995         struct ll_sb_info *sbi = ll_i2sbi(inode);
 996         struct ptlrpc_request *request = NULL;
 997         struct mdc_op_data *op_data;
 998         int ia_valid = attr->ia_valid;
 999         int rc = 0;
1000         ENTRY;
1001
1002         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1003         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1004
1005         if (ia_valid & ATTR_SIZE) {
1006                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1007                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1008                                attr->ia_size, ll_file_maxbytes(inode));
1009                         RETURN(-EFBIG);
1010                 }
1011
1012                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1013         }
1014
1015         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1016         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1017                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1018                         RETURN(-EPERM);
1019         }
1020
1021         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1022         if (attr->ia_valid & ATTR_CTIME) {
1023                 attr->ia_ctime = CURRENT_TIME;
1024                 attr->ia_valid |= ATTR_CTIME_SET;
1025         }
1026         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1027                 attr->ia_atime = CURRENT_TIME;
1028                 attr->ia_valid |= ATTR_ATIME_SET;
1029         }
1030         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1031                 attr->ia_mtime = CURRENT_TIME;
1032                 attr->ia_valid |= ATTR_MTIME_SET;
1033         }
1034
1035         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1036                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1037                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1038                        LTIME_S(CURRENT_TIME));
1039         if (lsm)
1040                 attr->ia_valid &= ~ATTR_SIZE;
1041
1042         /* If only OST attributes being set on objects, don't do MDS RPC.
1043          * In that case, we need to check permissions and update the local
1044          * inode ourselves so we can call obdo_from_inode() always. */
1045         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1046                 struct lustre_md md;
1047                 int save_valid;
1048
1049                 OBD_ALLOC(op_data, sizeof(*op_data));
1050                 if (op_data == NULL)
1051                         RETURN(-ENOMEM);
1052                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1053
1054                 rc = md_setattr(sbi->ll_md_exp, op_data,
1055                                 attr, NULL, 0, NULL, 0, &request);
1056                 OBD_FREE(op_data, sizeof(*op_data));
1057                 if (rc) {
1058                         ptlrpc_req_finished(request);
1059                         if (rc != -EPERM && rc != -EACCES)
1060                                 CERROR("md_setattr fails: rc = %d\n", rc);
1061                         RETURN(rc);
1062                 }
1063
1064                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1065                                        sbi->ll_dt_exp, &md);
1066                 if (rc) {
1067                         ptlrpc_req_finished(request);
1068                         RETURN(rc);
1069                 }
1070
1071                 /* We call inode_setattr to adjust timestamps, but we first
1072                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1073                  *
1074                  * NB: ATTR_SIZE will only be set at this point if the size
1075                  * resides on the MDS, ie, this file has no objects. */
1076                 save_valid = attr->ia_valid;
1077                 attr->ia_valid &= ~ATTR_SIZE;
1078                 inode_setattr(inode, attr);
1079                 attr->ia_valid = save_valid;
1080
1081                 ll_update_inode(inode, &md);
1082                 ptlrpc_req_finished(request);
1083
1084                 if (!lsm || !S_ISREG(inode->i_mode)) {
1085                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1086                         RETURN(0);
1087                 }
1088         } else {
1089                 /* The OST doesn't check permissions, but the alternative is
1090                  * a gratuitous RPC to the MDS.  We already rely on the client
1091                  * to do read/write/truncate permission checks, so is mtime OK?
1092                  */
1093                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1094                         /* from sys_utime() */
1095                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1096                                 if (current->fsuid != inode->i_uid &&
1097                                     (rc=ll_permission(inode,MAY_WRITE,NULL))!=0)
1098                                         RETURN(rc);
1099                         } else {
1100                                 /* from inode_change_ok() */
1101                                 if (current->fsuid != inode->i_uid &&
1102                                     !capable(CAP_FOWNER))
1103                                         RETURN(-EPERM);
1104                         }
1105                 }
1106
1107                 /* Won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1108                 inode_setattr(inode, attr);
1109         }
1110
1111         /* We really need to get our PW lock before we change inode->i_size.
1112          * If we don't we can race with other i_size updaters on our node, like
1113          * ll_file_read.  We can also race with i_size propogation to other
1114          * nodes through dirtying and writeback of final cached pages.  This
1115          * last one is especially bad for racing o_append users on other
1116          * nodes. */
1117         if (ia_valid & ATTR_SIZE) {
1118                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1119                                                            OBD_OBJECT_EOF } };
1120                 struct lustre_handle lockh = { 0 };
1121                 struct ll_inode_info *lli = ll_i2info(inode);
1122                 int err, ast_flags = 0;
1123                 /* XXX when we fix the AST intents to pass the discard-range
1124                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1125                  * XXX here. */
1126                 if (attr->ia_size == 0)
1127                         ast_flags = LDLM_AST_DISCARD_DATA;
1128
1129                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1130                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1131
1132                 if (rc != 0)
1133                         RETURN(rc);
1134
1135                 down(&lli->lli_size_sem);
1136                 rc = vmtruncate(inode, attr->ia_size);
1137                 if (rc != 0) {
1138                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1139                         up(&lli->lli_size_sem);
1140                 }
1141
1142                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1143                 if (err) {
1144                         CERROR("ll_extent_unlock failed: %d\n", err);
1145                         if (!rc)
1146                                 rc = err;
1147                 }
1148         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1149                 struct obdo *oa = NULL;
1150
1151                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1152                        inode->i_ino, LTIME_S(attr->ia_mtime));
1153
1154                 oa = obdo_alloc();
1155                 if (oa == NULL)
1156                         RETURN(-ENOMEM);
1157
1158                 oa->o_id = lsm->lsm_object_id;
1159                 oa->o_gr = lsm->lsm_object_gr;
1160                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1161                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1162                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1163                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1164                 obdo_free(oa);
1165                 if (rc)
1166                         CERROR("obd_setattr fails: rc = %d\n", rc);
1167         }
1168         RETURN(rc);
1169 }
1170
1171 int ll_setattr(struct dentry *de, struct iattr *attr)
1172 {
1173         LASSERT(de->d_inode);
1174         return ll_setattr_raw(de->d_inode, attr);
1175 }
1176
1177 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1178                        unsigned long max_age)
1179 {
1180         struct ll_sb_info *sbi = ll_s2sbi(sb);
1181         struct obd_statfs obd_osfs;
1182         int rc;
1183         ENTRY;
1184
1185         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1186         if (rc) {
1187                 CERROR("obd_statfs fails: rc = %d\n", rc);
1188                 RETURN(rc);
1189         }
1190
1191         osfs->os_type = sb->s_magic;
1192
1193         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1194                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1195
1196         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1197         if (rc) {
1198                 CERROR("obd_statfs fails: rc = %d\n", rc);
1199                 RETURN(rc);
1200         }
1201
1202         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1203                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1204                obd_osfs.os_files);
1205
1206         osfs->os_blocks = obd_osfs.os_blocks;
1207         osfs->os_bfree = obd_osfs.os_bfree;
1208         osfs->os_bavail = obd_osfs.os_bavail;
1209
1210         /* If we don't have as many objects free on the OST as inodes
1211          * on the MDS, we reduce the total number of inodes to
1212          * compensate, so that the "inodes in use" number is correct.
1213          */
1214         if (obd_osfs.os_ffree < osfs->os_ffree) {
1215                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1216                         obd_osfs.os_ffree;
1217                 osfs->os_ffree = obd_osfs.os_ffree;
1218         }
1219
1220         RETURN(rc);
1221 }
1222
1223 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1224 {
1225         struct obd_statfs osfs;
1226         int rc;
1227
1228         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1229         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1230
1231         /* For now we will always get up-to-date statfs values, but in the
1232          * future we may allow some amount of caching on the client (e.g.
1233          * from QOS or lprocfs updates). */
1234         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1235         if (rc)
1236                 return rc;
1237
1238         statfs_unpack(sfs, &osfs);
1239
1240         if (sizeof(sfs->f_blocks) == 4) {
1241                 while (osfs.os_blocks > ~0UL) {
1242                         sfs->f_bsize <<= 1;
1243
1244                         osfs.os_blocks >>= 1;
1245                         osfs.os_bfree >>= 1;
1246                         osfs.os_bavail >>= 1;
1247                 }
1248         }
1249
1250         sfs->f_blocks = osfs.os_blocks;
1251         sfs->f_bfree = osfs.os_bfree;
1252         sfs->f_bavail = osfs.os_bavail;
1253
1254         return 0;
1255 }
1256
1257 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1258 {
1259         struct ll_inode_info *lli = ll_i2info(inode);
1260         struct lov_stripe_md *lsm = md->lsm;
1261         struct mds_body *body = md->body;
1262         struct mea *mea = md->mea;
1263         struct posix_acl *ll_acl_access = md->acl_access;
1264         ENTRY;
1265
1266         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1267
1268         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1269                 /* check for default striping info for dir. */
1270                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1271         }
1272
1273         if (lsm != NULL) {
1274                 LASSERT(lsm->lsm_object_gr > 0);
1275                 if (lli->lli_smd == NULL) {
1276                         lli->lli_smd = lsm;
1277                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1278                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1279                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1280                 } else {
1281                         int i;
1282                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1283                                 CERROR("lsm mismatch for inode %ld\n",
1284                                        inode->i_ino);
1285                                 CERROR("lli_smd:\n");
1286                                 dump_lsm(D_ERROR, lli->lli_smd);
1287                                 CERROR("lsm:\n");
1288                                 dump_lsm(D_ERROR, lsm);
1289                                 LBUG();
1290                         }
1291                         /* XXX FIXME -- We should decide on a safer (atomic) and
1292                          * more elegant way to update the lsm */
1293                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1294                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1295                                         lsm->lsm_oinfo[i].loi_id;
1296                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1297                                         lsm->lsm_oinfo[i].loi_gr;
1298                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1299                                         lsm->lsm_oinfo[i].loi_ost_idx;
1300                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1301                                         lsm->lsm_oinfo[i].loi_ost_gen;
1302                         }
1303                 }
1304                 /* bug 2844 - limit i_blksize for broken user-space apps */
1305                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1306                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1307                 if (lli->lli_smd != lsm)
1308                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1309         }
1310
1311         if (mea != NULL) {
1312                 if (lli->lli_mea == NULL) {
1313                         lli->lli_mea = mea;
1314                 } else {
1315                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1316                                 CERROR("mea mismatch for inode %lu\n",
1317                                         inode->i_ino);
1318                                 LBUG();
1319                         }
1320                 }
1321                 if (lli->lli_mea != mea)
1322                         obd_free_memmd(ll_i2mdexp(inode),
1323                                        (struct lov_stripe_md **) &mea);
1324         }
1325
1326         if (body->valid & OBD_MD_FID)
1327                 id_assign_fid(&lli->lli_id, &body->id1);
1328
1329         if (body->valid & OBD_MD_FLID)
1330                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1331
1332         if (body->valid & OBD_MD_FLGENER)
1333                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1334
1335         spin_lock(&lli->lli_lock);
1336         if (ll_acl_access != NULL) {
1337                 if (lli->lli_acl_access != NULL)
1338                         posix_acl_release(lli->lli_acl_access);
1339                 lli->lli_acl_access = ll_acl_access;
1340         }
1341         spin_unlock(&lli->lli_lock);
1342
1343         if (body->valid & OBD_MD_FLID)
1344                 inode->i_ino = id_ino(&body->id1);
1345         if (body->valid & OBD_MD_FLGENER)
1346                 inode->i_generation = id_gen(&body->id1);
1347         if (body->valid & OBD_MD_FLATIME)
1348                 LTIME_S(inode->i_atime) = body->atime;
1349         if (body->valid & OBD_MD_FLMTIME &&
1350             body->mtime > LTIME_S(inode->i_mtime)) {
1351                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1352                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1353                 LTIME_S(inode->i_mtime) = body->mtime;
1354         }
1355         if (body->valid & OBD_MD_FLCTIME &&
1356             body->ctime > LTIME_S(inode->i_ctime))
1357                 LTIME_S(inode->i_ctime) = body->ctime;
1358         if (body->valid & OBD_MD_FLMODE) {
1359                 inode->i_mode = (inode->i_mode & S_IFMT) |
1360                         (body->mode & ~S_IFMT);
1361         }
1362         if (body->valid & OBD_MD_FLTYPE) {
1363                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1364                         (body->mode & S_IFMT);
1365         }
1366         if (body->valid & OBD_MD_FLUID)
1367                 inode->i_uid = body->uid;
1368         if (body->valid & OBD_MD_FLGID)
1369                 inode->i_gid = body->gid;
1370         if (body->valid & OBD_MD_FLFLAGS)
1371                 inode->i_flags = body->flags;
1372         if (body->valid & OBD_MD_FLNLINK)
1373                 inode->i_nlink = body->nlink;
1374         if (body->valid & OBD_MD_FLRDEV)
1375 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1376                 inode->i_rdev = body->rdev;
1377 #else
1378                 inode->i_rdev = old_decode_dev(body->rdev);
1379 #endif
1380         if (body->valid & OBD_MD_FLSIZE)
1381                 inode->i_size = body->size;
1382         if (body->valid & OBD_MD_FLBLOCKS)
1383                 inode->i_blocks = body->blocks;
1384
1385         if (body->valid & OBD_MD_FLSIZE)
1386                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1387
1388 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1389         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1390 #endif
1391         LASSERT(id_fid(&lli->lli_id) != 0);
1392 }
1393
1394 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1395 static struct backing_dev_info ll_backing_dev_info = {
1396         .ra_pages       = 0,    /* No readahead */
1397         .memory_backed  = 0,    /* Does contribute to dirty memory */
1398 };
1399 #endif
1400
1401 void ll_read_inode2(struct inode *inode, void *opaque)
1402 {
1403         struct lustre_md *md = opaque;
1404         struct ll_inode_info *lli = ll_i2info(inode);
1405         ENTRY;
1406
1407         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1408                inode->i_generation, inode);
1409
1410         ll_lli_init(lli);
1411
1412         LASSERT(!lli->lli_smd);
1413
1414         /* Core attributes from the MDS first.  This is a new inode, and
1415          * the VFS doesn't zero times in the core inode so we have to do
1416          * it ourselves.  They will be overwritten by either MDS or OST
1417          * attributes - we just need to make sure they aren't newer. */
1418         LTIME_S(inode->i_mtime) = 0;
1419         LTIME_S(inode->i_atime) = 0;
1420         LTIME_S(inode->i_ctime) = 0;
1421
1422         inode->i_rdev = 0;
1423         ll_update_inode(inode, md);
1424
1425         /* OIDEBUG(inode); */
1426
1427         if (S_ISREG(inode->i_mode)) {
1428                 inode->i_op = &ll_file_inode_operations;
1429                 inode->i_fop = &ll_file_operations;
1430                 inode->i_mapping->a_ops = &ll_aops;
1431                 EXIT;
1432         } else if (S_ISDIR(inode->i_mode)) {
1433                 inode->i_op = &ll_dir_inode_operations;
1434                 inode->i_fop = &ll_dir_operations;
1435                 inode->i_mapping->a_ops = &ll_dir_aops;
1436                 EXIT;
1437         } else if (S_ISLNK(inode->i_mode)) {
1438                 inode->i_op = &ll_fast_symlink_inode_operations;
1439                 EXIT;
1440         } else {
1441                 inode->i_op = &ll_special_inode_operations;
1442
1443 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1444                 init_special_inode(inode, inode->i_mode,
1445                                    kdev_t_to_nr(inode->i_rdev));
1446
1447                 /* initializing backing dev info. */
1448                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1449 #else
1450                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1451 #endif
1452                 lli->ll_save_ifop = inode->i_fop;
1453
1454                 if (S_ISCHR(inode->i_mode))
1455                         inode->i_fop = &ll_special_chr_inode_fops;
1456                 else if (S_ISBLK(inode->i_mode))
1457                         inode->i_fop = &ll_special_blk_inode_fops;
1458                 else if (S_ISFIFO(inode->i_mode))
1459                         inode->i_fop = &ll_special_fifo_inode_fops;
1460                 else if (S_ISSOCK(inode->i_mode))
1461                         inode->i_fop = &ll_special_sock_inode_fops;
1462
1463                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
1464                       inode->i_fop);
1465
1466                 if (lli->ll_save_ifop->owner) {
1467                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
1468                               lli->ll_save_ifop->owner);
1469                 }
1470                 EXIT;
1471         }
1472 }
1473
1474 void ll_delete_inode(struct inode *inode)
1475 {
1476         struct ll_sb_info *sbi = ll_i2sbi(inode);
1477         struct lustre_id id;
1478         int rc;
1479         ENTRY;
1480
1481         ll_inode2id(&id, inode);
1482
1483         rc = md_delete_inode(sbi->ll_md_exp, &id);
1484         if (rc) {
1485                 CERROR("md_delete_inode() failed, error %d\n",
1486                        rc);
1487         }
1488
1489         clear_inode(inode);
1490         EXIT;
1491 }
1492
1493 int ll_iocontrol(struct inode *inode, struct file *file,
1494                  unsigned int cmd, unsigned long arg)
1495 {
1496         struct ll_sb_info *sbi = ll_i2sbi(inode);
1497         struct ptlrpc_request *req = NULL;
1498         int rc, flags = 0;
1499         ENTRY;
1500
1501         switch(cmd) {
1502         case EXT3_IOC_GETFLAGS: {
1503                 struct lustre_id id;
1504                 __u64 valid = OBD_MD_FLFLAGS;
1505                 struct mds_body *body;
1506
1507                 ll_inode2id(&id, inode);
1508                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
1509                 if (rc) {
1510                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1511                         RETURN(-abs(rc));
1512                 }
1513
1514                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
1515
1516                 if (body->flags & S_APPEND)
1517                         flags |= EXT3_APPEND_FL;
1518                 if (body->flags & S_IMMUTABLE)
1519                         flags |= EXT3_IMMUTABLE_FL;
1520                 if (body->flags & S_NOATIME)
1521                         flags |= EXT3_NOATIME_FL;
1522
1523                 ptlrpc_req_finished (req);
1524
1525                 RETURN(put_user(flags, (int *)arg));
1526         }
1527         case EXT3_IOC_SETFLAGS: {
1528                 struct mdc_op_data *op_data;
1529                 struct iattr attr;
1530                 struct obdo *oa;
1531                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1532
1533                 if (get_user(flags, (int *)arg))
1534                         RETURN(-EFAULT);
1535
1536                 oa = obdo_alloc();
1537                 if (!oa)
1538                         RETURN(-ENOMEM);
1539
1540                 OBD_ALLOC(op_data, sizeof(*op_data));
1541                 if (op_data == NULL) {
1542                         obdo_free(oa);
1543                         RETURN(-ENOMEM);
1544                 }
1545                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1546
1547                 memset(&attr, 0x0, sizeof(attr));
1548                 attr.ia_attr_flags = flags;
1549                 attr.ia_valid |= ATTR_ATTR_FLAG;
1550
1551                 rc = md_setattr(sbi->ll_md_exp, op_data,
1552                                 &attr, NULL, 0, NULL, 0, &req);
1553                 OBD_FREE(op_data, sizeof(*op_data));
1554                 if (rc) {
1555                         ptlrpc_req_finished(req);
1556                         if (rc != -EPERM && rc != -EACCES)
1557                                 CERROR("md_setattr fails: rc = %d\n", rc);
1558                         obdo_free(oa);
1559                         RETURN(rc);
1560                 }
1561                 ptlrpc_req_finished(req);
1562
1563                 oa->o_id = lsm->lsm_object_id;
1564                 oa->o_gr = lsm->lsm_object_gr;
1565                 oa->o_flags = flags;
1566                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
1567
1568                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1569                 obdo_free(oa);
1570                 if (rc) {
1571                         if (rc != -EPERM && rc != -EACCES)
1572                                 CERROR("md_setattr fails: rc = %d\n", rc);
1573                         RETURN(rc);
1574                 }
1575
1576                 if (flags & EXT3_APPEND_FL)
1577                         inode->i_flags |= S_APPEND;
1578                 else
1579                         inode->i_flags &= ~S_APPEND;
1580                 if (flags & EXT3_IMMUTABLE_FL)
1581                         inode->i_flags |= S_IMMUTABLE;
1582                 else
1583                         inode->i_flags &= ~S_IMMUTABLE;
1584                 if (flags & EXT3_NOATIME_FL)
1585                         inode->i_flags |= S_NOATIME;
1586                 else
1587                         inode->i_flags &= ~S_NOATIME;
1588
1589                 RETURN(0);
1590         }
1591         default:
1592                 RETURN(-ENOSYS);
1593         }
1594
1595         RETURN(0);
1596 }
1597
1598 /* this is only called in the case of forced umount. */
1599 void ll_umount_begin(struct super_block *sb)
1600 {
1601         struct ll_sb_info *sbi = ll_s2sbi(sb);
1602         struct obd_ioctl_data ioc_data = { 0 };
1603         struct obd_device *obd;
1604         ENTRY;
1605
1606         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1607                sb->s_count, atomic_read(&sb->s_active));
1608
1609         obd = class_exp2obd(sbi->ll_md_exp);
1610         if (obd == NULL) {
1611                 CERROR("Invalid MDC connection handle "LPX64"\n",
1612                        sbi->ll_md_exp->exp_handle.h_cookie);
1613                 EXIT;
1614                 return;
1615         }
1616         obd->obd_no_recov = 1;
1617         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
1618                       sizeof(ioc_data), &ioc_data, NULL);
1619
1620         obd = class_exp2obd(sbi->ll_dt_exp);
1621         if (obd == NULL) {
1622                 CERROR("Invalid LOV connection handle "LPX64"\n",
1623                        sbi->ll_dt_exp->exp_handle.h_cookie);
1624                 EXIT;
1625                 return;
1626         }
1627
1628         obd->obd_no_recov = 1;
1629         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
1630                       sizeof(ioc_data), &ioc_data, NULL);
1631
1632         /*
1633          * really, we'd like to wait until there are no requests outstanding,
1634          * and then continue.  For now, we just invalidate the requests,
1635          * schedule, and hope.
1636          */
1637         schedule();
1638
1639         EXIT;
1640 }
1641
1642 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
1643                   struct inode **inode, struct ptlrpc_request *req,
1644                   int offset, struct super_block *sb)
1645 {
1646         struct lustre_md md;
1647         int rc = 0;
1648
1649         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
1650         if (rc)
1651                 RETURN(rc);
1652
1653         if (*inode) {
1654                 ll_update_inode(*inode, &md);
1655         } else {
1656                 LASSERT(sb);
1657                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
1658                 if (*inode == NULL || is_bad_inode(*inode)) {
1659                         /* free the lsm if we allocated one above */
1660                         if (md.lsm != NULL)
1661                                 obd_free_memmd(dt_exp, &md.lsm);
1662                         if (md.mea != NULL)
1663                                 obd_free_memmd(md_exp,
1664                                                (struct lov_stripe_md**)&md.mea);
1665                         rc = -ENOMEM;
1666                         CERROR("new_inode -fatal: rc %d\n", rc);
1667                 }
1668         }
1669
1670         RETURN(rc);
1671 }
1672
1673 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
1674                char *filename, struct lustre_id *ret)
1675 {
1676         struct ptlrpc_request *request = NULL;
1677         struct mds_body *body;
1678         int rc;
1679
1680         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
1681                              OBD_MD_FID, 0, &request);
1682         if (rc < 0) {
1683                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
1684                        filename, rc);
1685                 return rc;
1686         }
1687
1688         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
1689         LASSERT(body != NULL);
1690         LASSERT_REPSWABBED(request, 0);
1691
1692         *ret = body->id1;
1693         ptlrpc_req_finished(request);
1694
1695         return rc;
1696 }