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 #include <linux/seq_file.h>
  31
  32 #include <linux/lustre_lite.h>
  33 #include <linux/lustre_ha.h>
  34 #include <linux/lustre_dlm.h>
  35 #include <linux/lprocfs_status.h>
  36 #include <linux/lustre_acl.h>
  37 #include "llite_internal.h"
  38
  39 kmem_cache_t *ll_file_data_slab;
  40 kmem_cache_t *ll_intent_slab;
  41
  42 extern struct address_space_operations ll_aops;
  43 extern struct address_space_operations ll_dir_aops;
  44
  45 #ifndef log2
  46 #define log2(n) ffz(~(n))
  47 #endif
  48
  49 struct ll_sb_info *lustre_init_sbi(struct super_block *sb)
  50 {
  51         struct ll_sb_info *sbi = NULL;
  52         class_uuid_t uuid;
  53         ENTRY;
  54
  55         OBD_ALLOC(sbi, sizeof(*sbi));
  56         if (!sbi)
  57                 RETURN(NULL);
  58
  59         spin_lock_init(&sbi->ll_lock);
  60         INIT_LIST_HEAD(&sbi->ll_pglist);
  61         sbi->ll_pglist_gen = 0;
  62         if (num_physpages < SBI_DEFAULT_RA_MAX / 4)
  63                 sbi->ll_ra_info.ra_max_pages = num_physpages / 4;
  64         else
  65                 sbi->ll_ra_info.ra_max_pages = SBI_DEFAULT_RA_MAX;
  66         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  67         INIT_HLIST_HEAD(&sbi->ll_orphan_dentry_list);
  68         INIT_LIST_HEAD(&sbi->ll_mnt_list);
  69
  70         sema_init(&sbi->ll_gns_sem, 1);
  71         spin_lock_init(&sbi->ll_gns_lock);
  72         INIT_LIST_HEAD(&sbi->ll_gns_sbi_head);
  73         init_waitqueue_head(&sbi->ll_gns_waitq);
  74         init_completion(&sbi->ll_gns_mount_finished);
  75
  76         /* this later may be reset via /proc/fs/... */
  77         memcpy(sbi->ll_gns_oname, ".mntinfo", strlen(".mntinfo"));
  78         sbi->ll_gns_oname[strlen(sbi->ll_gns_oname)] = '\0';
  79
  80         /* this later may be reset via /proc/fs/... */
  81         memcpy(sbi->ll_gns_upcall, "/usr/sbin/gns_upcall",
  82                strlen("/usr/sbin/gns_upcall"));
  83         sbi->ll_gns_upcall[strlen(sbi->ll_gns_upcall)] = '\0';
  84
  85         /* default values, may be changed via /proc/fs/... */
  86         sbi->ll_gns_state = LL_GNS_IDLE;
  87         sbi->ll_gns_pending_dentry = NULL;
  88         atomic_set(&sbi->ll_gns_enabled, 1);
  89         sbi->ll_gns_tick = GNS_TICK_TIMEOUT;
  90         sbi->ll_gns_timeout = GNS_MOUNT_TIMEOUT;
  91
  92         sbi->ll_gns_timer.data = (unsigned long)sbi;
  93         sbi->ll_gns_timer.function = ll_gns_timer_callback;
  94         init_timer(&sbi->ll_gns_timer);
  95
  96         ll_set_sbi(sb, sbi);
  97
  98         generate_random_uuid(uuid);
  99         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 100         RETURN(sbi);
 101 }
 102
 103 void lustre_free_sbi(struct super_block *sb)
 104 {
 105         struct ll_sb_info *sbi = ll_s2sbi(sb);
 106         ENTRY;
 107
 108         if (sbi != NULL) {
 109                 list_del(&sbi->ll_gns_sbi_head);
 110                 del_timer(&sbi->ll_gns_timer);
 111                 OBD_FREE(sbi, sizeof(*sbi));
 112         }
 113         ll_set_sbi(sb, NULL);
 114         EXIT;
 115 }
 116
 117 int lustre_init_dt_desc(struct ll_sb_info *sbi)
 118 {
 119         __u32 valsize;
 120         int rc = 0;
 121         ENTRY;
 122
 123         valsize = sizeof(sbi->ll_dt_desc);
 124         memset(&sbi->ll_dt_desc, 0, sizeof(sbi->ll_dt_desc));
 125         rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 126                           "lovdesc", &valsize, &sbi->ll_dt_desc);
 127         RETURN(rc);
 128 }
 129
 130 extern struct dentry_operations ll_d_ops;
 131
 132 int lustre_common_fill_super(struct super_block *sb, char *lmv, char *lov,
 133                              int async,  char *mds_security,  char *oss_security,
 134                              __u32 *nllu, __u64 *remote)
 135 {
 136         struct ll_sb_info *sbi = ll_s2sbi(sb);
 137         struct ptlrpc_request *request = NULL;
 138         struct lustre_handle dt_conn = {0, };
 139         struct lustre_handle md_conn = {0, };
 140         struct obd_connect_data *data;
 141         struct inode *root = NULL;
 142         struct obd_device *obd;
 143         struct obd_statfs osfs;
 144         struct lustre_md md;
 145         __u32 valsize;
 146         int err;
 147         ENTRY;
 148
 149         obd = class_name2obd(lmv);
 150         if (!obd) {
 151                 CERROR("MDC %s: not setup or attached\n", lmv);
 152                 RETURN(-EINVAL);
 153         }
 154         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 155                      sizeof(async), &async);
 156
 157         if ((*remote & (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) ==
 158             (OBD_CONNECT_LOCAL | OBD_CONNECT_REMOTE)) {
 159                 CERROR("wrong remote flag "LPX64"\n", *remote);
 160                 RETURN(-EINVAL);
 161         }
 162
 163         OBD_ALLOC(data, sizeof(*data));
 164         if (!data)
 165                 RETURN(-ENOMEM);
 166
 167         data->ocd_connect_flags |= *remote & (OBD_CONNECT_LOCAL |
 168                                               OBD_CONNECT_REMOTE);
 169         memcpy(data->ocd_nllu, nllu, sizeof(data->ocd_nllu));
 170
 171         if (mds_security == NULL)
 172                 mds_security = "null";
 173
 174         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 175                            strlen(mds_security), mds_security);
 176         if (err) {
 177                 CERROR("LMV %s: failed to set security %s, err %d\n",
 178                         lmv, mds_security, err);
 179                 OBD_FREE(data, sizeof(*data));
 180                 RETURN(err);
 181         }
 182
 183         if (proc_lustre_fs_root) {
 184                 err = lprocfs_register_mountpoint(proc_lustre_fs_root,
 185                                                   sb, lov, lmv);
 186                 if (err < 0)
 187                         CERROR("could not register mount in /proc/lustre");
 188         }
 189
 190         err = obd_connect(&md_conn, obd, &sbi->ll_sb_uuid, data,
 191                           OBD_OPT_REAL_CLIENT);
 192         if (err == -EBUSY) {
 193                 CERROR("An MDS (lmv %s) is performing recovery, of which this"
 194                        " client is not a part.  Please wait for recovery to "
 195                        "complete, abort, or time out.\n", lmv);
 196                 GOTO(out, err);
 197         } else if (err) {
 198                 CERROR("cannot connect to %s: rc = %d\n", lmv, err);
 199                 GOTO(out, err);
 200         }
 201         sbi->ll_md_exp = class_conn2export(&md_conn);
 202         err = obd_statfs(obd, &osfs, jiffies - HZ);
 203         if (err)
 204                 GOTO(out_lmv, err);
 205
 206         if (!osfs.os_bsize) {
 207                 CERROR("Invalid block size is detected.");
 208                 GOTO(out_lmv, err);
 209         }
 210
 211         sb->s_magic = LL_SUPER_MAGIC;
 212         sb->s_blocksize = osfs.os_bsize;
 213         sb->s_blocksize_bits = log2(osfs.os_bsize);
 214         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 215
 216         /* in 2.6.x FS is not allowed to form s_dev */
 217 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 218         {
 219                 kdev_t devno;
 220
 221                 devno = get_uuid2int((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid,
 222                                      strlen((char *)sbi->ll_md_exp->exp_obd->obd_uuid.uuid));
 223
 224                 sb->s_dev = devno;
 225         }
 226 #endif
 227
 228         /* after statfs, we are supposed to have connected to MDSs,
 229          * so it's ok to check remote flag returned.
 230          */
 231         valsize = sizeof(&sbi->ll_remote);
 232         err = obd_get_info(sbi->ll_md_exp, strlen("remote_flag"), "remote_flag",
 233                            &valsize, &sbi->ll_remote);
 234         if (err) {
 235                 CERROR("fail to obtain remote flag\n");
 236                 GOTO(out, err);
 237         }
 238
 239         obd = class_name2obd(lov);
 240         if (!obd) {
 241                 CERROR("OSC %s: not setup or attached\n", lov);
 242                 GOTO(out_lmv, err);
 243         }
 244         obd_set_info(obd->obd_self_export, strlen("async"), "async",
 245                      sizeof(async), &async);
 246
 247        if (oss_security == NULL)
 248                 oss_security = "null";
 249
 250         err = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 251                            strlen(oss_security), oss_security);
 252         if (err) {
 253                 CERROR("LOV %s: failed to set security %s, err %d\n",
 254                         lov, oss_security, err);
 255                 OBD_FREE(data, sizeof(*data));
 256                 RETURN(err);
 257         }
 258
 259         err = obd_connect(&dt_conn, obd, &sbi->ll_sb_uuid, data, 0);
 260         if (err == -EBUSY) {
 261                 CERROR("An OST (lov %s) is performing recovery, of which this"
 262                        " client is not a part.  Please wait for recovery to "
 263                        "complete, abort, or time out.\n", lov);
 264                 GOTO(out, err);
 265         } else if (err) {
 266                 CERROR("cannot connect to %s: rc = %d\n", lov, err);
 267                 GOTO(out_lmv, err);
 268         }
 269         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 270
 271         err = lustre_init_dt_desc(sbi);
 272         if (err == 0) {
 273                 int mdsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 274                 obd_init_ea_size(sbi->ll_md_exp, mdsize,
 275                                  sbi->ll_dt_desc.ld_tgt_count *
 276                                  sizeof(struct llog_cookie));
 277         }
 278
 279         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_rootid);
 280         if (err) {
 281                 CERROR("cannot mds_connect: rc = %d\n", err);
 282                 GOTO(out_lov, err);
 283         }
 284         CDEBUG(D_SUPER, "rootid "DLID4"\n", OLID4(&sbi->ll_rootid));
 285
 286         sb->s_op = &lustre_super_operations;
 287
 288         /* make root inode */
 289         err = md_getattr(sbi->ll_md_exp, &sbi->ll_rootid,
 290                          (OBD_MD_FLNOTOBD | OBD_MD_FLBLOCKS | OBD_MD_FID),
 291                          NULL, 0, 0, &request);
 292         if (err) {
 293                 CERROR("md_getattr failed for root: rc = %d\n", err);
 294                 GOTO(out_lov, err);
 295         }
 296
 297         err = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
 298                                 sbi->ll_dt_exp, &md);
 299         if (err) {
 300                 CERROR("failed to understand root inode md: rc = %d\n", err);
 301                 ptlrpc_req_finished(request);
 302                 GOTO(out_lov, err);
 303         }
 304
 305         LASSERT(id_ino(&sbi->ll_rootid) != 0);
 306         root = ll_iget(sb, id_ino(&sbi->ll_rootid), &md);
 307
 308         ptlrpc_req_finished(request);
 309
 310         if (root == NULL || is_bad_inode(root)) {
 311                 if (md.lsm != NULL)
 312                     obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
 313                 if (md.mea != NULL)
 314                     obd_free_memmd(sbi->ll_md_exp,
 315                                    (struct lov_stripe_md**)&md.mea);
 316                 CERROR("lustre_lite: bad iget4 for root\n");
 317                 GOTO(out_root, err = -EBADF);
 318         }
 319
 320         err = ll_close_thread_start(&sbi->ll_lcq);
 321         if (err) {
 322                 CERROR("cannot start close thread: rc %d\n", err);
 323                 GOTO(out_root, err);
 324         }
 325
 326         ll_gns_add_timer(sbi);
 327
 328         /* making vm readahead 0 for 2.4.x. In the case of 2.6.x,
 329            backing dev info assigned to inode mapping is used for
 330            determining maximal readahead. */
 331 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0)) && \
 332     !defined(KERNEL_HAS_AS_MAX_READAHEAD)
 333         /* bug 2805 - set VM readahead to zero */
 334         vm_max_readahead = vm_min_readahead = 0;
 335 #endif
 336
 337         sb->s_root = d_alloc_root(root);
 338         sb->s_root->d_op = &ll_d_ops;
 339
 340         sb->s_flags |= MS_POSIXACL;
 341 #ifdef S_PDIROPS
 342         CWARN("Enabling PDIROPS\n");
 343         sb->s_flags |= S_PDIROPS;
 344 #endif
 345
 346         if (data != NULL)
 347                 OBD_FREE(data, sizeof(*data));
 348         RETURN(err);
 349 out_root:
 350         if (root)
 351                 iput(root);
 352 out_lov:
 353         obd_disconnect(sbi->ll_dt_exp, 0);
 354 out_lmv:
 355         obd_disconnect(sbi->ll_md_exp, 0);
 356 out:
 357         if (data != NULL)
 358                 OBD_FREE(data, sizeof(*data));
 359         lprocfs_unregister_mountpoint(sbi);
 360         return err;
 361 }
 362
 363 void lustre_common_put_super(struct super_block *sb)
 364 {
 365         struct ll_sb_info *sbi = ll_s2sbi(sb);
 366         struct hlist_node *tmp, *next;
 367         ENTRY;
 368
 369         ll_gns_del_timer(sbi);
 370         ll_close_thread_shutdown(sbi->ll_lcq);
 371
 372         list_del(&sbi->ll_conn_chain);
 373         obd_disconnect(sbi->ll_dt_exp, 0);
 374
 375         lprocfs_unregister_mountpoint(sbi);
 376         if (sbi->ll_proc_root) {
 377                 lprocfs_remove(sbi->ll_proc_root);
 378                 sbi->ll_proc_root = NULL;
 379         }
 380
 381         obd_disconnect(sbi->ll_md_exp, 0);
 382
 383         // We do this to get rid of orphaned dentries. That is not really trw.
 384         hlist_for_each_safe(tmp, next, &sbi->ll_orphan_dentry_list) {
 385                 struct dentry *dentry = hlist_entry(tmp, struct dentry, d_hash);
 386                 CWARN("orphan dentry %.*s (%p->%p) at unmount\n",
 387                       dentry->d_name.len, dentry->d_name.name, dentry, next);
 388                 shrink_dcache_parent(dentry);
 389         }
 390         EXIT;
 391 }
 392
 393 char *ll_read_opt(const char *opt, char *data)
 394 {
 395         char *value;
 396         char *retval;
 397         ENTRY;
 398
 399         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 400         if (strncmp(opt, data, strlen(opt)))
 401                 RETURN(NULL);
 402         if ((value = strchr(data, '=')) == NULL)
 403                 RETURN(NULL);
 404
 405         value++;
 406         OBD_ALLOC(retval, strlen(value) + 1);
 407         if (!retval) {
 408                 CERROR("out of memory!\n");
 409                 RETURN(NULL);
 410         }
 411
 412         memcpy(retval, value, strlen(value)+1);
 413         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 414         RETURN(retval);
 415 }
 416
 417 int ll_set_opt(const char *opt, char *data, int fl)
 418 {
 419         ENTRY;
 420
 421         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 422         if (strncmp(opt, data, strlen(opt)))
 423                 RETURN(0);
 424         else
 425                 RETURN(fl);
 426 }
 427
 428 void ll_options(char *options, char **lov, char **lmv, char **mds_sec,
 429                 char **oss_sec, int *async, int *flags)
 430 {
 431         char *this_char;
 432 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 433         char *opt_ptr = options;
 434 #endif
 435         ENTRY;
 436
 437         if (!options) {
 438                 EXIT;
 439                 return;
 440         }
 441
 442         *async = 0;
 443 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
 444         for (this_char = strtok (options, ",");
 445              this_char != NULL;
 446              this_char = strtok (NULL, ",")) {
 447 #else
 448         while ((this_char = strsep (&opt_ptr, ",")) != NULL) {
 449 #endif
 450                 CDEBUG(D_SUPER, "this_char %s\n", this_char);
 451                 if (!*lov && (*lov = ll_read_opt("osc", this_char)))
 452                         continue;
 453                 if (!*lmv && (*lmv = ll_read_opt("mdc", this_char)))
 454                         continue;
 455                 if (!strncmp(this_char, "lasync", strlen("lasync"))) {
 456                         *async = 1;
 457                         continue;
 458                 }
 459                 if (!*mds_sec && (*mds_sec = ll_read_opt("mds_sec", this_char)))
 460                         continue;
 461                 if (!*oss_sec && (*oss_sec = ll_read_opt("oss_sec", this_char)))
 462                         continue;
 463                 if (!(*flags & LL_SBI_NOLCK) &&
 464                     ((*flags) = (*flags) |
 465                                 ll_set_opt("nolock", this_char,
 466                                            LL_SBI_NOLCK)))
 467                         continue;
 468         }
 469
 470         EXIT;
 471 }
 472
 473 void ll_lli_init(struct ll_inode_info *lli)
 474 {
 475         sema_init(&lli->lli_open_sem, 1);
 476         sema_init(&lli->lli_size_sem, 1);
 477         lli->lli_flags = 0;
 478         lli->lli_size_pid = 0;
 479         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 480         spin_lock_init(&lli->lli_lock);
 481         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 482         lli->lli_inode_magic = LLI_INODE_MAGIC;
 483         memset(&lli->lli_id, 0, sizeof(lli->lli_id));
 484         sema_init(&lli->lli_och_sem, 1);
 485         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 486         lli->lli_mds_exec_och = NULL;
 487         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 488         lli->lli_open_fd_exec_count = 0;
 489 }
 490
 491 int ll_fill_super(struct super_block *sb, void *data, int silent)
 492 {
 493         struct ll_sb_info *sbi;
 494         char *lov = NULL;
 495         char *lmv = NULL;
 496         char *mds_sec = NULL;
 497         char *oss_sec = NULL;
 498         int async, err;
 499         __u32 nllu[2] = { NOBODY_UID, NOBODY_GID };
 500         __u64 remote_flag = 0;
 501         ENTRY;
 502
 503         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 504
 505         sbi = lustre_init_sbi(sb);
 506         if (!sbi)
 507                 RETURN(-ENOMEM);
 508
 509         sbi->ll_flags |= LL_SBI_READAHEAD;
 510         ll_options(data, &lov, &lmv, &mds_sec, &oss_sec,
 511                    &async, &sbi->ll_flags);
 512
 513         if (!lov) {
 514                 CERROR("no osc\n");
 515                 GOTO(out, err = -EINVAL);
 516         }
 517
 518         if (!lmv) {
 519                 CERROR("no mdc\n");
 520                 GOTO(out, err = -EINVAL);
 521         }
 522
 523         err = lustre_common_fill_super(sb, lmv, lov, async, mds_sec, oss_sec,
 524                                        nllu, &remote_flag);
 525         EXIT;
 526 out:
 527         if (err)
 528                 lustre_free_sbi(sb);
 529
 530         if (lmv)
 531                 OBD_FREE(lmv, strlen(lmv) + 1);
 532         if (lov)
 533                 OBD_FREE(lov, strlen(lov) + 1);
 534         if (mds_sec)
 535                 OBD_FREE(mds_sec, strlen(mds_sec) + 1);
 536         if (oss_sec)
 537                 OBD_FREE(oss_sec, strlen(oss_sec) + 1);
 538
 539         return err;
 540 } /* ll_read_super */
 541
 542 static int lustre_process_log(struct lustre_mount_data *lmd, char *profile,
 543                               struct config_llog_instance *cfg, int allow_recov)
 544 {
 545         struct lustre_cfg *lcfg = NULL;
 546         struct lustre_cfg_bufs bufs;
 547         struct portals_cfg pcfg;
 548         char *peer = "MDS_PEER_UUID";
 549         struct obd_device *obd;
 550         struct lustre_handle md_conn = {0, };
 551         struct obd_export *exp;
 552         char *name = "mdc_dev";
 553         class_uuid_t uuid;
 554         struct obd_uuid lmv_uuid;
 555         struct llog_ctxt *ctxt;
 556         int rc = 0, err = 0;
 557         ENTRY;
 558
 559         if (lmd_bad_magic(lmd))
 560                 RETURN(-EINVAL);
 561
 562         generate_random_uuid(uuid);
 563         class_uuid_unparse(uuid, &lmv_uuid);
 564
 565         if (lmd->lmd_local_nid) {
 566                 PCFG_INIT(pcfg, NAL_CMD_REGISTER_MYNID);
 567                 pcfg.pcfg_nal = lmd->lmd_nal;
 568                 pcfg.pcfg_nid = lmd->lmd_local_nid;
 569                 rc = libcfs_nal_cmd(&pcfg);
 570                 if (rc < 0)
 571                         GOTO(out, rc);
 572         }
 573
 574         if (lmd->lmd_nal == SOCKNAL ||
 575             lmd->lmd_nal == OPENIBNAL ||
 576             lmd->lmd_nal == IIBNAL ||
 577             lmd->lmd_nal == VIBNAL ||
 578             lmd->lmd_nal == RANAL) {
 579                 PCFG_INIT(pcfg, NAL_CMD_ADD_PEER);
 580                 pcfg.pcfg_nal     = lmd->lmd_nal;
 581                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 582                 pcfg.pcfg_id      = lmd->lmd_server_ipaddr;
 583                 pcfg.pcfg_misc    = lmd->lmd_port;
 584                 rc = libcfs_nal_cmd(&pcfg);
 585                 if (rc < 0)
 586                         GOTO(out, rc);
 587         }
 588         lustre_cfg_bufs_reset(&bufs, name);
 589         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 590
 591         lcfg = lustre_cfg_new(LCFG_ADD_UUID, &bufs);
 592         lcfg->lcfg_nal = lmd->lmd_nal;
 593         lcfg->lcfg_nid = lmd->lmd_server_nid;
 594         LASSERT(lcfg->lcfg_nal);
 595         LASSERT(lcfg->lcfg_nid);
 596         err = class_process_config(lcfg);
 597         lustre_cfg_free(lcfg);
 598         if (err < 0)
 599                 GOTO(out_del_conn, err);
 600
 601         lustre_cfg_bufs_reset(&bufs, name);
 602         lustre_cfg_bufs_set_string(&bufs, 1, OBD_MDC_DEVICENAME);
 603         lustre_cfg_bufs_set_string(&bufs, 2, (char *)lmv_uuid.uuid);
 604
 605         lcfg = lustre_cfg_new(LCFG_ATTACH, &bufs);
 606         err = class_process_config(lcfg);
 607         lustre_cfg_free(lcfg);
 608         if (err < 0)
 609                 GOTO(out_del_uuid, err);
 610
 611         lustre_cfg_bufs_reset(&bufs, name);
 612         lustre_cfg_bufs_set_string(&bufs, 1, lmd->lmd_mds);
 613         lustre_cfg_bufs_set_string(&bufs, 2, peer);
 614
 615         lcfg = lustre_cfg_new(LCFG_SETUP, &bufs);
 616         err = class_process_config(lcfg);
 617         lustre_cfg_free(lcfg);
 618         if (err < 0)
 619                 GOTO(out_detach, err);
 620
 621         obd = class_name2obd(name);
 622         if (obd == NULL)
 623                 GOTO(out_cleanup, rc = -EINVAL);
 624
 625         rc = obd_set_info(obd->obd_self_export, strlen("sec"), "sec",
 626                           strlen(lmd->lmd_mds_security), lmd->lmd_mds_security);
 627         if (rc)
 628                 GOTO(out_cleanup, rc);
 629
 630         /* Disable initial recovery on this import */
 631         rc = obd_set_info(obd->obd_self_export,
 632                           strlen("initial_recov"), "initial_recov",
 633                           sizeof(allow_recov), &allow_recov);
 634         if (rc)
 635                 GOTO(out_cleanup, rc);
 636
 637         rc = obd_connect(&md_conn, obd, &lmv_uuid, NULL, 0);
 638         if (rc) {
 639                 CERROR("cannot connect to %s: rc = %d\n", lmd->lmd_mds, rc);
 640                 GOTO(out_cleanup, rc);
 641         }
 642
 643         exp = class_conn2export(&md_conn);
 644
 645         ctxt = llog_get_context(&exp->exp_obd->obd_llogs,LLOG_CONFIG_REPL_CTXT);
 646         rc = class_config_process_llog(ctxt, profile, cfg);
 647         if (rc)
 648                 CERROR("class_config_process_llog failed: rc = %d\n", rc);
 649
 650         err = obd_disconnect(exp, 0);
 651
 652         EXIT;
 653 out_cleanup:
 654         lustre_cfg_bufs_reset(&bufs, name);
 655         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 656         err = class_process_config(lcfg);
 657         lustre_cfg_free(lcfg);
 658         if (err < 0)
 659                 GOTO(out, err);
 660 out_detach:
 661         lustre_cfg_bufs_reset(&bufs, name);
 662         lcfg = lustre_cfg_new(LCFG_DETACH, &bufs);
 663         err = class_process_config(lcfg);
 664         lustre_cfg_free(lcfg);
 665         if (err < 0)
 666                 GOTO(out, err);
 667
 668 out_del_uuid:
 669         lustre_cfg_bufs_reset(&bufs, name);
 670         lustre_cfg_bufs_set_string(&bufs, 1, peer);
 671         lcfg = lustre_cfg_new(LCFG_DEL_UUID, &bufs);
 672         err = class_process_config(lcfg);
 673         lustre_cfg_free(lcfg);
 674
 675 out_del_conn:
 676         if (lmd->lmd_nal == SOCKNAL ||
 677             lmd->lmd_nal == OPENIBNAL ||
 678             lmd->lmd_nal == IIBNAL ||
 679             lmd->lmd_nal == VIBNAL ||
 680             lmd->lmd_nal == RANAL) {
 681                 int err2;
 682
 683                 PCFG_INIT(pcfg, NAL_CMD_DEL_PEER);
 684                 pcfg.pcfg_nal     = lmd->lmd_nal;
 685                 pcfg.pcfg_nid     = lmd->lmd_server_nid;
 686                 pcfg.pcfg_flags   = 1;          /* single_share */
 687                 err2 = libcfs_nal_cmd(&pcfg);
 688                 if (err2 && !err)
 689                         err = err2;
 690                 if (err < 0)
 691                         GOTO(out, err);
 692         }
 693 out:
 694         if (rc == 0)
 695                 rc = err;
 696
 697         return rc;
 698 }
 699
 700 static void lustre_manual_cleanup(struct ll_sb_info *sbi)
 701 {
 702         struct lustre_cfg *lcfg;
 703         struct lustre_cfg_bufs bufs;
 704         struct obd_device *obd;
 705         int next = 0;
 706
 707         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) != NULL)
 708         {
 709                 int err;
 710
 711                 lustre_cfg_bufs_reset(&bufs, obd->obd_name);
 712                 lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
 713                 err = class_process_config(lcfg);
 714                 if (err) {
 715                         CERROR("cleanup failed: %s\n", obd->obd_name);
 716                         //continue;
 717                 }
 718
 719                 lcfg->lcfg_command = LCFG_DETACH;
 720                 err = class_process_config(lcfg);
 721                 lustre_cfg_free(lcfg);
 722                 if (err) {
 723                         CERROR("detach failed: %s\n", obd->obd_name);
 724                         //continue;
 725                 }
 726         }
 727
 728         if (sbi->ll_lmd != NULL)
 729                 class_del_profile(sbi->ll_lmd->lmd_profile);
 730 }
 731
 732 int lustre_fill_super(struct super_block *sb, void *data, int silent)
 733 {
 734         struct lustre_mount_data * lmd = data;
 735         char *lov = NULL, *lmv = NULL;
 736         struct ll_sb_info *sbi;
 737         int err;
 738         ENTRY;
 739
 740         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 741         if (lmd_bad_magic(lmd))
 742                 RETURN(-EINVAL);
 743
 744         sbi = lustre_init_sbi(sb);
 745         if (!sbi)
 746                 RETURN(-ENOMEM);
 747
 748         sbi->ll_flags |= LL_SBI_READAHEAD;
 749
 750         if (lmd->lmd_profile) {
 751                 struct lustre_profile *lprof;
 752                 struct config_llog_instance cfg;
 753                 int len;
 754
 755                 if (lmd->lmd_mds[0] == '\0') {
 756                         CERROR("no mds name\n");
 757                         GOTO(out_free, err = -EINVAL);
 758                 }
 759                 lmd->lmd_mds_security[sizeof(lmd->lmd_mds_security) - 1] = 0;
 760                 lmd->lmd_oss_security[sizeof(lmd->lmd_oss_security) - 1] = 0;
 761
 762                 OBD_ALLOC(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 763                 if (sbi->ll_lmd == NULL)
 764                         GOTO(out_free, err = -ENOMEM);
 765                 memcpy(sbi->ll_lmd, lmd, sizeof(*lmd));
 766
 767                 /* generate a string unique to this super, let's try
 768                  the address of the super itself.*/
 769                 len = (sizeof(sb) * 2) + 1;
 770                 OBD_ALLOC(sbi->ll_instance, len);
 771                 if (sbi->ll_instance == NULL)
 772                         GOTO(out_free, err = -ENOMEM);
 773                 sprintf(sbi->ll_instance, "%p", sb);
 774
 775                 cfg.cfg_instance = sbi->ll_instance;
 776                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 777                 cfg.cfg_local_nid = lmd->lmd_local_nid;
 778                 err = lustre_process_log(lmd, lmd->lmd_profile, &cfg, 0);
 779                 if (err < 0) {
 780                         CERROR("Unable to process log: %s\n", lmd->lmd_profile);
 781                         GOTO(out_free, err);
 782                 }
 783
 784                 lprof = class_get_profile(lmd->lmd_profile);
 785                 if (lprof == NULL) {
 786                         CERROR("No profile found: %s\n", lmd->lmd_profile);
 787                         GOTO(out_free, err = -EINVAL);
 788                 }
 789                 if (lov)
 790                         OBD_FREE(lov, strlen(lov) + 1);
 791                 OBD_ALLOC(lov, strlen(lprof->lp_lov) +
 792                           strlen(sbi->ll_instance) + 2);
 793                 sprintf(lov, "%s-%s", lprof->lp_lov, sbi->ll_instance);
 794
 795                 if (lmv)
 796                         OBD_FREE(lmv, strlen(lmv) + 1);
 797                 OBD_ALLOC(lmv, strlen(lprof->lp_lmv) +
 798                           strlen(sbi->ll_instance) + 2);
 799                 sprintf(lmv, "%s-%s", lprof->lp_lmv, sbi->ll_instance);
 800         }
 801
 802         if (!lov) {
 803                 CERROR("no osc\n");
 804                 GOTO(out_free, err = -EINVAL);
 805         }
 806
 807         if (!lmv) {
 808                 CERROR("no mdc\n");
 809                 GOTO(out_free, err = -EINVAL);
 810         }
 811
 812         err = lustre_common_fill_super(sb, lmv, lov, lmd->lmd_async,
 813                                        lmd->lmd_mds_security,
 814                                        lmd->lmd_oss_security,
 815                                        &lmd->lmd_nllu, &lmd->lmd_remote_flag);
 816
 817         if (err)
 818                 GOTO(out_free, err);
 819
 820 out_dev:
 821         if (lmv)
 822                 OBD_FREE(lmv, strlen(lmv) + 1);
 823         if (lov)
 824                 OBD_FREE(lov, strlen(lov) + 1);
 825
 826         RETURN(err);
 827
 828 out_free:
 829         if (sbi->ll_lmd) {
 830                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 831                 int err;
 832
 833                 if (sbi->ll_instance != NULL) {
 834                         struct lustre_mount_data *lmd = sbi->ll_lmd;
 835                         struct config_llog_instance cfg;
 836                         char *cl_prof;
 837
 838                         cfg.cfg_instance = sbi->ll_instance;
 839                         cfg.cfg_uuid = sbi->ll_sb_uuid;
 840
 841                         OBD_ALLOC(cl_prof, len);
 842                         sprintf(cl_prof, "%s-clean", lmd->lmd_profile);
 843                         err = lustre_process_log(lmd, cl_prof, &cfg, 0);
 844                         if (err < 0) {
 845                                 CERROR("Unable to process log: %s\n", cl_prof);
 846                                 lustre_manual_cleanup(sbi);
 847                         }
 848                         OBD_FREE(cl_prof, len);
 849                         OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 850                 }
 851                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 852         }
 853         lustre_free_sbi(sb);
 854         goto out_dev;
 855 } /* lustre_fill_super */
 856
 857 void lustre_put_super(struct super_block *sb)
 858 {
 859         struct obd_device *obd;
 860         struct ll_sb_info *sbi = ll_s2sbi(sb);
 861         int force_umount = 0;
 862         ENTRY;
 863
 864         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 865         obd = class_exp2obd(sbi->ll_md_exp);
 866         if (obd)
 867                 force_umount = obd->obd_no_recov;
 868         obd = NULL;
 869
 870         lustre_common_put_super(sb);
 871         if (sbi->ll_lmd != NULL) {
 872                 char *cl_prof;
 873                 int len = strlen(sbi->ll_lmd->lmd_profile) + sizeof("-clean")+1;
 874                 int err;
 875                 struct config_llog_instance cfg;
 876
 877                 if (force_umount) {
 878                         CERROR("force umount, doing manual cleanup\n");
 879                         lustre_manual_cleanup(sbi);
 880                         GOTO(free_lmd, 0);
 881                 }
 882
 883                 cfg.cfg_instance = sbi->ll_instance;
 884                 cfg.cfg_uuid = sbi->ll_sb_uuid;
 885
 886                 OBD_ALLOC(cl_prof, len);
 887                 sprintf(cl_prof, "%s-clean", sbi->ll_lmd->lmd_profile);
 888                 err = lustre_process_log(sbi->ll_lmd, cl_prof, &cfg, 0);
 889                 if (err < 0) {
 890                         CERROR("Unable to process log: %s, doing manual cleanup"
 891                                "\n", cl_prof);
 892                         lustre_manual_cleanup(sbi);
 893                 }
 894
 895                 OBD_FREE(cl_prof, len);
 896         free_lmd:
 897                 OBD_FREE(sbi->ll_lmd, sizeof(*sbi->ll_lmd));
 898                 OBD_FREE(sbi->ll_instance, strlen(sbi->ll_instance) + 1);
 899         }
 900
 901         lustre_free_sbi(sb);
 902
 903         EXIT;
 904 } /* lustre_put_super */
 905
 906 int ll_process_config_update(struct ll_sb_info *sbi, int clean)
 907 {
 908         struct lustre_mount_data *lmd = sbi->ll_lmd;
 909         char *profile = lmd->lmd_profile, *name = NULL;
 910         struct config_llog_instance cfg;
 911         int rc, namelen =  0, version;
 912         struct llog_ctxt *ctxt;
 913         ENTRY;
 914
 915         if (profile == NULL)
 916                 RETURN(0);
 917         if (lmd == NULL) {
 918                 CERROR("Client not mounted with zero-conf; cannot "
 919                        "process update log.\n");
 920                 RETURN(0);
 921         }
 922
 923         rc = obd_cancel_unused(sbi->ll_md_exp, NULL,
 924                                LDLM_FL_CONFIG_CHANGE, NULL);
 925         if (rc != 0)
 926                 CWARN("obd_cancel_unused(mdc): %d\n", rc);
 927
 928         rc = obd_cancel_unused(sbi->ll_dt_exp, NULL,
 929                                LDLM_FL_CONFIG_CHANGE, NULL);
 930         if (rc != 0)
 931                 CWARN("obd_cancel_unused(lov): %d\n", rc);
 932
 933         cfg.cfg_instance = sbi->ll_instance;
 934         cfg.cfg_uuid = sbi->ll_sb_uuid;
 935         cfg.cfg_local_nid = lmd->lmd_local_nid;
 936
 937         namelen = strlen(profile) + 20; /* -clean-######### */
 938         OBD_ALLOC(name, namelen);
 939         if (name == NULL)
 940                 RETURN(-ENOMEM);
 941
 942         if (clean) {
 943                 version = sbi->ll_config_version - 1;
 944                 sprintf(name, "%s-clean-%d", profile, version);
 945         } else {
 946                 version = sbi->ll_config_version + 1;
 947                 sprintf(name, "%s-%d", profile, version);
 948         }
 949
 950         CWARN("Applying configuration log %s\n", name);
 951
 952         ctxt = llog_get_context(&sbi->ll_md_exp->exp_obd->obd_llogs,
 953                                 LLOG_CONFIG_REPL_CTXT);
 954         rc = class_config_process_llog(ctxt, name, &cfg);
 955         if (rc == 0)
 956                 sbi->ll_config_version = version;
 957         CWARN("Finished applying configuration log %s: %d\n", name, rc);
 958
 959         if (rc == 0 && clean == 0) {
 960                 struct lov_desc desc;
 961                 __u32 valsize;
 962                 int rc = 0;
 963
 964                 valsize = sizeof(desc);
 965                 rc = obd_get_info(sbi->ll_dt_exp, strlen("lovdesc") + 1,
 966                                   "lovdesc", &valsize, &desc);
 967
 968                 rc = obd_init_ea_size(sbi->ll_md_exp,
 969                                       obd_size_diskmd(sbi->ll_dt_exp, NULL),
 970                                       (desc.ld_tgt_count *
 971                                        sizeof(struct llog_cookie)));
 972         }
 973         OBD_FREE(name, namelen);
 974         RETURN(rc);
 975 }
 976
 977 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
 978 {
 979         struct inode *inode = NULL;
 980
 981         /* NOTE: we depend on atomic igrab() -bzzz */
 982         lock_res(lock->l_resource);
 983         if (lock->l_ast_data) {
 984                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
 985                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
 986                         inode = igrab(lock->l_ast_data);
 987                 } else {
 988                         inode = lock->l_ast_data;
 989                         CDEBUG(inode->i_state & I_FREEING ? D_INFO : D_WARNING,
 990                                "l_ast_data %p is bogus: magic %0x8\n",
 991                                lock->l_ast_data, lli->lli_inode_magic);
 992                         inode = NULL;
 993                 }
 994         }
 995         unlock_res(lock->l_resource);
 996         return inode;
 997 }
 998
 999 int null_if_equal(struct ldlm_lock *lock, void *data)
1000 {
1001         if (data == lock->l_ast_data) {
1002                 lock->l_ast_data = NULL;
1003
1004                 if (lock->l_req_mode != lock->l_granted_mode)
1005                         LDLM_ERROR(lock,"clearing inode with ungranted lock\n");
1006         }
1007
1008         return LDLM_ITER_CONTINUE;
1009 }
1010
1011 static void remote_acl_free(struct remote_acl *racl);
1012
1013 void ll_clear_inode(struct inode *inode)
1014 {
1015         struct lustre_id id;
1016         struct ll_inode_info *lli = ll_i2info(inode);
1017         struct ll_sb_info *sbi = ll_i2sbi(inode);
1018         ENTRY;
1019
1020         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1021                inode->i_generation, inode);
1022
1023         lli->lli_inode_magic = LLI_INODE_DEAD;
1024         ll_inode2id(&id, inode);
1025
1026         clear_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &(ll_i2info(inode)->lli_flags));
1027         md_change_cbdata(sbi->ll_md_exp, &id, null_if_equal, inode);
1028
1029         LASSERT(!lli->lli_open_fd_write_count);
1030         LASSERT(!lli->lli_open_fd_read_count);
1031         LASSERT(!lli->lli_open_fd_exec_count);
1032         if (lli->lli_mds_write_och)
1033                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_WRITE);
1034         if (lli->lli_mds_exec_och)
1035                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_EXEC);
1036         if (lli->lli_mds_read_och)
1037                 ll_md_real_close(sbi->ll_md_exp, inode, FMODE_READ);
1038         if (lli->lli_smd)
1039                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1040                                   null_if_equal, inode);
1041
1042         if (lli->lli_smd) {
1043                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1044                 lli->lli_smd = NULL;
1045         }
1046
1047         if (lli->lli_mea) {
1048                 obd_free_memmd(sbi->ll_md_exp,
1049                                (struct lov_stripe_md **) &lli->lli_mea);
1050                 lli->lli_mea = NULL;
1051         }
1052
1053         if (lli->lli_symlink_name) {
1054                 OBD_FREE(lli->lli_symlink_name,
1055                          strlen(lli->lli_symlink_name) + 1);
1056                 lli->lli_symlink_name = NULL;
1057         }
1058
1059         if (lli->lli_posix_acl) {
1060                 LASSERT(lli->lli_remote_acl == NULL);
1061                 posix_acl_release(lli->lli_posix_acl);
1062                 lli->lli_posix_acl = NULL;
1063         }
1064
1065         if (lli->lli_remote_acl) {
1066                 LASSERT(lli->lli_posix_acl == NULL);
1067                 remote_acl_free(lli->lli_remote_acl);
1068                 lli->lli_remote_acl = NULL;
1069         }
1070
1071         lli->lli_inode_magic = LLI_INODE_DEAD;
1072
1073         EXIT;
1074 }
1075
1076 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1077  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1078  * keep these values until such a time that objects are allocated for it.
1079  * We do the MDS operations first, as it is checking permissions for us.
1080  * We don't to the MDS RPC if there is nothing that we want to store there,
1081  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1082  * going to do an RPC anyways.
1083  *
1084  * If we are doing a truncate, we will send the mtime and ctime updates
1085  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1086  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1087  * at the same time.
1088  */
1089 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1090 {
1091         struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1092         struct ll_inode_info *lli = ll_i2info(inode);
1093         struct ll_sb_info *sbi = ll_i2sbi(inode);
1094         struct ptlrpc_request *request = NULL;
1095         struct mdc_op_data *op_data;
1096         int ia_valid = attr->ia_valid;
1097         int err, rc = 0;
1098         ENTRY;
1099
1100         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu\n", inode->i_ino);
1101         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_SETATTR);
1102
1103         if (ia_valid & ATTR_SIZE) {
1104                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1105                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1106                                attr->ia_size, ll_file_maxbytes(inode));
1107                         RETURN(-EFBIG);
1108                 }
1109
1110                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1111         }
1112
1113         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1114         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1115                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1116                         RETURN(-EPERM);
1117         }
1118
1119         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1120         if (attr->ia_valid & ATTR_CTIME) {
1121                 attr->ia_ctime = CURRENT_TIME;
1122                 attr->ia_valid |= ATTR_CTIME_SET;
1123         }
1124         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1125                 attr->ia_atime = CURRENT_TIME;
1126                 attr->ia_valid |= ATTR_ATIME_SET;
1127         }
1128         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1129                 attr->ia_mtime = CURRENT_TIME;
1130                 attr->ia_valid |= ATTR_MTIME_SET;
1131         }
1132
1133         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1134                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1135                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1136                        LTIME_S(CURRENT_TIME));
1137
1138         if (lsm)
1139                 attr->ia_valid &= ~ATTR_SIZE;
1140
1141         /* If only OST attributes being set on objects, don't do MDS RPC.
1142          * In that case, we need to check permissions and update the local
1143          * inode ourselves so we can call obdo_from_inode() always. */
1144         if (ia_valid & (lsm ? ~(ATTR_SIZE | ATTR_FROM_OPEN /*| ATTR_RAW*/) : ~0)) {
1145                 struct lustre_md md;
1146
1147                 OBD_ALLOC(op_data, sizeof(*op_data));
1148                 if (op_data == NULL)
1149                         RETURN(-ENOMEM);
1150                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
1151
1152                 rc = md_setattr(sbi->ll_md_exp, op_data,
1153                                 attr, NULL, 0, NULL, 0, &request);
1154                 OBD_FREE(op_data, sizeof(*op_data));
1155                 if (rc) {
1156                         ptlrpc_req_finished(request);
1157                         if (rc != -EPERM && rc != -EACCES)
1158                                 CERROR("md_setattr fails: rc = %d\n", rc);
1159                         RETURN(rc);
1160                 }
1161
1162                 rc = mdc_req2lustre_md(sbi->ll_md_exp, request, 0,
1163                                        sbi->ll_dt_exp, &md);
1164                 if (rc) {
1165                         ptlrpc_req_finished(request);
1166                         RETURN(rc);
1167                 }
1168
1169                 /* We call inode_setattr to adjust timestamps, but we first
1170                  * clear ATTR_SIZE to avoid invoking vmtruncate.
1171                  *
1172                  * NB: ATTR_SIZE will only be set at this point if the size
1173                  * resides on the MDS, ie, this file has no objects. */
1174                 attr->ia_valid &= ~ATTR_SIZE;
1175
1176                 /*
1177                  * assigning inode_setattr() to @err to disable warning that
1178                  * function's result should be checked by by caller. error is
1179                  * impossible here, as vmtruncate() control path is disabled.
1180                  */
1181                 err = inode_setattr(inode, attr);
1182                 ll_update_inode(inode, &md);
1183                 ptlrpc_req_finished(request);
1184
1185                 if (!lsm || !S_ISREG(inode->i_mode)) {
1186                         CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1187                         RETURN(0);
1188                 }
1189         } else {
1190                 /* The OST doesn't check permissions, but the alternative is
1191                  * a gratuitous RPC to the MDS.  We already rely on the client
1192                  * to do read/write/truncate permission checks, so is mtime OK?
1193                  */
1194                 if (ia_valid & (ATTR_MTIME | ATTR_ATIME)) {
1195                         /* from sys_utime() */
1196                         if (!(ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))) {
1197                                 if (current->fsuid != inode->i_uid &&
1198                                     (rc = ll_permission(inode, MAY_WRITE, NULL)) != 0)
1199                                         RETURN(rc);
1200                         } else {
1201                                 /* from inode_change_ok() */
1202                                 if (current->fsuid != inode->i_uid &&
1203                                     !capable(CAP_FOWNER))
1204                                         RETURN(-EPERM);
1205                         }
1206                 }
1207
1208                 /* won't invoke vmtruncate, as we already cleared ATTR_SIZE */
1209                 err = inode_setattr(inode, attr);
1210                 /*
1211                  * assigning inode_setattr() to @err to disable warning that
1212                  * function's result should be checked by by caller. error is
1213                  * impossible here, as vmtruncate() control path is disabled.
1214                  */
1215         }
1216
1217         /* We really need to get our PW lock before we change inode->i_size.
1218          * If we don't we can race with other i_size updaters on our node, like
1219          * ll_file_read.  We can also race with i_size propogation to other
1220          * nodes through dirtying and writeback of final cached pages.  This
1221          * last one is especially bad for racing o_append users on other
1222          * nodes. */
1223         if (ia_valid & ATTR_SIZE) {
1224                 ldlm_policy_data_t policy = { .l_extent = {attr->ia_size,
1225                                                            OBD_OBJECT_EOF } };
1226                 struct lustre_handle lockh = { 0 };
1227                 int err, ast_flags = 0;
1228                 /* XXX when we fix the AST intents to pass the discard-range
1229                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1230                  * XXX here. */
1231                 if (attr->ia_size == 0)
1232                         ast_flags = LDLM_AST_DISCARD_DATA;
1233
1234                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy, &lockh,
1235                                     ast_flags, &ll_i2sbi(inode)->ll_seek_stime);
1236
1237                 if (rc != 0)
1238                         RETURN(rc);
1239
1240                 down(&lli->lli_size_sem);
1241                 lli->lli_size_pid = current->pid;
1242                 rc = vmtruncate(inode, attr->ia_size);
1243                 if (rc != 0) {
1244                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1245                         lli->lli_size_pid = 0;
1246                         up(&lli->lli_size_sem);
1247                 }
1248
1249                 err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1250                 if (err) {
1251                         CERROR("ll_extent_unlock failed: %d\n", err);
1252                         if (!rc)
1253                                 rc = err;
1254                 }
1255         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET | ATTR_UID | ATTR_GID)) {
1256                 struct obdo *oa = NULL;
1257
1258                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1259                        inode->i_ino, LTIME_S(attr->ia_mtime));
1260
1261                 oa = obdo_alloc();
1262                 if (oa == NULL)
1263                         RETURN(-ENOMEM);
1264
1265                 oa->o_id = lsm->lsm_object_id;
1266                 oa->o_gr = lsm->lsm_object_gr;
1267                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1268
1269                 /* adding uid and gid, needed for quota */
1270                 if (ia_valid & ATTR_UID) {
1271                         oa->o_uid = inode->i_uid;
1272                         oa->o_valid |= OBD_MD_FLUID;
1273                 }
1274
1275                 if (ia_valid & ATTR_GID) {
1276                         oa->o_gid = inode->i_gid;
1277                         oa->o_valid |= OBD_MD_FLGID;
1278                 }
1279
1280                 /* putting there also fid, needed for quota too. */
1281                 memcpy(obdo_id(oa), &lli->lli_id, sizeof(lli->lli_id));
1282                 oa->o_valid |= OBD_MD_FLINLINE;
1283
1284                 obdo_from_inode(oa, inode, OBD_MD_FLTYPE | OBD_MD_FLATIME |
1285                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME);
1286                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
1287                 obdo_free(oa);
1288                 if (rc)
1289                         CERROR("obd_setattr fails: rc = %d\n", rc);
1290         }
1291
1292         RETURN(rc);
1293 }
1294
1295 int ll_setattr(struct dentry *de, struct iattr *attr)
1296 {
1297         LASSERT(de->d_inode);
1298         return ll_setattr_raw(de->d_inode, attr);
1299 }
1300
1301 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1302                        unsigned long max_age)
1303 {
1304         struct ll_sb_info *sbi = ll_s2sbi(sb);
1305         struct obd_statfs obd_osfs;
1306         int rc;
1307         ENTRY;
1308
1309         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age);
1310         if (rc) {
1311                 CERROR("obd_statfs fails: rc = %d\n", rc);
1312                 RETURN(rc);
1313         }
1314
1315         osfs->os_type = sb->s_magic;
1316
1317         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1318                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1319
1320         rc = obd_statfs(class_exp2obd(sbi->ll_dt_exp), &obd_osfs, max_age);
1321         if (rc) {
1322                 CERROR("obd_statfs fails: rc = %d\n", rc);
1323                 RETURN(rc);
1324         }
1325
1326         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1327                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1328                obd_osfs.os_files);
1329
1330         osfs->os_blocks = obd_osfs.os_blocks;
1331         osfs->os_bfree = obd_osfs.os_bfree;
1332         osfs->os_bavail = obd_osfs.os_bavail;
1333
1334         /* If we don't have as many objects free on the OST as inodes
1335          * on the MDS, we reduce the total number of inodes to
1336          * compensate, so that the "inodes in use" number is correct.
1337          */
1338         if (obd_osfs.os_ffree < osfs->os_ffree) {
1339                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1340                         obd_osfs.os_ffree;
1341                 osfs->os_ffree = obd_osfs.os_ffree;
1342         }
1343
1344         RETURN(rc);
1345 }
1346
1347 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1348 {
1349         struct obd_statfs osfs;
1350         int rc;
1351
1352         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p\n", sb);
1353         lprocfs_counter_incr(ll_s2sbi(sb)->ll_stats, LPROC_LL_STAFS);
1354
1355         /* For now we will always get up-to-date statfs values, but in the
1356          * future we may allow some amount of caching on the client (e.g.
1357          * from QOS or lprocfs updates). */
1358         rc = ll_statfs_internal(sb, &osfs, jiffies - 1);
1359         if (rc)
1360                 return rc;
1361
1362         statfs_unpack(sfs, &osfs);
1363
1364         if (sizeof(sfs->f_blocks) == 4) {
1365                 while (osfs.os_blocks > ~0UL) {
1366                         sfs->f_bsize <<= 1;
1367
1368                         osfs.os_blocks >>= 1;
1369                         osfs.os_bfree >>= 1;
1370                         osfs.os_bavail >>= 1;
1371                 }
1372         }
1373
1374         sfs->f_blocks = osfs.os_blocks;
1375         sfs->f_bfree = osfs.os_bfree;
1376         sfs->f_bavail = osfs.os_bavail;
1377
1378         return 0;
1379 }
1380
1381
1382 /********************************
1383  * remote acl                   *
1384  ********************************/
1385
1386 static struct remote_acl *remote_acl_alloc(void)
1387 {
1388         struct remote_acl *racl;
1389         int i;
1390
1391         OBD_ALLOC(racl, sizeof(*racl));
1392         if (!racl)
1393                 return NULL;
1394
1395         spin_lock_init(&racl->ra_lock);
1396         init_MUTEX(&racl->ra_update_sem);
1397
1398         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++)
1399                 INIT_LIST_HEAD(&racl->ra_perm_cache[i]);
1400
1401         return racl;
1402 }
1403
1404 /*
1405  * caller should guarantee no race here.
1406  */
1407 static void remote_perm_flush_xperms(struct lustre_remote_perm *perm)
1408 {
1409         struct remote_perm_setxid *xperm;
1410
1411         while (!list_empty(&perm->lrp_setxid_perms)) {
1412                 xperm = list_entry(perm->lrp_setxid_perms.next,
1413                                    struct remote_perm_setxid,
1414                                    list);
1415                 list_del(&xperm->list);
1416                 OBD_FREE(xperm, sizeof(*xperm));
1417         }
1418 }
1419
1420 /*
1421  * caller should guarantee no race here.
1422  */
1423 static void remote_acl_flush(struct remote_acl *racl)
1424 {
1425         struct list_head *head;
1426         struct lustre_remote_perm *perm, *tmp;
1427         int i;
1428
1429         for (i = 0; i < REMOTE_ACL_HASHSIZE; i++) {
1430                 head = &racl->ra_perm_cache[i];
1431
1432                 list_for_each_entry_safe(perm, tmp, head, lrp_list) {
1433                         remote_perm_flush_xperms(perm);
1434                         list_del(&perm->lrp_list);
1435                         OBD_FREE(perm, sizeof(*perm));
1436                 }
1437         }
1438 }
1439
1440 static void remote_acl_free(struct remote_acl *racl)
1441 {
1442         if (!racl)
1443                 return;
1444
1445         down(&racl->ra_update_sem);
1446         spin_lock(&racl->ra_lock);
1447         remote_acl_flush(racl);
1448         spin_unlock(&racl->ra_lock);
1449         up(&racl->ra_update_sem);
1450
1451         OBD_FREE(racl, sizeof(*racl));
1452 }
1453
1454 static inline int remote_acl_hashfunc(__u32 id)
1455 {
1456         return (id & (REMOTE_ACL_HASHSIZE - 1));
1457 }
1458
1459 static
1460 int __remote_acl_check(struct remote_acl *racl, unsigned int *perm)
1461 {
1462         struct list_head *head;
1463         struct lustre_remote_perm *lperm;
1464         struct remote_perm_setxid *xperm;
1465         int found = 0, rc = -ENOENT;
1466
1467         LASSERT(racl);
1468         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1469         spin_lock(&racl->ra_lock);
1470
1471         list_for_each_entry(lperm, head, lrp_list) {
1472                 if (lperm->lrp_auth_uid == current->uid) {
1473                         found = 1;
1474                         break;
1475                 }
1476         }
1477
1478         if (!found)
1479                 goto out;
1480
1481         if (lperm->lrp_auth_uid == current->fsuid &&
1482             lperm->lrp_auth_gid == current->fsgid) {
1483                 if (lperm->lrp_valid) {
1484                         *perm = lperm->lrp_perm;
1485                         rc = 0;
1486                 }
1487                 goto out;
1488         } else if ((!lperm->lrp_setuid &&
1489                     lperm->lrp_auth_uid != current->fsuid) ||
1490                    (!lperm->lrp_setgid &&
1491                     lperm->lrp_auth_gid != current->fsgid))  {
1492                 *perm = 0;
1493                 rc = 0;
1494                 goto out;
1495         }
1496
1497         list_for_each_entry(xperm, &lperm->lrp_setxid_perms, list) {
1498                 if (xperm->uid == current->fsuid &&
1499                     xperm->gid == current->fsgid) {
1500                         *perm = xperm->perm;
1501                         rc = 0;
1502                         goto out;
1503                 }
1504         }
1505
1506 out:
1507         spin_unlock(&racl->ra_lock);
1508         return rc;
1509 }
1510
1511 static
1512 int __remote_acl_update(struct remote_acl *racl,
1513                         struct mds_remote_perm *mperm,
1514                         struct lustre_remote_perm *lperm,
1515                         struct remote_perm_setxid *xperm)
1516 {
1517         struct list_head *head;
1518         struct lustre_remote_perm *lp;
1519         struct remote_perm_setxid *xp;
1520         int found = 0, setuid = 0, setgid = 0;
1521
1522         LASSERT(racl);
1523         LASSERT(mperm);
1524         LASSERT(lperm);
1525         LASSERT(current->uid == mperm->mrp_auth_uid);
1526
1527         if (current->fsuid != mperm->mrp_auth_uid)
1528                 setuid = 1;
1529         if (current->fsgid != mperm->mrp_auth_gid)
1530                 setgid = 1;
1531
1532         head = &racl->ra_perm_cache[remote_acl_hashfunc(current->uid)];
1533         spin_lock(&racl->ra_lock);
1534
1535         list_for_each_entry(lp, head, lrp_list) {
1536                 if (lp->lrp_auth_uid == current->uid) {
1537                         found = 1;
1538                         break;
1539                 }
1540         }
1541
1542         if (found) {
1543                 OBD_FREE(lperm, sizeof(*lperm));
1544
1545                 if (!lp->lrp_valid && !setuid && !setgid) {
1546                         lp->lrp_perm = mperm->mrp_perm;
1547                         lp->lrp_valid = 1;
1548                 }
1549
1550                 /* sanity check for changes of setxid rules */
1551                 if ((lp->lrp_setuid != 0) != (mperm->mrp_allow_setuid != 0)) {
1552                         CWARN("setuid changes: %d => %d\n",
1553                               (lp->lrp_setuid != 0),
1554                               (mperm->mrp_allow_setuid != 0));
1555                         lp->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1556                 }
1557
1558                 if ((lp->lrp_setgid != 0) != (mperm->mrp_allow_setgid != 0)) {
1559                         CWARN("setgid changes: %d => %d\n",
1560                               (lp->lrp_setgid != 0),
1561                               (mperm->mrp_allow_setgid != 0));
1562                         lp->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1563                 }
1564
1565                 if (!lp->lrp_setuid && !lp->lrp_setgid &&
1566                     !list_empty(&lp->lrp_setxid_perms)) {
1567                         remote_perm_flush_xperms(lp);
1568                 }
1569         } else {
1570                 /* initialize lperm and linked into hashtable
1571                  */
1572                 INIT_LIST_HEAD(&lperm->lrp_setxid_perms);
1573                 lperm->lrp_auth_uid = mperm->mrp_auth_uid;
1574                 lperm->lrp_auth_gid = mperm->mrp_auth_gid;
1575                 lperm->lrp_setuid = (mperm->mrp_allow_setuid != 0);
1576                 lperm->lrp_setgid = (mperm->mrp_allow_setgid != 0);
1577                 list_add(&lperm->lrp_list, head);
1578
1579                 if (!setuid && !setgid) {
1580                         /* in this case, i'm the authenticated user,
1581                          * and mrp_perm is for me.
1582                          */
1583                         lperm->lrp_perm = mperm->mrp_perm;
1584                         lperm->lrp_valid = 1;
1585                         spin_unlock(&racl->ra_lock);
1586
1587                         if (xperm)
1588                                 OBD_FREE(xperm, sizeof(*xperm));
1589                         return 0;
1590                 }
1591
1592                 lp = lperm;
1593                 /* fall through */
1594         }
1595
1596         LASSERT(lp->lrp_setuid || lp->lrp_setgid ||
1597                 list_empty(&lp->lrp_setxid_perms));
1598
1599         /* if no xperm supplied, we are all done here */
1600         if (!xperm) {
1601                 spin_unlock(&racl->ra_lock);
1602                 return 0;
1603         }
1604
1605         /* whether we allow setuid/setgid */
1606         if ((!lp->lrp_setuid && setuid) || (!lp->lrp_setgid && setgid)) {
1607                 OBD_FREE(xperm, sizeof(*xperm));
1608                 spin_unlock(&racl->ra_lock);
1609                 return 0;
1610         }
1611
1612         /* traverse xperm list */
1613         list_for_each_entry(xp, &lp->lrp_setxid_perms, list) {
1614                 if (xp->uid == current->fsuid &&
1615                     xp->gid == current->fsgid) {
1616                         if (xp->perm != mperm->mrp_perm) {
1617                                 /* actually this should not happen */
1618                                 CWARN("perm changed: %o => %o\n",
1619                                       xp->perm, mperm->mrp_perm);
1620                                 xp->perm = mperm->mrp_perm;
1621                         }
1622                         OBD_FREE(xperm, sizeof(*xperm));
1623                         spin_unlock(&racl->ra_lock);
1624                         return 0;
1625                 }
1626         }
1627
1628         /* finally insert this xperm */
1629         xperm->uid = current->fsuid;
1630         xperm->gid = current->fsgid;
1631         xperm->perm = mperm->mrp_perm;
1632         list_add(&xperm->list, &lp->lrp_setxid_perms);
1633
1634         spin_unlock(&racl->ra_lock);
1635         return 0;
1636 }
1637
1638 /*
1639  * remote_acl semaphore must be held by caller
1640  */
1641 static
1642 int remote_acl_update_locked(struct remote_acl *racl,
1643                              struct mds_remote_perm *mperm)
1644 {
1645         struct lustre_remote_perm *lperm;
1646         struct remote_perm_setxid *xperm;
1647         int setuid = 0, setgid = 0;
1648
1649         might_sleep();
1650
1651         if (current->uid != mperm->mrp_auth_uid) {
1652                 CERROR("current uid %u while authenticated as %u\n",
1653                        current->uid, mperm->mrp_auth_uid);
1654                 return -EINVAL;
1655         }
1656
1657         if (current->fsuid != mperm->mrp_auth_uid)
1658                 setuid = 1;
1659         if (current->fsgid == mperm->mrp_auth_gid)
1660                 setgid = 1;
1661
1662         OBD_ALLOC(lperm, sizeof(*lperm));
1663         if (!lperm)
1664                 return -ENOMEM;
1665
1666         if ((setuid || setgid) &&
1667             !(setuid && !mperm->mrp_allow_setuid) &&
1668             !(setgid && !mperm->mrp_allow_setgid)) {
1669                 OBD_ALLOC(xperm, sizeof(*xperm));
1670                 if (!xperm) {
1671                         OBD_FREE(lperm, sizeof(*lperm));
1672                         return -ENOMEM;
1673                 }
1674         } else
1675                 xperm = NULL;
1676
1677         return __remote_acl_update(racl, mperm, lperm, xperm);
1678 }
1679
1680 /*
1681  * return -EACCES at any error cases
1682  */
1683 int ll_remote_acl_permission(struct inode *inode, int mode)
1684 {
1685         struct ll_sb_info *sbi = ll_i2sbi(inode);
1686         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1687         struct ptlrpc_request *req = NULL;
1688         struct lustre_id id;
1689         struct mds_remote_perm *mperm;
1690         int rc = -EACCES, perm;
1691
1692         if (!racl)
1693                 return -EACCES;
1694
1695         if (__remote_acl_check(racl, &perm) == 0) {
1696                 return ((perm & mode) == mode ? 0 : -EACCES);
1697         }
1698
1699         might_sleep();
1700
1701         /* doing update
1702          */
1703         down(&racl->ra_update_sem);
1704
1705         /* we might lose the race when obtain semaphore,
1706          * so check again.
1707          */
1708         if (__remote_acl_check(racl, &perm) == 0) {
1709                 if ((perm & mode) == mode)
1710                         rc = 0;
1711                 goto out;
1712         }
1713
1714         /* really fetch from mds
1715          */
1716         ll_inode2id(&id, inode);
1717         if (md_access_check(sbi->ll_md_exp, &id, &req))
1718                 goto out;
1719
1720         /* status non-zero indicate there's more apparent error
1721          * detected by mds, e.g. didn't allow this user at all.
1722          * we simply ignore and didn't cache it.
1723          */
1724         if (req->rq_repmsg->status)
1725                 goto out;
1726
1727         mperm = lustre_swab_repbuf(req, 1, sizeof(*mperm),
1728                                    lustre_swab_remote_perm);
1729         LASSERT(mperm);
1730         LASSERT_REPSWABBED(req, 1);
1731
1732         if ((mperm->mrp_perm & mode) == mode)
1733                 rc = 0;
1734
1735         remote_acl_update_locked(racl, mperm);
1736 out:
1737         if (req)
1738                 ptlrpc_req_finished(req);
1739
1740         up(&racl->ra_update_sem);
1741         return rc;
1742 }
1743
1744 int ll_remote_acl_update(struct inode *inode, struct mds_remote_perm *perm)
1745 {
1746         struct remote_acl *racl = ll_i2info(inode)->lli_remote_acl;
1747         int rc;
1748
1749         LASSERT(perm);
1750
1751         if (!racl)
1752                 return -EACCES;
1753
1754         down(&racl->ra_update_sem);
1755         rc = remote_acl_update_locked(racl, perm);
1756         up(&racl->ra_update_sem);
1757
1758         return rc;
1759 }
1760
1761 void ll_inode_invalidate_acl(struct inode *inode)
1762 {
1763         struct ll_sb_info *sbi = ll_i2sbi(inode);
1764         struct ll_inode_info *lli = ll_i2info(inode);
1765
1766         if (sbi->ll_remote) {
1767                 struct remote_acl *racl = lli->lli_remote_acl;
1768
1769                 LASSERT(!lli->lli_posix_acl);
1770                 if (racl) {
1771                         down(&racl->ra_update_sem);
1772                         spin_lock(&racl->ra_lock);
1773                         remote_acl_flush(lli->lli_remote_acl);
1774                         spin_unlock(&racl->ra_lock);
1775                         up(&racl->ra_update_sem);
1776                 }
1777         } else {
1778                 LASSERT(!lli->lli_remote_acl);
1779                 spin_lock(&lli->lli_lock);
1780                 posix_acl_release(lli->lli_posix_acl);
1781                 lli->lli_posix_acl = NULL;
1782                 spin_unlock(&lli->lli_lock);
1783         }
1784 }
1785
1786 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1787 {
1788         struct ll_inode_info *lli = ll_i2info(inode);
1789         struct lov_stripe_md *lsm = md->lsm;
1790         struct mds_body *body = md->body;
1791         struct mea *mea = md->mea;
1792         struct posix_acl *posix_acl = md->posix_acl;
1793         struct ll_sb_info *sbi = ll_i2sbi(inode);
1794         ENTRY;
1795
1796         LASSERT((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1797
1798         if (md->lsm && md->lsm->lsm_magic != LOV_MAGIC) {
1799                 /* check for default striping info for dir. */
1800                 LASSERT((mea != NULL) == ((body->valid & OBD_MD_FLDIREA) != 0));
1801         }
1802
1803         if (lsm != NULL) {
1804                 LASSERT(lsm->lsm_object_gr > 0);
1805                 if (lli->lli_smd == NULL) {
1806                         lli->lli_smd = lsm;
1807                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1808                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1809                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1810                 } else {
1811                         int i;
1812                         if (memcmp(lli->lli_smd, lsm, sizeof(*lsm))) {
1813                                 CERROR("lsm mismatch for inode %ld\n",
1814                                        inode->i_ino);
1815                                 CERROR("lli_smd:\n");
1816                                 dump_lsm(D_ERROR, lli->lli_smd);
1817                                 CERROR("lsm:\n");
1818                                 dump_lsm(D_ERROR, lsm);
1819                                 LBUG();
1820                         }
1821                         /* XXX FIXME -- We should decide on a safer (atomic) and
1822                          * more elegant way to update the lsm */
1823                         for (i = 0; i < lsm->lsm_stripe_count; i++) {
1824                                 lli->lli_smd->lsm_oinfo[i].loi_id =
1825                                         lsm->lsm_oinfo[i].loi_id;
1826                                 lli->lli_smd->lsm_oinfo[i].loi_gr =
1827                                         lsm->lsm_oinfo[i].loi_gr;
1828                                 lli->lli_smd->lsm_oinfo[i].loi_ost_idx =
1829                                         lsm->lsm_oinfo[i].loi_ost_idx;
1830                                 lli->lli_smd->lsm_oinfo[i].loi_ost_gen =
1831                                         lsm->lsm_oinfo[i].loi_ost_gen;
1832                         }
1833                 }
1834                 /* bug 2844 - limit i_blksize for broken user-space apps */
1835                 LASSERTF(lsm->lsm_xfersize != 0, "%lu\n", lsm->lsm_xfersize);
1836                 inode->i_blksize = min(lsm->lsm_xfersize, LL_MAX_BLKSIZE);
1837                 if (lli->lli_smd != lsm)
1838                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1839         }
1840
1841         if (mea != NULL) {
1842                 if (lli->lli_mea == NULL) {
1843                         lli->lli_mea = mea;
1844                 } else {
1845                         if (memcmp(lli->lli_mea, mea, body->eadatasize)) {
1846                                 CERROR("mea mismatch for inode %lu\n",
1847                                         inode->i_ino);
1848                                 LBUG();
1849                         }
1850                 }
1851                 if (lli->lli_mea != mea)
1852                         obd_free_memmd(ll_i2mdexp(inode),
1853                                        (struct lov_stripe_md **) &mea);
1854         }
1855
1856         if (body->valid & OBD_MD_FID)
1857                 id_assign_fid(&lli->lli_id, &body->id1);
1858
1859         if (body->valid & OBD_MD_FLID)
1860                 id_ino(&lli->lli_id) = id_ino(&body->id1);
1861
1862         if (body->valid & OBD_MD_FLGENER)
1863                 id_gen(&lli->lli_id) = id_gen(&body->id1);
1864
1865         /* local/remote ACL */
1866         if (sbi->ll_remote) {
1867                 LASSERT(md->posix_acl == NULL);
1868                 if (md->remote_perm) {
1869                         ll_remote_acl_update(inode, md->remote_perm);
1870                         OBD_FREE(md->remote_perm, sizeof(*md->remote_perm));
1871                         md->remote_perm = NULL;
1872                 }
1873         } else {
1874                 LASSERT(md->remote_perm == NULL);
1875                 spin_lock(&lli->lli_lock);
1876                 if (posix_acl != NULL) {
1877                         if (lli->lli_posix_acl != NULL)
1878                                 posix_acl_release(lli->lli_posix_acl);
1879                         lli->lli_posix_acl = posix_acl;
1880                 }
1881                 spin_unlock(&lli->lli_lock);
1882         }
1883
1884         if (body->valid & OBD_MD_FLID)
1885                 inode->i_ino = id_ino(&body->id1);
1886         if (body->valid & OBD_MD_FLGENER)
1887                 inode->i_generation = id_gen(&body->id1);
1888         if (body->valid & OBD_MD_FLATIME)
1889                 LTIME_S(inode->i_atime) = body->atime;
1890         if (body->valid & OBD_MD_FLMTIME &&
1891             body->mtime > LTIME_S(inode->i_mtime)) {
1892                 CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %u\n",
1893                        inode->i_ino, LTIME_S(inode->i_mtime), body->mtime);
1894                 LTIME_S(inode->i_mtime) = body->mtime;
1895         }
1896         if (body->valid & OBD_MD_FLCTIME &&
1897             body->ctime > LTIME_S(inode->i_ctime))
1898                 LTIME_S(inode->i_ctime) = body->ctime;
1899         if (body->valid & OBD_MD_FLMODE) {
1900                 inode->i_mode = (inode->i_mode & S_IFMT) |
1901                         (body->mode & ~S_IFMT);
1902         }
1903         if (body->valid & OBD_MD_FLTYPE) {
1904                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1905                         (body->mode & S_IFMT);
1906         }
1907         if (body->valid & OBD_MD_FLUID)
1908                 inode->i_uid = body->uid;
1909         if (body->valid & OBD_MD_FLGID)
1910                 inode->i_gid = body->gid;
1911         if (body->valid & OBD_MD_FLFLAGS)
1912                 inode->i_flags = body->flags;
1913         if (body->valid & OBD_MD_FLNLINK)
1914                 inode->i_nlink = body->nlink;
1915         if (body->valid & OBD_MD_FLRDEV)
1916 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1917                 inode->i_rdev = body->rdev;
1918 #else
1919                 inode->i_rdev = old_decode_dev(body->rdev);
1920 #endif
1921         if (body->valid & OBD_MD_FLSIZE)
1922                 inode->i_size = body->size;
1923         if (body->valid & OBD_MD_FLBLOCKS)
1924                 inode->i_blocks = body->blocks;
1925
1926         if (body->valid & OBD_MD_FLSIZE)
1927                 set_bit(LLI_F_HAVE_MDS_SIZE_LOCK, &lli->lli_flags);
1928
1929 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1930         inode->i_dev = (kdev_t)id_group(&lli->lli_id);
1931 #endif
1932         LASSERT(id_fid(&lli->lli_id) != 0);
1933 }
1934
1935 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1936 static struct backing_dev_info ll_backing_dev_info = {
1937         .ra_pages       = 0,    /* No readahead */
1938         .memory_backed  = 0,    /* Does contribute to dirty memory */
1939 };
1940 #endif
1941
1942 void ll_read_inode2(struct inode *inode, void *opaque)
1943 {
1944         struct lustre_md *md = opaque;
1945         struct ll_inode_info *lli = ll_i2info(inode);
1946         ENTRY;
1947
1948         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1949                inode->i_generation, inode);
1950
1951         ll_lli_init(lli);
1952
1953         LASSERT(!lli->lli_smd);
1954
1955         if (ll_i2sbi(inode)->ll_remote) {
1956                 lli->lli_remote_acl = remote_acl_alloc();
1957                 /* if failed alloc, nobody will be able to access this inode */
1958         }
1959
1960         /* Core attributes from the MDS first.  This is a new inode, and
1961          * the VFS doesn't zero times in the core inode so we have to do
1962          * it ourselves.  They will be overwritten by either MDS or OST
1963          * attributes - we just need to make sure they aren't newer. */
1964         LTIME_S(inode->i_mtime) = 0;
1965         LTIME_S(inode->i_atime) = 0;
1966         LTIME_S(inode->i_ctime) = 0;
1967
1968         inode->i_rdev = 0;
1969         ll_update_inode(inode, md);
1970
1971         /* OIDEBUG(inode); */
1972
1973         if (S_ISREG(inode->i_mode)) {
1974                 inode->i_op = &ll_file_inode_operations;
1975                 inode->i_fop = &ll_file_operations;
1976                 inode->i_mapping->a_ops = &ll_aops;
1977                 EXIT;
1978         } else if (S_ISDIR(inode->i_mode)) {
1979                 inode->i_op = &ll_dir_inode_operations;
1980                 inode->i_fop = &ll_dir_operations;
1981                 inode->i_mapping->a_ops = &ll_dir_aops;
1982                 EXIT;
1983         } else if (S_ISLNK(inode->i_mode)) {
1984                 inode->i_op = &ll_fast_symlink_inode_operations;
1985                 EXIT;
1986         } else {
1987                 inode->i_op = &ll_special_inode_operations;
1988
1989 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
1990                 init_special_inode(inode, inode->i_mode,
1991                                    kdev_t_to_nr(inode->i_rdev));
1992
1993                 /* initializing backing dev info. */
1994                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1995 #else
1996                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
1997 #endif
1998                 lli->ll_save_ifop = inode->i_fop;
1999
2000                 if (S_ISCHR(inode->i_mode))
2001                         inode->i_fop = &ll_special_chr_inode_fops;
2002                 else if (S_ISBLK(inode->i_mode))
2003                         inode->i_fop = &ll_special_blk_inode_fops;
2004                 else if (S_ISFIFO(inode->i_mode))
2005                         inode->i_fop = &ll_special_fifo_inode_fops;
2006                 else if (S_ISSOCK(inode->i_mode))
2007                         inode->i_fop = &ll_special_sock_inode_fops;
2008
2009                 CWARN("saved %p, replaced with %p\n", lli->ll_save_ifop,
2010                       inode->i_fop);
2011
2012                 if (lli->ll_save_ifop->owner) {
2013                         CWARN("%p has owner %p\n", lli->ll_save_ifop,
2014                               lli->ll_save_ifop->owner);
2015                 }
2016                 EXIT;
2017         }
2018 }
2019
2020 void ll_delete_inode(struct inode *inode)
2021 {
2022         struct ll_sb_info *sbi = ll_i2sbi(inode);
2023         struct lustre_id id;
2024         int rc;
2025         ENTRY;
2026
2027         ll_inode2id(&id, inode);
2028
2029         rc = md_delete_inode(sbi->ll_md_exp, &id);
2030         if (rc) {
2031                 CERROR("md_delete_inode() failed, error %d\n",
2032                        rc);
2033         }
2034
2035         clear_inode(inode);
2036         EXIT;
2037 }
2038
2039 int ll_iocontrol(struct inode *inode, struct file *file,
2040                  unsigned int cmd, unsigned long arg)
2041 {
2042         struct ll_sb_info *sbi = ll_i2sbi(inode);
2043         struct ptlrpc_request *req = NULL;
2044         int rc, flags = 0;
2045         ENTRY;
2046
2047         switch(cmd) {
2048         case EXT3_IOC_GETFLAGS: {
2049                 struct lustre_id id;
2050                 __u64 valid = OBD_MD_FLFLAGS;
2051                 struct mds_body *body;
2052
2053                 ll_inode2id(&id, inode);
2054                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, 0, 0, &req);
2055                 if (rc) {
2056                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2057                         RETURN(-abs(rc));
2058                 }
2059
2060                 body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
2061
2062                 if (body->flags & S_APPEND)
2063                         flags |= EXT3_APPEND_FL;
2064                 if (body->flags & S_IMMUTABLE)
2065                         flags |= EXT3_IMMUTABLE_FL;
2066                 if (body->flags & S_NOATIME)
2067                         flags |= EXT3_NOATIME_FL;
2068
2069                 ptlrpc_req_finished (req);
2070
2071                 RETURN(put_user(flags, (int *)arg));
2072         }
2073         case EXT3_IOC_SETFLAGS: {
2074                 struct mdc_op_data *op_data;
2075                 struct iattr attr;
2076                 struct obdo *oa;
2077                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2078
2079                 if (get_user(flags, (int *)arg))
2080                         RETURN(-EFAULT);
2081
2082                 oa = obdo_alloc();
2083                 if (!oa)
2084                         RETURN(-ENOMEM);
2085
2086                 OBD_ALLOC(op_data, sizeof(*op_data));
2087                 if (op_data == NULL) {
2088                         obdo_free(oa);
2089                         RETURN(-ENOMEM);
2090                 }
2091                 ll_prepare_mdc_data(op_data, inode, NULL, NULL, 0, 0);
2092
2093                 memset(&attr, 0x0, sizeof(attr));
2094                 attr.ia_attr_flags = flags;
2095                 attr.ia_valid |= ATTR_ATTR_FLAG;
2096
2097                 rc = md_setattr(sbi->ll_md_exp, op_data,
2098                                 &attr, NULL, 0, NULL, 0, &req);
2099                 OBD_FREE(op_data, sizeof(*op_data));
2100                 if (rc) {
2101                         ptlrpc_req_finished(req);
2102                         if (rc != -EPERM && rc != -EACCES)
2103                                 CERROR("md_setattr fails: rc = %d\n", rc);
2104                         obdo_free(oa);
2105                         RETURN(rc);
2106                 }
2107                 ptlrpc_req_finished(req);
2108
2109                 oa->o_id = lsm->lsm_object_id;
2110                 oa->o_gr = lsm->lsm_object_gr;
2111                 oa->o_flags = flags;
2112                 oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS | OBD_MD_FLGROUP;
2113
2114                 rc = obd_setattr(sbi->ll_dt_exp, oa, lsm, NULL);
2115                 obdo_free(oa);
2116                 if (rc) {
2117                         if (rc != -EPERM && rc != -EACCES)
2118                                 CERROR("md_setattr fails: rc = %d\n", rc);
2119                         RETURN(rc);
2120                 }
2121
2122                 if (flags & EXT3_APPEND_FL)
2123                         inode->i_flags |= S_APPEND;
2124                 else
2125                         inode->i_flags &= ~S_APPEND;
2126                 if (flags & EXT3_IMMUTABLE_FL)
2127                         inode->i_flags |= S_IMMUTABLE;
2128                 else
2129                         inode->i_flags &= ~S_IMMUTABLE;
2130                 if (flags & EXT3_NOATIME_FL)
2131                         inode->i_flags |= S_NOATIME;
2132                 else
2133                         inode->i_flags &= ~S_NOATIME;
2134
2135                 RETURN(0);
2136         }
2137         default:
2138                 RETURN(-ENOSYS);
2139         }
2140
2141         RETURN(0);
2142 }
2143
2144 /* this is only called in the case of forced umount. */
2145 void ll_umount_begin(struct super_block *sb)
2146 {
2147         struct ll_sb_info *sbi = ll_s2sbi(sb);
2148         struct obd_ioctl_data ioc_data = { 0 };
2149         struct obd_device *obd;
2150         ENTRY;
2151
2152         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2153                sb->s_count, atomic_read(&sb->s_active));
2154
2155         obd = class_exp2obd(sbi->ll_md_exp);
2156         if (obd == NULL) {
2157                 CERROR("Invalid MDC connection handle "LPX64"\n",
2158                        sbi->ll_md_exp->exp_handle.h_cookie);
2159                 EXIT;
2160                 return;
2161         }
2162         obd->obd_no_recov = 1;
2163         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2164                       sizeof(ioc_data), &ioc_data, NULL);
2165
2166         obd = class_exp2obd(sbi->ll_dt_exp);
2167         if (obd == NULL) {
2168                 CERROR("Invalid LOV connection handle "LPX64"\n",
2169                        sbi->ll_dt_exp->exp_handle.h_cookie);
2170                 EXIT;
2171                 return;
2172         }
2173
2174         obd->obd_no_recov = 1;
2175         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2176                       sizeof(ioc_data), &ioc_data, NULL);
2177
2178         /*
2179          * really, we'd like to wait until there are no requests outstanding,
2180          * and then continue.  For now, we just invalidate the requests,
2181          * schedule, and hope.
2182          */
2183         schedule();
2184
2185         EXIT;
2186 }
2187
2188 int ll_prep_inode(struct obd_export *dt_exp, struct obd_export *md_exp,
2189                   struct inode **inode, struct ptlrpc_request *req,
2190                   int offset, struct super_block *sb)
2191 {
2192         struct lustre_md md;
2193         int rc = 0;
2194
2195         rc = mdc_req2lustre_md(md_exp, req, offset, dt_exp, &md);
2196         if (rc)
2197                 RETURN(rc);
2198
2199         if (*inode) {
2200                 ll_update_inode(*inode, &md);
2201         } else {
2202                 LASSERT(sb);
2203                 *inode = ll_iget(sb, id_ino(&md.body->id1), &md);
2204                 if (*inode == NULL || is_bad_inode(*inode)) {
2205                         /* free the lsm if we allocated one above */
2206                         if (md.lsm != NULL)
2207                                 obd_free_memmd(dt_exp, &md.lsm);
2208                         if (md.mea != NULL)
2209                                 obd_free_memmd(md_exp,
2210                                                (struct lov_stripe_md**)&md.mea);
2211                         rc = -ENOMEM;
2212                         CERROR("new_inode -fatal: rc %d\n", rc);
2213                 }
2214         }
2215
2216         RETURN(rc);
2217 }
2218
2219 int ll_show_options(struct seq_file *m, struct vfsmount *mnt)
2220 {
2221         struct ll_sb_info *sbi = ll_s2sbi(mnt->mnt_sb);
2222         struct lustre_mount_data *lmd = sbi->ll_lmd;
2223
2224         if (lmd) {
2225                 seq_printf(m, ",mds_sec=%s,oss_sec=%s",
2226                            lmd->lmd_mds_security, lmd->lmd_oss_security);
2227         }
2228         seq_printf(m, ",%s", sbi->ll_remote ? "remote" : "local");
2229         if (sbi->ll_remote && lmd) {
2230                 seq_printf(m, ",nllu=%u:%u", lmd->lmd_nllu, lmd->lmd_nllg);
2231         }
2232
2233         return 0;
2234 }
2235
2236 int ll_get_fid(struct obd_export *exp, struct lustre_id *idp,
2237                char *filename, struct lustre_id *ret)
2238 {
2239         struct ptlrpc_request *request = NULL;
2240         struct mds_body *body;
2241         int rc;
2242
2243         rc = md_getattr_lock(exp, idp, filename, strlen(filename) + 1,
2244                              OBD_MD_FID, 0, &request);
2245         if (rc < 0) {
2246                 CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
2247                        filename, rc);
2248                 return rc;
2249         }
2250
2251         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
2252         LASSERT(body != NULL);
2253         LASSERT_REPSWABBED(request, 0);
2254
2255         *ret = body->id1;
2256         ptlrpc_req_finished(request);
2257
2258         return rc;
2259 }
2260
2261 int ll_flush_cred(struct inode *inode)
2262 {
2263         struct ll_sb_info *sbi = ll_i2sbi(inode);
2264         uid_t   uid = current->fsuid;
2265         int     rc = 0;
2266
2267         /* XXX to avoid adding api, we simply use set_info() interface
2268          * to notify underlying obds. set_info() is more like a ioctl() now...
2269          */
2270         if (sbi->ll_md_exp) {
2271                 rc = obd_set_info(sbi->ll_md_exp,
2272                                   strlen("flush_cred"), "flush_cred",
2273                                   sizeof(uid), &uid);
2274                 if (rc)
2275                         return rc;
2276         }
2277
2278         if (sbi->ll_dt_exp) {
2279                 rc = obd_set_info(sbi->ll_dt_exp,
2280                                   strlen("flush_cred"), "flush_cred",
2281                                   sizeof(uid), &uid);
2282                 if (rc)
2283                         return rc;
2284         }
2285
2286         return rc;
2287 }