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) 2002-2005 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 <lustre_lite.h>
  32 #include <lustre_ha.h>
  33 #include <lustre_dlm.h>
  34 #include <lprocfs_status.h>
  35 #include <lustre_disk.h>
  36 #include <lustre_param.h>
  37 #include <lustre_log.h>
  38 #include <obd_cksum.h>
  39 #include <lustre_cache.h>
  40 #include "llite_internal.h"
  41
  42 cfs_mem_cache_t *ll_file_data_slab;
  43
  44 LIST_HEAD(ll_super_blocks);
  45 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  46
  47 extern struct address_space_operations ll_aops;
  48 extern struct address_space_operations ll_dir_aops;
  49
  50 #ifndef log2
  51 #define log2(n) ffz(~(n))
  52 #endif
  53
  54
  55 static struct ll_sb_info *ll_init_sbi(void)
  56 {
  57         struct ll_sb_info *sbi = NULL;
  58         class_uuid_t uuid;
  59         int i;
  60         ENTRY;
  61
  62         OBD_ALLOC(sbi, sizeof(*sbi));
  63         if (!sbi)
  64                 RETURN(NULL);
  65
  66         spin_lock_init(&sbi->ll_lock);
  67         spin_lock_init(&sbi->ll_lco.lco_lock);
  68         spin_lock_init(&sbi->ll_pp_extent_lock);
  69         spin_lock_init(&sbi->ll_process_lock);
  70         sbi->ll_rw_stats_on = 0;
  71         INIT_LIST_HEAD(&sbi->ll_pglist);
  72         if (num_physpages >> (20 - CFS_PAGE_SHIFT) < 512)
  73                 sbi->ll_async_page_max = num_physpages / 2;
  74         else
  75                 sbi->ll_async_page_max = (num_physpages / 4) * 3;
  76         sbi->ll_ra_info.ra_max_pages = min(num_physpages / 8,
  77                                            SBI_DEFAULT_READAHEAD_MAX);
  78         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  79                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  80         sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
  81         sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
  82         INIT_LIST_HEAD(&sbi->ll_conn_chain);
  83         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
  84
  85         ll_generate_random_uuid(uuid);
  86         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
  87         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
  88
  89         spin_lock(&ll_sb_lock);
  90         list_add_tail(&sbi->ll_list, &ll_super_blocks);
  91         spin_unlock(&ll_sb_lock);
  92
  93 #ifdef ENABLE_LLITE_CHECKSUM
  94         sbi->ll_flags |= LL_SBI_CHECKSUM;
  95 #endif
  96
  97 #ifdef HAVE_LRU_RESIZE_SUPPORT
  98         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
  99 #endif
 100
 101 #ifdef HAVE_EXPORT___IGET
 102         INIT_LIST_HEAD(&sbi->ll_deathrow);
 103         spin_lock_init(&sbi->ll_deathrow_lock);
 104 #endif
 105         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 106                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 107                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 108         }
 109
 110         /* metadata statahead is enabled by default */
 111         sbi->ll_sa_max = LL_SA_RPC_DEF;
 112
 113         RETURN(sbi);
 114 }
 115
 116 void ll_free_sbi(struct super_block *sb)
 117 {
 118         struct ll_sb_info *sbi = ll_s2sbi(sb);
 119         ENTRY;
 120
 121         if (sbi != NULL) {
 122                 spin_lock(&ll_sb_lock);
 123                 list_del(&sbi->ll_list);
 124                 spin_unlock(&ll_sb_lock);
 125                 OBD_FREE(sbi, sizeof(*sbi));
 126         }
 127         EXIT;
 128 }
 129
 130 static struct dentry_operations ll_d_root_ops = {
 131 #ifdef DCACHE_LUSTRE_INVALID
 132         .d_compare = ll_dcompare,
 133 #endif
 134 };
 135
 136 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 137  * us to make MDS RPCs with large enough reply buffers to hold the
 138  * maximum-sized (= maximum striped) EA and cookie without having to
 139  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 140 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 141 {
 142         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 143         __u32 valsize = sizeof(struct lov_desc);
 144         int rc, easize, def_easize, cookiesize;
 145         struct lov_desc desc;
 146         __u32 stripes;
 147         ENTRY;
 148
 149         rc = obd_get_info(dt_exp, strlen(KEY_LOVDESC) + 1, KEY_LOVDESC,
 150                           &valsize, &desc);
 151         if (rc)
 152                 RETURN(rc);
 153
 154         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 155         lsm.lsm_stripe_count = stripes;
 156         easize = obd_size_diskmd(dt_exp, &lsm);
 157
 158         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 159         def_easize = obd_size_diskmd(dt_exp, &lsm);
 160
 161         cookiesize = stripes * sizeof(struct llog_cookie);
 162
 163         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 164                easize, cookiesize);
 165
 166         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 167         RETURN(rc);
 168 }
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 171 {
 172         struct inode *root = 0;
 173         struct ll_sb_info *sbi = ll_s2sbi(sb);
 174         struct obd_device *obd;
 175         struct lu_fid rootfid;
 176         struct obd_capa *oc = NULL;
 177         struct obd_statfs osfs;
 178         struct ptlrpc_request *request = NULL;
 179         struct lustre_handle dt_conn = {0, };
 180         struct lustre_handle md_conn = {0, };
 181         struct obd_connect_data *data = NULL;
 182         struct lustre_md lmd;
 183         obd_valid valid;
 184         int size, err, checksum;
 185         ENTRY;
 186
 187         obd = class_name2obd(md);
 188         if (!obd) {
 189                 CERROR("MD %s: not setup or attached\n", md);
 190                 RETURN(-EINVAL);
 191         }
 192
 193         OBD_ALLOC_PTR(data);
 194         if (data == NULL)
 195                 RETURN(-ENOMEM);
 196
 197         if (proc_lustre_fs_root) {
 198                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 199                                                   dt, md);
 200                 if (err < 0)
 201                         CERROR("could not register mount in /proc/fs/lustre\n");
 202         }
 203
 204         /* indicate the features supported by this client */
 205         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 206                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 207                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 208                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 209                                   OBD_CONNECT_FID;
 210
 211 #ifdef HAVE_LRU_RESIZE_SUPPORT
 212         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 213                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 214 #endif
 215 #ifdef CONFIG_FS_POSIX_ACL
 216         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 217 #endif
 218         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 219         data->ocd_version = LUSTRE_VERSION_CODE;
 220
 221         if (sb->s_flags & MS_RDONLY)
 222                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 223         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 224                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 225
 226 #ifdef HAVE_MS_FLOCK_LOCK
 227         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 228         sb->s_flags |= MS_FLOCK_LOCK;
 229 #endif
 230
 231         if (sbi->ll_flags & LL_SBI_FLOCK)
 232                 sbi->ll_fop = &ll_file_operations_flock;
 233         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 234                 sbi->ll_fop = &ll_file_operations;
 235         else
 236                 sbi->ll_fop = &ll_file_operations_noflock;
 237
 238         /* real client */
 239         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 240         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 241                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 242                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 243         } else {
 244                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 245                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 246         }
 247
 248         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 249         if (err == -EBUSY) {
 250                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 251                                    "recovery, of which this client is not a "
 252                                    "part. Please wait for recovery to complete,"
 253                                    " abort, or time out.\n", md);
 254                 GOTO(out, err);
 255         } else if (err) {
 256                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 257                 GOTO(out, err);
 258         }
 259         sbi->ll_md_exp = class_conn2export(&md_conn);
 260
 261         err = obd_fid_init(sbi->ll_md_exp);
 262         if (err) {
 263                 CERROR("Can't init metadata layer FID infrastructure, "
 264                        "rc %d\n", err);
 265                 GOTO(out_md, err);
 266         }
 267
 268         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 269         if (err)
 270                 GOTO(out_md_fid, err);
 271
 272         size = sizeof(*data);
 273         err = obd_get_info(sbi->ll_md_exp, strlen(KEY_CONN_DATA),
 274                            KEY_CONN_DATA,  &size, data);
 275         if (err) {
 276                 CERROR("Get connect data failed: %d \n", err);
 277                 GOTO(out_md, err);
 278         }
 279
 280         LASSERT(osfs.os_bsize);
 281         sb->s_blocksize = osfs.os_bsize;
 282         sb->s_blocksize_bits = log2(osfs.os_bsize);
 283         sb->s_magic = LL_SUPER_MAGIC;
 284
 285         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 286          *         retval = in_file->f_op->sendfile(...);
 287          *         if (*ppos > max)
 288          *                 retval = -EOVERFLOW;
 289          *
 290          * it will check if *ppos is greater than max. However, max equals to
 291          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 292          * has been defined as a signed long long ineger in linux kernel. */
 293 #if BITS_PER_LONG == 64
 294         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 295 #else
 296         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 297 #endif
 298         sbi->ll_namelen = osfs.os_namelen;
 299         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 300
 301         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 302             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 303                 LCONSOLE_INFO("Disabling user_xattr feature because "
 304                               "it is not supported on the server\n");
 305                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 306         }
 307
 308         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 309 #ifdef MS_POSIXACL
 310                 sb->s_flags |= MS_POSIXACL;
 311 #endif
 312                 sbi->ll_flags |= LL_SBI_ACL;
 313         } else {
 314                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 315 #ifdef MS_POSIXACL
 316                 sb->s_flags &= ~MS_POSIXACL;
 317 #endif
 318                 sbi->ll_flags &= ~LL_SBI_ACL;
 319         }
 320
 321         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 322                 sbi->ll_flags |= LL_SBI_JOIN;
 323
 324         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 325                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 326                         /* sometimes local client claims to be remote, but mdt
 327                          * will disagree when client gss not applied. */
 328                         LCONSOLE_INFO("client claims to be remote, but server "
 329                                       "rejected, forced to be local.\n");
 330                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 331                 }
 332         } else {
 333                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 334                         /* with gss applied, remote client can not claim to be
 335                          * local, so mdt maybe force client to be remote. */
 336                         LCONSOLE_INFO("client claims to be local, but server "
 337                                       "rejected, forced to be remote.\n");
 338                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 339                 }
 340         }
 341
 342         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 343                 LCONSOLE_INFO("client enabled MDS capability!\n");
 344                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 345         }
 346
 347         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 348                 LCONSOLE_INFO("client enabled OSS capability!\n");
 349                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 350         }
 351
 352         sbi->ll_sdev_orig = sb->s_dev;
 353 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 354         /* We set sb->s_dev equal on all lustre clients in order to support
 355          * NFS export clustering.  NFSD requires that the FSID be the same
 356          * on all clients. */
 357         /* s_dev is also used in lt_compare() to compare two fs, but that is
 358          * only a node-local comparison. */
 359
 360         /* XXX: this will not work with LMV */
 361         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 362                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 363 #endif
 364
 365         obd = class_name2obd(dt);
 366         if (!obd) {
 367                 CERROR("DT %s: not setup or attached\n", dt);
 368                 GOTO(out_md_fid, err = -ENODEV);
 369         }
 370
 371         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 372                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 373                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 374                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK;
 375         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 376                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 377
 378         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 379                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 380                  * disabled by default, because it can still be enabled on the
 381                  * fly via /proc. As a consequence, we still need to come to an
 382                  * agreement on the supported algorithms at connect time */
 383                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 384
 385                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 386                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 387                 else
 388                         /* send the list of supported checksum types */
 389                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 390         }
 391
 392 #ifdef HAVE_LRU_RESIZE_SUPPORT
 393         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 394 #endif
 395         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 396                "ocd_grant: %d\n", data->ocd_connect_flags,
 397                data->ocd_version, data->ocd_grant);
 398
 399         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 400         obd->obd_upcall.onu_upcall = ll_ocd_update;
 401         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 402
 403         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 404         if (err == -EBUSY) {
 405                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 406                                    "recovery, of which this client is not a "
 407                                    "part.  Please wait for recovery to "
 408                                    "complete, abort, or time out.\n", dt);
 409                 GOTO(out_md_fid, err);
 410         } else if (err) {
 411                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 412                 GOTO(out_md_fid, err);
 413         }
 414
 415         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 416
 417         err = obd_fid_init(sbi->ll_dt_exp);
 418         if (err) {
 419                 CERROR("Can't init data layer FID infrastructure, "
 420                        "rc %d\n", err);
 421                 GOTO(out_dt, err);
 422         }
 423
 424         spin_lock(&sbi->ll_lco.lco_lock);
 425         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 426         spin_unlock(&sbi->ll_lco.lco_lock);
 427
 428         err = obd_register_page_removal_cb(sbi->ll_dt_exp,
 429                                            ll_page_removal_cb,
 430                                            ll_pin_extent_cb);
 431         if (err) {
 432                 CERROR("cannot register page removal callback: rc = %d\n",err);
 433                 GOTO(out_dt, err);
 434         }
 435         err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
 436                                           ll_extent_lock_cancel_cb);
 437         if (err) {
 438                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
 439                 GOTO(out_page_rm_cb, err);
 440         }
 441
 442         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 443         if (err) {
 444                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 445                 GOTO(out_lock_cn_cb, err);
 446         }
 447
 448         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 449                                   0, NULL, NULL, NULL, 0, NULL);
 450         if (err < 0) {
 451                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 452                                    "filesystem. There must be at least one "
 453                                    "active OST for a client to start.\n");
 454                 GOTO(out_lock_cn_cb, err);
 455         }
 456
 457         if (!ll_async_page_slab) {
 458                 ll_async_page_slab_size =
 459                         size_round(sizeof(struct ll_async_page)) + err;
 460                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 461                                                           ll_async_page_slab_size,
 462                                                           0, 0);
 463                 if (!ll_async_page_slab)
 464                         GOTO(out_lock_cn_cb, err = -ENOMEM);
 465         }
 466
 467         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 468         if (err) {
 469                 CERROR("cannot mds_connect: rc = %d\n", err);
 470                 GOTO(out_lock_cn_cb, err);
 471         }
 472         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 473         sbi->ll_root_fid = rootfid;
 474
 475         sb->s_op = &lustre_super_operations;
 476         sb->s_export_op = &lustre_export_operations;
 477
 478         /* make root inode
 479          * XXX: move this to after cbd setup? */
 480         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 481         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 482                 valid |= OBD_MD_FLRMTPERM;
 483         else if (sbi->ll_flags & LL_SBI_ACL)
 484                 valid |= OBD_MD_FLACL;
 485
 486         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 487         if (oc)
 488                 free_capa(oc);
 489         if (err) {
 490                 CERROR("md_getattr failed for root: rc = %d\n", err);
 491                 GOTO(out_lock_cn_cb, err);
 492         }
 493         memset(&lmd, 0, sizeof(lmd));
 494         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 495                                sbi->ll_md_exp, &lmd);
 496         if (err) {
 497                 CERROR("failed to understand root inode md: rc = %d\n", err);
 498                 ptlrpc_req_finished (request);
 499                 GOTO(out_lock_cn_cb, err);
 500         }
 501
 502         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 503         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 504         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 505         ptlrpc_req_finished(request);
 506
 507         if (root == NULL || is_bad_inode(root)) {
 508                 if (lmd.lsm)
 509                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 510 #ifdef CONFIG_FS_POSIX_ACL
 511                 if (lmd.posix_acl) {
 512                         posix_acl_release(lmd.posix_acl);
 513                         lmd.posix_acl = NULL;
 514                 }
 515 #endif
 516                 CERROR("lustre_lite: bad iget4 for root\n");
 517                 GOTO(out_root, err = -EBADF);
 518         }
 519
 520         err = ll_close_thread_start(&sbi->ll_lcq);
 521         if (err) {
 522                 CERROR("cannot start close thread: rc %d\n", err);
 523                 GOTO(out_root, err);
 524         }
 525
 526 #ifdef CONFIG_FS_POSIX_ACL
 527         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 528                 rct_init(&sbi->ll_rct);
 529                 et_init(&sbi->ll_et);
 530         }
 531 #endif
 532
 533         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 534         err = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"),"checksum",
 535                                  sizeof(checksum), &checksum, NULL);
 536
 537         sb->s_root = d_alloc_root(root);
 538         if (data != NULL)
 539                 OBD_FREE(data, sizeof(*data));
 540         sb->s_root->d_op = &ll_d_root_ops;
 541         RETURN(err);
 542 out_root:
 543         if (root)
 544                 iput(root);
 545 out_lock_cn_cb:
 546         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
 547                                       ll_extent_lock_cancel_cb);
 548 out_page_rm_cb:
 549         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 550                                        ll_page_removal_cb);
 551         obd_fid_fini(sbi->ll_dt_exp);
 552 out_dt:
 553         obd_disconnect(sbi->ll_dt_exp);
 554         sbi->ll_dt_exp = NULL;
 555 out_md_fid:
 556         obd_fid_fini(sbi->ll_md_exp);
 557 out_md:
 558         obd_disconnect(sbi->ll_md_exp);
 559         sbi->ll_md_exp = NULL;
 560 out:
 561         if (data != NULL)
 562                 OBD_FREE_PTR(data);
 563         lprocfs_unregister_mountpoint(sbi);
 564         return err;
 565 }
 566
 567 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 568 {
 569         int size, rc;
 570
 571         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 572         size = sizeof(int);
 573         rc = obd_get_info(sbi->ll_md_exp, strlen("max_easize"), "max_easize",
 574                           &size, lmmsize);
 575         if (rc)
 576                 CERROR("Get max mdsize error rc %d \n", rc);
 577
 578         RETURN(rc);
 579 }
 580
 581 void ll_dump_inode(struct inode *inode)
 582 {
 583         struct list_head *tmp;
 584         int dentry_count = 0;
 585
 586         LASSERT(inode != NULL);
 587
 588         list_for_each(tmp, &inode->i_dentry)
 589                 dentry_count++;
 590
 591         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 592                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 593                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 594 }
 595
 596 void lustre_dump_dentry(struct dentry *dentry, int recur)
 597 {
 598         struct list_head *tmp;
 599         int subdirs = 0;
 600
 601         LASSERT(dentry != NULL);
 602
 603         list_for_each(tmp, &dentry->d_subdirs)
 604                 subdirs++;
 605
 606         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 607                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 608                dentry->d_name.len, dentry->d_name.name,
 609                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 610                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 611                dentry->d_flags, dentry->d_fsdata, subdirs);
 612         if (dentry->d_inode != NULL)
 613                 ll_dump_inode(dentry->d_inode);
 614
 615         if (recur == 0)
 616                 return;
 617
 618         list_for_each(tmp, &dentry->d_subdirs) {
 619                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 620                 lustre_dump_dentry(d, recur - 1);
 621         }
 622 }
 623
 624 #ifdef HAVE_EXPORT___IGET
 625 static void prune_dir_dentries(struct inode *inode)
 626 {
 627         struct dentry *dentry, *prev = NULL;
 628
 629         /* due to lustre specific logic, a directory
 630          * can have few dentries - a bug from VFS POV */
 631 restart:
 632         spin_lock(&dcache_lock);
 633         if (!list_empty(&inode->i_dentry)) {
 634                 dentry = list_entry(inode->i_dentry.prev,
 635                                     struct dentry, d_alias);
 636                 /* in order to prevent infinite loops we
 637                  * break if previous dentry is busy */
 638                 if (dentry != prev) {
 639                         prev = dentry;
 640                         dget_locked(dentry);
 641                         spin_unlock(&dcache_lock);
 642
 643                         /* try to kill all child dentries */
 644                         lock_dentry(dentry);
 645                         shrink_dcache_parent(dentry);
 646                         unlock_dentry(dentry);
 647                         dput(dentry);
 648
 649                         /* now try to get rid of current dentry */
 650                         d_prune_aliases(inode);
 651                         goto restart;
 652                 }
 653         }
 654         spin_unlock(&dcache_lock);
 655 }
 656
 657 static void prune_deathrow_one(struct ll_inode_info *lli)
 658 {
 659         struct inode *inode = ll_info2i(lli);
 660
 661         /* first, try to drop any dentries - they hold a ref on the inode */
 662         if (S_ISDIR(inode->i_mode))
 663                 prune_dir_dentries(inode);
 664         else
 665                 d_prune_aliases(inode);
 666
 667
 668         /* if somebody still uses it, leave it */
 669         LASSERT(atomic_read(&inode->i_count) > 0);
 670         if (atomic_read(&inode->i_count) > 1)
 671                 goto out;
 672
 673         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 674                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 675
 676         /* seems nobody uses it anymore */
 677         inode->i_nlink = 0;
 678
 679 out:
 680         iput(inode);
 681         return;
 682 }
 683
 684 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 685 {
 686         struct ll_inode_info *lli;
 687         int empty;
 688
 689         do {
 690                 if (need_resched() && try)
 691                         break;
 692
 693                 if (try) {
 694                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 695                                 break;
 696                 } else {
 697                         spin_lock(&sbi->ll_deathrow_lock);
 698                 }
 699
 700                 empty = 1;
 701                 lli = NULL;
 702                 if (!list_empty(&sbi->ll_deathrow)) {
 703                         lli = list_entry(sbi->ll_deathrow.next,
 704                                          struct ll_inode_info,
 705                                          lli_dead_list);
 706                         list_del_init(&lli->lli_dead_list);
 707                         if (!list_empty(&sbi->ll_deathrow))
 708                                 empty = 0;
 709                 }
 710                 spin_unlock(&sbi->ll_deathrow_lock);
 711
 712                 if (lli)
 713                         prune_deathrow_one(lli);
 714
 715         } while (empty == 0);
 716 }
 717 #else /* !HAVE_EXPORT___IGET */
 718 #define prune_deathrow(sbi, try) do {} while (0)
 719 #endif /* HAVE_EXPORT___IGET */
 720
 721 void client_common_put_super(struct super_block *sb)
 722 {
 723         struct ll_sb_info *sbi = ll_s2sbi(sb);
 724         ENTRY;
 725
 726 #ifdef CONFIG_FS_POSIX_ACL
 727         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 728                 et_fini(&sbi->ll_et);
 729                 rct_fini(&sbi->ll_rct);
 730         }
 731 #endif
 732
 733         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 734
 735         ll_close_thread_shutdown(sbi->ll_lcq);
 736
 737         /* destroy inodes in deathrow */
 738         prune_deathrow(sbi, 0);
 739
 740         list_del(&sbi->ll_conn_chain);
 741
 742         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 743                                        ll_page_removal_cb);
 744         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,ll_extent_lock_cancel_cb);
 745
 746         obd_fid_fini(sbi->ll_dt_exp);
 747         obd_disconnect(sbi->ll_dt_exp);
 748         sbi->ll_dt_exp = NULL;
 749
 750         lprocfs_unregister_mountpoint(sbi);
 751
 752         obd_fid_fini(sbi->ll_md_exp);
 753         obd_disconnect(sbi->ll_md_exp);
 754         sbi->ll_md_exp = NULL;
 755
 756         EXIT;
 757 }
 758
 759 void ll_kill_super(struct super_block *sb)
 760 {
 761         struct ll_sb_info *sbi;
 762
 763         ENTRY;
 764
 765         /* not init sb ?*/
 766         if (!(sb->s_flags & MS_ACTIVE))
 767                 return;
 768
 769         sbi = ll_s2sbi(sb);
 770         /* we need restore s_dev from changed for clustred NFS before put_super
 771          * because new kernels have cached s_dev and change sb->s_dev in
 772          * put_super not affected real removing devices */
 773         if (sbi)
 774                 sb->s_dev = sbi->ll_sdev_orig;
 775         EXIT;
 776 }
 777
 778 char *ll_read_opt(const char *opt, char *data)
 779 {
 780         char *value;
 781         char *retval;
 782         ENTRY;
 783
 784         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 785         if (strncmp(opt, data, strlen(opt)))
 786                 RETURN(NULL);
 787         if ((value = strchr(data, '=')) == NULL)
 788                 RETURN(NULL);
 789
 790         value++;
 791         OBD_ALLOC(retval, strlen(value) + 1);
 792         if (!retval) {
 793                 CERROR("out of memory!\n");
 794                 RETURN(NULL);
 795         }
 796
 797         memcpy(retval, value, strlen(value)+1);
 798         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 799         RETURN(retval);
 800 }
 801
 802 static inline int ll_set_opt(const char *opt, char *data, int fl)
 803 {
 804         if (strncmp(opt, data, strlen(opt)) != 0)
 805                 return(0);
 806         else
 807                 return(fl);
 808 }
 809
 810 /* non-client-specific mount options are parsed in lmd_parse */
 811 static int ll_options(char *options, int *flags)
 812 {
 813         int tmp;
 814         char *s1 = options, *s2;
 815         ENTRY;
 816
 817         if (!options)
 818                 RETURN(0);
 819
 820         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 821
 822         while (*s1) {
 823                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 824                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 830                 if (tmp) {
 831                         *flags |= tmp;
 832                         goto next;
 833                 }
 834                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 835                 if (tmp) {
 836                         *flags |= tmp;
 837                         goto next;
 838                 }
 839                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 840                 if (tmp) {
 841                         *flags &= ~tmp;
 842                         goto next;
 843                 }
 844                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 845                 if (tmp) {
 846                         *flags |= tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 850                 if (tmp) {
 851                         *flags &= ~tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 855                 if (tmp) {
 856                         /* Ignore deprecated mount option.  The client will
 857                          * always try to mount with ACL support, whether this
 858                          * is used depends on whether server supports it. */
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 862                 if (tmp) {
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870
 871                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 872                 if (tmp) {
 873                         *flags |= tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 877                 if (tmp) {
 878                         *flags &= ~tmp;
 879                         goto next;
 880                 }
 881                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 882                 if (tmp) {
 883                         *flags |= tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 887                 if (tmp) {
 888                         *flags &= ~tmp;
 889                         goto next;
 890                 }
 891
 892                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 893                                    s1);
 894                 RETURN(-EINVAL);
 895
 896 next:
 897                 /* Find next opt */
 898                 s2 = strchr(s1, ',');
 899                 if (s2 == NULL)
 900                         break;
 901                 s1 = s2 + 1;
 902         }
 903         RETURN(0);
 904 }
 905
 906 void ll_lli_init(struct ll_inode_info *lli)
 907 {
 908         lli->lli_inode_magic = LLI_INODE_MAGIC;
 909         sema_init(&lli->lli_size_sem, 1);
 910         sema_init(&lli->lli_write_sem, 1);
 911         lli->lli_flags = 0;
 912         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
 913         spin_lock_init(&lli->lli_lock);
 914         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
 915         INIT_LIST_HEAD(&lli->lli_close_list);
 916         lli->lli_inode_magic = LLI_INODE_MAGIC;
 917         sema_init(&lli->lli_och_sem, 1);
 918         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
 919         lli->lli_mds_exec_och = NULL;
 920         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
 921         lli->lli_open_fd_exec_count = 0;
 922         INIT_LIST_HEAD(&lli->lli_dead_list);
 923         lli->lli_remote_perms = NULL;
 924         lli->lli_rmtperm_utime = 0;
 925         sema_init(&lli->lli_rmtperm_sem, 1);
 926         INIT_LIST_HEAD(&lli->lli_oss_capas);
 927 }
 928
 929 int ll_fill_super(struct super_block *sb)
 930 {
 931         struct lustre_profile *lprof;
 932         struct lustre_sb_info *lsi = s2lsi(sb);
 933         struct ll_sb_info *sbi;
 934         char  *dt = NULL, *md = NULL;
 935         char  *profilenm = get_profile_name(sb);
 936         struct config_llog_instance cfg = {0, };
 937         char   ll_instance[sizeof(sb) * 2 + 1];
 938         int    err;
 939         ENTRY;
 940
 941         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 942
 943         cfs_module_get();
 944
 945         /* client additional sb info */
 946         lsi->lsi_llsbi = sbi = ll_init_sbi();
 947         if (!sbi) {
 948                 cfs_module_put();
 949                 RETURN(-ENOMEM);
 950         }
 951
 952         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 953         if (err)
 954                 GOTO(out_free, err);
 955
 956         /* Generate a string unique to this super, in case some joker tries
 957            to mount the same fs at two mount points.
 958            Use the address of the super itself.*/
 959         sprintf(ll_instance, "%p", sb);
 960         cfg.cfg_instance = ll_instance;
 961         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 962
 963         /* set up client obds */
 964         err = lustre_process_log(sb, profilenm, &cfg);
 965         if (err < 0) {
 966                 CERROR("Unable to process log: %d\n", err);
 967                 GOTO(out_free, err);
 968         }
 969
 970         lprof = class_get_profile(profilenm);
 971         if (lprof == NULL) {
 972                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
 973                                    " read from the MGS.  Does that filesystem "
 974                                    "exist?\n", profilenm);
 975                 GOTO(out_free, err = -EINVAL);
 976         }
 977         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 978                lprof->lp_md, lprof->lp_dt);
 979
 980         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
 981                   strlen(ll_instance) + 2);
 982         if (!dt)
 983                 GOTO(out_free, err = -ENOMEM);
 984         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
 985
 986         OBD_ALLOC(md, strlen(lprof->lp_md) +
 987                   strlen(ll_instance) + 2);
 988         if (!md)
 989                 GOTO(out_free, err = -ENOMEM);
 990         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
 991
 992         /* connections, registrations, sb setup */
 993         err = client_common_fill_super(sb, md, dt);
 994
 995 out_free:
 996         if (md)
 997                 OBD_FREE(md, strlen(md) + 1);
 998         if (dt)
 999                 OBD_FREE(dt, strlen(dt) + 1);
1000         if (err)
1001                 ll_put_super(sb);
1002         else
1003                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1004
1005         RETURN(err);
1006 } /* ll_fill_super */
1007
1008
1009 void ll_put_super(struct super_block *sb)
1010 {
1011         struct config_llog_instance cfg;
1012         char   ll_instance[sizeof(sb) * 2 + 1];
1013         struct obd_device *obd;
1014         struct lustre_sb_info *lsi = s2lsi(sb);
1015         struct ll_sb_info *sbi = ll_s2sbi(sb);
1016         char *profilenm = get_profile_name(sb);
1017         int force = 1, next;
1018         ENTRY;
1019
1020         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1021
1022         ll_print_capa_stat(sbi);
1023
1024         sprintf(ll_instance, "%p", sb);
1025         cfg.cfg_instance = ll_instance;
1026         lustre_end_log(sb, NULL, &cfg);
1027
1028         if (sbi->ll_md_exp) {
1029                 obd = class_exp2obd(sbi->ll_md_exp);
1030                 if (obd)
1031                         force = obd->obd_force;
1032         }
1033
1034         /* We need to set force before the lov_disconnect in
1035            lustre_common_put_super, since l_d cleans up osc's as well. */
1036         if (force) {
1037                 next = 0;
1038                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1039                                                      &next)) != NULL) {
1040                         obd->obd_force = force;
1041                 }
1042         }
1043
1044         if (sbi->ll_lcq) {
1045                 /* Only if client_common_fill_super succeeded */
1046                 client_common_put_super(sb);
1047         }
1048         next = 0;
1049         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1050                 class_manual_cleanup(obd);
1051         }
1052
1053         if (profilenm)
1054                 class_del_profile(profilenm);
1055
1056         ll_free_sbi(sb);
1057         lsi->lsi_llsbi = NULL;
1058
1059         lustre_common_put_super(sb);
1060
1061         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1062
1063         cfs_module_put();
1064
1065         EXIT;
1066 } /* client_put_super */
1067
1068 #ifdef HAVE_REGISTER_CACHE
1069 #include <linux/cache_def.h>
1070 #ifdef HAVE_CACHE_RETURN_INT
1071 static int
1072 #else
1073 static void
1074 #endif
1075 ll_shrink_cache(int priority, unsigned int gfp_mask)
1076 {
1077         struct ll_sb_info *sbi;
1078         int count = 0;
1079
1080         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1081                 count += llap_shrink_cache(sbi, priority);
1082
1083 #ifdef HAVE_CACHE_RETURN_INT
1084         return count;
1085 #endif
1086 }
1087
1088 struct cache_definition ll_cache_definition = {
1089         .name = "llap_cache",
1090         .shrink = ll_shrink_cache
1091 };
1092 #endif /* HAVE_REGISTER_CACHE */
1093
1094 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1095 {
1096         struct inode *inode = NULL;
1097         /* NOTE: we depend on atomic igrab() -bzzz */
1098         lock_res_and_lock(lock);
1099         if (lock->l_ast_data) {
1100                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1101                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1102                         inode = igrab(lock->l_ast_data);
1103                 } else {
1104                         inode = lock->l_ast_data;
1105                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1106                                                 D_INFO : D_WARNING,
1107                                         lock, __FILE__, __func__, __LINE__,
1108                                         "l_ast_data %p is bogus: magic %08x",
1109                                         lock->l_ast_data, lli->lli_inode_magic);
1110                         inode = NULL;
1111                 }
1112         }
1113         unlock_res_and_lock(lock);
1114         return inode;
1115 }
1116
1117 static int null_if_equal(struct ldlm_lock *lock, void *data)
1118 {
1119         if (data == lock->l_ast_data) {
1120                 lock->l_ast_data = NULL;
1121
1122                 if (lock->l_req_mode != lock->l_granted_mode)
1123                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1124         }
1125
1126         return LDLM_ITER_CONTINUE;
1127 }
1128
1129 void ll_clear_inode(struct inode *inode)
1130 {
1131         struct ll_inode_info *lli = ll_i2info(inode);
1132         struct ll_sb_info *sbi = ll_i2sbi(inode);
1133         ENTRY;
1134
1135         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1136                inode->i_generation, inode);
1137
1138         if (S_ISDIR(inode->i_mode)) {
1139                 /* these should have been cleared in ll_file_release */
1140                 LASSERT(lli->lli_sai == NULL);
1141                 LASSERT(lli->lli_opendir_key == NULL);
1142                 LASSERT(lli->lli_opendir_pid == 0);
1143         }
1144
1145         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1146         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1147                          null_if_equal, inode);
1148
1149         LASSERT(!lli->lli_open_fd_write_count);
1150         LASSERT(!lli->lli_open_fd_read_count);
1151         LASSERT(!lli->lli_open_fd_exec_count);
1152
1153         if (lli->lli_mds_write_och)
1154                 ll_md_real_close(inode, FMODE_WRITE);
1155         if (lli->lli_mds_exec_och)
1156                 ll_md_real_close(inode, FMODE_EXEC);
1157         if (lli->lli_mds_read_och)
1158                 ll_md_real_close(inode, FMODE_READ);
1159
1160         if (lli->lli_smd) {
1161                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1162                                   null_if_equal, inode);
1163
1164                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1165                 lli->lli_smd = NULL;
1166         }
1167
1168         if (lli->lli_symlink_name) {
1169                 OBD_FREE(lli->lli_symlink_name,
1170                          strlen(lli->lli_symlink_name) + 1);
1171                 lli->lli_symlink_name = NULL;
1172         }
1173
1174         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1175                 LASSERT(lli->lli_posix_acl == NULL);
1176                 if (lli->lli_remote_perms) {
1177                         free_rmtperm_hash(lli->lli_remote_perms);
1178                         lli->lli_remote_perms = NULL;
1179                 }
1180         }
1181 #ifdef CONFIG_FS_POSIX_ACL
1182         else if (lli->lli_posix_acl) {
1183                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1184                 LASSERT(lli->lli_remote_perms == NULL);
1185                 posix_acl_release(lli->lli_posix_acl);
1186                 lli->lli_posix_acl = NULL;
1187         }
1188 #endif
1189         lli->lli_inode_magic = LLI_INODE_DEAD;
1190
1191 #ifdef HAVE_EXPORT___IGET
1192         spin_lock(&sbi->ll_deathrow_lock);
1193         list_del_init(&lli->lli_dead_list);
1194         spin_unlock(&sbi->ll_deathrow_lock);
1195 #endif
1196         ll_clear_inode_capas(inode);
1197
1198         EXIT;
1199 }
1200
1201 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1202                   struct md_open_data **mod)
1203 {
1204         struct lustre_md md;
1205         struct ll_sb_info *sbi = ll_i2sbi(inode);
1206         struct ptlrpc_request *request = NULL;
1207         int rc;
1208         ENTRY;
1209
1210         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1211                                      LUSTRE_OPC_ANY, NULL);
1212         if (IS_ERR(op_data))
1213                 RETURN(PTR_ERR(op_data));
1214
1215         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1216                         &request, mod);
1217         if (rc) {
1218                 ptlrpc_req_finished(request);
1219                 if (rc == -ENOENT) {
1220                         inode->i_nlink = 0;
1221                         /* Unlinked special device node? Or just a race?
1222                          * Pretend we done everything. */
1223                         if (!S_ISREG(inode->i_mode) &&
1224                             !S_ISDIR(inode->i_mode))
1225                                 rc = inode_setattr(inode, &op_data->op_attr);
1226                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1227                         CERROR("md_setattr fails: rc = %d\n", rc);
1228                 }
1229                 RETURN(rc);
1230         }
1231
1232         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1233                               sbi->ll_md_exp, &md);
1234         if (rc) {
1235                 ptlrpc_req_finished(request);
1236                 RETURN(rc);
1237         }
1238
1239         /* We call inode_setattr to adjust timestamps.
1240          * If there is at least some data in file, we cleared ATTR_SIZE
1241          * above to avoid invoking vmtruncate, otherwise it is important
1242          * to call vmtruncate in inode_setattr to update inode->i_size
1243          * (bug 6196) */
1244         rc = inode_setattr(inode, &op_data->op_attr);
1245
1246         /* Extract epoch data if obtained. */
1247         op_data->op_handle = md.body->handle;
1248         op_data->op_ioepoch = md.body->ioepoch;
1249
1250         ll_update_inode(inode, &md);
1251         ptlrpc_req_finished(request);
1252
1253         RETURN(rc);
1254 }
1255
1256 /* Close IO epoch and send Size-on-MDS attribute update. */
1257 static int ll_setattr_done_writing(struct inode *inode,
1258                                    struct md_op_data *op_data,
1259                                    struct md_open_data *mod)
1260 {
1261         struct ll_inode_info *lli = ll_i2info(inode);
1262         int rc = 0;
1263         ENTRY;
1264
1265         LASSERT(op_data != NULL);
1266         if (!S_ISREG(inode->i_mode))
1267                 RETURN(0);
1268
1269         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1270                op_data->op_ioepoch, PFID(&lli->lli_fid));
1271
1272         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1273         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1274         if (rc == -EAGAIN) {
1275                 /* MDS has instructed us to obtain Size-on-MDS attribute
1276                  * from OSTs and send setattr to back to MDS. */
1277                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1278                                          op_data->op_ioepoch);
1279         } else if (rc) {
1280                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1281                        inode->i_ino, rc);
1282         }
1283         RETURN(rc);
1284 }
1285
1286 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1287 {
1288         struct ll_sb_info *sbi = ll_i2sbi(inode);
1289         struct ll_inode_info *lli = ll_i2info(inode);
1290         struct lov_stripe_md *lsm = lli->lli_smd;
1291         int rc;
1292         ldlm_policy_data_t policy = { .l_extent = {new_size,
1293                                                    OBD_OBJECT_EOF } };
1294         struct lustre_handle lockh = { 0 };
1295         int local_lock = 0; /* 0 - no local lock;
1296                              * 1 - lock taken by lock_extent;
1297                              * 2 - by obd_match*/
1298         int ast_flags;
1299         int err;
1300         ENTRY;
1301
1302         UNLOCK_INODE_MUTEX(inode);
1303         UP_WRITE_I_ALLOC_SEM(inode);
1304
1305         if (sbi->ll_lockless_truncate_enable &&
1306             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1307                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1308                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1309                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1310                 if (rc > 0) {
1311                         local_lock = 2;
1312                         rc = 0;
1313                 } else if (rc == 0) {
1314                         rc = ll_file_punch(inode, new_size, 1);
1315                 }
1316         } else {
1317                 /* XXX when we fix the AST intents to pass the discard-range
1318                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1319                  * XXX here. */
1320                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1321                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1322                                     &lockh, ast_flags);
1323                 if (likely(rc == 0))
1324                         local_lock = 1;
1325         }
1326
1327 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
1328         DOWN_WRITE_I_ALLOC_SEM(inode);
1329         LOCK_INODE_MUTEX(inode);
1330 #else
1331         LOCK_INODE_MUTEX(inode);
1332         DOWN_WRITE_I_ALLOC_SEM(inode);
1333 #endif
1334         if (likely(rc == 0)) {
1335                 /* Only ll_inode_size_lock is taken at this level.
1336                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1337                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1338                  * ll_truncate dropped ll_inode_size_lock() */
1339                 ll_inode_size_lock(inode, 0);
1340                 if (!local_lock) {
1341                         spin_lock(&lli->lli_lock);
1342                         lli->lli_flags |= LLIF_SRVLOCK;
1343                         spin_unlock(&lli->lli_lock);
1344                 }
1345                 rc = vmtruncate(inode, new_size);
1346                 if (!local_lock) {
1347                         spin_lock(&lli->lli_lock);
1348                         lli->lli_flags &= ~LLIF_SRVLOCK;
1349                         spin_unlock(&lli->lli_lock);
1350                 }
1351                 if (rc != 0) {
1352                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1353                         ll_inode_size_unlock(inode, 0);
1354                 }
1355         }
1356
1357         if (local_lock) {
1358                 if (local_lock == 2)
1359                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1360                 else
1361                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1362                 if (unlikely(err != 0)){
1363                         CERROR("extent unlock failed: err=%d,"
1364                                " unlock method =%d\n", err, local_lock);
1365                         if (rc == 0)
1366                                 rc = err;
1367                 }
1368         }
1369         RETURN(rc);
1370 }
1371
1372 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1373  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1374  * keep these values until such a time that objects are allocated for it.
1375  * We do the MDS operations first, as it is checking permissions for us.
1376  * We don't to the MDS RPC if there is nothing that we want to store there,
1377  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1378  * going to do an RPC anyways.
1379  *
1380  * If we are doing a truncate, we will send the mtime and ctime updates
1381  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1382  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1383  * at the same time.
1384  */
1385 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1386 {
1387         struct ll_inode_info *lli = ll_i2info(inode);
1388         struct lov_stripe_md *lsm = lli->lli_smd;
1389         struct ll_sb_info *sbi = ll_i2sbi(inode);
1390         struct md_op_data *op_data = NULL;
1391         struct md_open_data *mod = NULL;
1392         int ia_valid = attr->ia_valid;
1393         int rc = 0, rc1 = 0;
1394         ENTRY;
1395
1396         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1397                attr->ia_valid);
1398         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1399
1400         if (ia_valid & ATTR_SIZE) {
1401                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1402                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1403                                attr->ia_size, ll_file_maxbytes(inode));
1404                         RETURN(-EFBIG);
1405                 }
1406
1407                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1408         }
1409
1410         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1411         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1412                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1413                         RETURN(-EPERM);
1414         }
1415
1416         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1417         if (attr->ia_valid & ATTR_CTIME) {
1418                 attr->ia_ctime = CURRENT_TIME;
1419                 attr->ia_valid |= ATTR_CTIME_SET;
1420         }
1421         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1422                 attr->ia_atime = CURRENT_TIME;
1423                 attr->ia_valid |= ATTR_ATIME_SET;
1424         }
1425         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1426                 attr->ia_mtime = CURRENT_TIME;
1427                 attr->ia_valid |= ATTR_MTIME_SET;
1428         }
1429         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1430                 /* To avoid stale mtime on mds, obtain it from ost and send
1431                    to mds. */
1432                 rc = ll_glimpse_size(inode, 0);
1433                 if (rc)
1434                         RETURN(rc);
1435
1436                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1437                 attr->ia_mtime = inode->i_mtime;
1438         }
1439
1440         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1441                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1442                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1443                        cfs_time_current_sec());
1444
1445         /* NB: ATTR_SIZE will only be set after this point if the size
1446          * resides on the MDS, ie, this file has no objects. */
1447         if (lsm)
1448                 attr->ia_valid &= ~ATTR_SIZE;
1449
1450         /* We always do an MDS RPC, even if we're only changing the size;
1451          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1452
1453         OBD_ALLOC_PTR(op_data);
1454         if (op_data == NULL)
1455                 RETURN(-ENOMEM);
1456
1457         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1458
1459         /* Open epoch for truncate. */
1460         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1461             (ia_valid & ATTR_SIZE))
1462                 op_data->op_flags = MF_EPOCH_OPEN;
1463
1464         rc = ll_md_setattr(inode, op_data, &mod);
1465         if (rc)
1466                 GOTO(out, rc);
1467
1468         if (op_data->op_ioepoch)
1469                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1470                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1471
1472         if (!lsm || !S_ISREG(inode->i_mode)) {
1473                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1474                 GOTO(out, rc = 0);
1475         }
1476
1477         /* We really need to get our PW lock before we change inode->i_size.
1478          * If we don't we can race with other i_size updaters on our node, like
1479          * ll_file_read.  We can also race with i_size propogation to other
1480          * nodes through dirtying and writeback of final cached pages.  This
1481          * last one is especially bad for racing o_append users on other
1482          * nodes. */
1483         if (ia_valid & ATTR_SIZE) {
1484                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1485         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1486                 obd_flag flags;
1487                 struct obd_info oinfo = { { { 0 } } };
1488                 struct obdo *oa;
1489
1490                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1491                        inode->i_ino, LTIME_S(attr->ia_mtime));
1492
1493                 OBDO_ALLOC(oa);
1494                 if (oa) {
1495                         oa->o_id = lsm->lsm_object_id;
1496                         oa->o_gr = lsm->lsm_object_gr;
1497                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1498
1499                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1500                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1501                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1502                                 OBD_MD_FLGROUP;
1503
1504                         obdo_from_inode(oa, inode, flags);
1505
1506                         oinfo.oi_oa = oa;
1507                         oinfo.oi_md = lsm;
1508                         oinfo.oi_capa = ll_mdscapa_get(inode);
1509
1510                         /* XXX: this looks unnecessary now. */
1511                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1512                         capa_put(oinfo.oi_capa);
1513                         if (rc)
1514                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1515                         OBDO_FREE(oa);
1516                 } else {
1517                         rc = -ENOMEM;
1518                 }
1519         }
1520         EXIT;
1521 out:
1522         if (op_data) {
1523                 if (op_data->op_ioepoch)
1524                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1525                 ll_finish_md_op_data(op_data);
1526         }
1527         return rc ? rc : rc1;
1528 }
1529
1530 int ll_setattr(struct dentry *de, struct iattr *attr)
1531 {
1532         int mode;
1533
1534         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1535             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1536                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1537         if ((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1538             (ATTR_SIZE|ATTR_MODE)) {
1539                 mode = de->d_inode->i_mode;
1540                 if (((mode & S_ISUID) && (!(attr->ia_mode & S_ISUID))) ||
1541                     ((mode & S_ISGID) && (mode & S_IXGRP) &&
1542                     (!(attr->ia_mode & S_ISGID))))
1543                         attr->ia_valid |= ATTR_FORCE;
1544         }
1545
1546         return ll_setattr_raw(de->d_inode, attr);
1547 }
1548
1549 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1550                        __u64 max_age, __u32 flags)
1551 {
1552         struct ll_sb_info *sbi = ll_s2sbi(sb);
1553         struct obd_statfs obd_osfs;
1554         int rc;
1555         ENTRY;
1556
1557         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1558         if (rc) {
1559                 CERROR("md_statfs fails: rc = %d\n", rc);
1560                 RETURN(rc);
1561         }
1562
1563         osfs->os_type = sb->s_magic;
1564
1565         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1566                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1567
1568         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1569                               &obd_osfs, max_age, flags);
1570         if (rc) {
1571                 CERROR("obd_statfs fails: rc = %d\n", rc);
1572                 RETURN(rc);
1573         }
1574
1575         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1576                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1577                obd_osfs.os_files);
1578
1579         osfs->os_bsize = obd_osfs.os_bsize;
1580         osfs->os_blocks = obd_osfs.os_blocks;
1581         osfs->os_bfree = obd_osfs.os_bfree;
1582         osfs->os_bavail = obd_osfs.os_bavail;
1583
1584         /* If we don't have as many objects free on the OST as inodes
1585          * on the MDS, we reduce the total number of inodes to
1586          * compensate, so that the "inodes in use" number is correct.
1587          */
1588         if (obd_osfs.os_ffree < osfs->os_ffree) {
1589                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1590                         obd_osfs.os_ffree;
1591                 osfs->os_ffree = obd_osfs.os_ffree;
1592         }
1593
1594         RETURN(rc);
1595 }
1596 #ifndef HAVE_STATFS_DENTRY_PARAM
1597 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1598 {
1599 #else
1600 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1601 {
1602         struct super_block *sb = de->d_sb;
1603 #endif
1604         struct obd_statfs osfs;
1605         int rc;
1606
1607         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1608         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1609
1610         /* For now we will always get up-to-date statfs values, but in the
1611          * future we may allow some amount of caching on the client (e.g.
1612          * from QOS or lprocfs updates). */
1613         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1614         if (rc)
1615                 return rc;
1616
1617         statfs_unpack(sfs, &osfs);
1618
1619         /* We need to downshift for all 32-bit kernels, because we can't
1620          * tell if the kernel is being called via sys_statfs64() or not.
1621          * Stop before overflowing f_bsize - in which case it is better
1622          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1623         if (sizeof(long) < 8) {
1624                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1625                         sfs->f_bsize <<= 1;
1626
1627                         osfs.os_blocks >>= 1;
1628                         osfs.os_bfree >>= 1;
1629                         osfs.os_bavail >>= 1;
1630                 }
1631         }
1632
1633         sfs->f_blocks = osfs.os_blocks;
1634         sfs->f_bfree = osfs.os_bfree;
1635         sfs->f_bavail = osfs.os_bavail;
1636
1637         return 0;
1638 }
1639
1640 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1641 {
1642         struct ll_inode_info *lli;
1643         struct lov_stripe_md *lsm;
1644
1645         lli = ll_i2info(inode);
1646         LASSERT(lli->lli_size_sem_owner != current);
1647         down(&lli->lli_size_sem);
1648         LASSERT(lli->lli_size_sem_owner == NULL);
1649         lli->lli_size_sem_owner = current;
1650         lsm = lli->lli_smd;
1651         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1652                  lsm, lock_lsm);
1653         if (lock_lsm)
1654                 lov_stripe_lock(lsm);
1655 }
1656
1657 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1658 {
1659         struct ll_inode_info *lli;
1660         struct lov_stripe_md *lsm;
1661
1662         lli = ll_i2info(inode);
1663         lsm = lli->lli_smd;
1664         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1665                  lsm, unlock_lsm);
1666         if (unlock_lsm)
1667                 lov_stripe_unlock(lsm);
1668         LASSERT(lli->lli_size_sem_owner == current);
1669         lli->lli_size_sem_owner = NULL;
1670         up(&lli->lli_size_sem);
1671 }
1672
1673 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1674 {
1675         struct ll_inode_info *lli = ll_i2info(inode);
1676
1677         dump_lsm(D_INODE, lsm);
1678         dump_lsm(D_INODE, lli->lli_smd);
1679         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1680                  "lsm must be joined lsm %p\n", lsm);
1681         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1682         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1683                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1684         lli->lli_smd = lsm;
1685         lli->lli_maxbytes = lsm->lsm_maxbytes;
1686         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1687                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1688 }
1689
1690 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1691 {
1692         struct ll_inode_info *lli = ll_i2info(inode);
1693         struct mdt_body *body = md->body;
1694         struct lov_stripe_md *lsm = md->lsm;
1695         struct ll_sb_info *sbi = ll_i2sbi(inode);
1696
1697         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1698         if (lsm != NULL) {
1699                 if (lli->lli_smd == NULL) {
1700                         if (lsm->lsm_magic != LOV_MAGIC &&
1701                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1702                                 dump_lsm(D_ERROR, lsm);
1703                                 LBUG();
1704                         }
1705                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1706                                lsm, inode->i_ino, inode->i_generation, inode);
1707                         /* ll_inode_size_lock() requires it is only called
1708                          * with lli_smd != NULL or lock_lsm == 0 or we can
1709                          * race between lock/unlock.  bug 9547 */
1710                         lli->lli_smd = lsm;
1711                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1712                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1713                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1714                 } else {
1715                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1716                              lli->lli_smd->lsm_stripe_count ==
1717                                         lsm->lsm_stripe_count) {
1718                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1719                                         CERROR("lsm mismatch for inode %ld\n",
1720                                                 inode->i_ino);
1721                                         CERROR("lli_smd:\n");
1722                                         dump_lsm(D_ERROR, lli->lli_smd);
1723                                         CERROR("lsm:\n");
1724                                         dump_lsm(D_ERROR, lsm);
1725                                         LBUG();
1726                                 }
1727                         } else
1728                                 ll_replace_lsm(inode, lsm);
1729                 }
1730                 if (lli->lli_smd != lsm)
1731                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1732         }
1733
1734         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1735                 if (body->valid & OBD_MD_FLRMTPERM)
1736                         ll_update_remote_perm(inode, md->remote_perm);
1737         }
1738 #ifdef CONFIG_FS_POSIX_ACL
1739         else if (body->valid & OBD_MD_FLACL) {
1740                 spin_lock(&lli->lli_lock);
1741                 if (lli->lli_posix_acl)
1742                         posix_acl_release(lli->lli_posix_acl);
1743                 lli->lli_posix_acl = md->posix_acl;
1744                 spin_unlock(&lli->lli_lock);
1745         }
1746 #endif
1747         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1748
1749         if (body->valid & OBD_MD_FLATIME &&
1750             body->atime > LTIME_S(inode->i_atime))
1751                 LTIME_S(inode->i_atime) = body->atime;
1752
1753         /* mtime is always updated with ctime, but can be set in past.
1754            As write and utime(2) may happen within 1 second, and utime's
1755            mtime has a priority over write's one, so take mtime from mds
1756            for the same ctimes. */
1757         if (body->valid & OBD_MD_FLCTIME &&
1758             body->ctime >= LTIME_S(inode->i_ctime)) {
1759                 LTIME_S(inode->i_ctime) = body->ctime;
1760                 if (body->valid & OBD_MD_FLMTIME) {
1761                         CDEBUG(D_INODE, "setting ino %lu mtime "
1762                                "from %lu to "LPU64"\n", inode->i_ino,
1763                                LTIME_S(inode->i_mtime), body->mtime);
1764                         LTIME_S(inode->i_mtime) = body->mtime;
1765                 }
1766         }
1767         if (body->valid & OBD_MD_FLMODE)
1768                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1769         if (body->valid & OBD_MD_FLTYPE)
1770                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1771         if (S_ISREG(inode->i_mode)) {
1772                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1773         } else {
1774                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1775         }
1776 #ifdef HAVE_INODE_BLKSIZE
1777         inode->i_blksize = 1<<inode->i_blkbits;
1778 #endif
1779         if (body->valid & OBD_MD_FLUID)
1780                 inode->i_uid = body->uid;
1781         if (body->valid & OBD_MD_FLGID)
1782                 inode->i_gid = body->gid;
1783         if (body->valid & OBD_MD_FLFLAGS)
1784                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1785         if (body->valid & OBD_MD_FLNLINK)
1786                 inode->i_nlink = body->nlink;
1787         if (body->valid & OBD_MD_FLRDEV)
1788                 inode->i_rdev = old_decode_dev(body->rdev);
1789
1790         if (body->valid & OBD_MD_FLID) {
1791                 /* FID shouldn't be changed! */
1792                 if (fid_is_sane(&lli->lli_fid)) {
1793                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1794                                  "Trying to change FID "DFID
1795                                  " to the "DFID", inode %lu/%u(%p)\n",
1796                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1797                                  inode->i_ino, inode->i_generation, inode);
1798                 } else
1799                         lli->lli_fid = body->fid1;
1800         }
1801
1802         LASSERT(fid_seq(&lli->lli_fid) != 0);
1803
1804         if (body->valid & OBD_MD_FLSIZE) {
1805                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1806                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1807                         struct lustre_handle lockh;
1808                         ldlm_mode_t mode;
1809
1810                         /* As it is possible a blocking ast has been processed
1811                          * by this time, we need to check there is an UPDATE
1812                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1813                          * it. */
1814                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1815                                                &lockh);
1816                         if (mode) {
1817                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1818                                                       LLIF_EPOCH_PENDING |
1819                                                       LLIF_SOM_DIRTY)) {
1820                                         CERROR("ino %lu flags %lu still has "
1821                                                "size authority! do not trust "
1822                                                "the size got from MDS\n",
1823                                                inode->i_ino, lli->lli_flags);
1824                                 } else {
1825                                         /* Use old size assignment to avoid
1826                                          * deadlock bz14138 & bz14326 */
1827                                         inode->i_size = body->size;
1828                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1829                                 }
1830                                 ldlm_lock_decref(&lockh, mode);
1831                         }
1832                 } else {
1833                         /* Use old size assignment to avoid
1834                          * deadlock bz14138 & bz14326 */
1835                         inode->i_size = body->size;
1836                 }
1837
1838                 if (body->valid & OBD_MD_FLBLOCKS)
1839                         inode->i_blocks = body->blocks;
1840         }
1841
1842         if (body->valid & OBD_MD_FLMDSCAPA) {
1843                 LASSERT(md->mds_capa);
1844                 ll_add_capa(inode, md->mds_capa);
1845         }
1846         if (body->valid & OBD_MD_FLOSSCAPA) {
1847                 LASSERT(md->oss_capa);
1848                 ll_add_capa(inode, md->oss_capa);
1849         }
1850 }
1851
1852 static struct backing_dev_info ll_backing_dev_info = {
1853         .ra_pages       = 0,    /* No readahead */
1854 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1855         .capabilities   = 0,    /* Does contribute to dirty memory */
1856 #else
1857         .memory_backed  = 0,    /* Does contribute to dirty memory */
1858 #endif
1859 };
1860
1861 void ll_read_inode2(struct inode *inode, void *opaque)
1862 {
1863         struct lustre_md *md = opaque;
1864         struct ll_inode_info *lli = ll_i2info(inode);
1865         ENTRY;
1866
1867         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1868                inode->i_ino, inode->i_generation, inode);
1869
1870         ll_lli_init(lli);
1871
1872         LASSERT(!lli->lli_smd);
1873
1874         /* Core attributes from the MDS first.  This is a new inode, and
1875          * the VFS doesn't zero times in the core inode so we have to do
1876          * it ourselves.  They will be overwritten by either MDS or OST
1877          * attributes - we just need to make sure they aren't newer. */
1878         LTIME_S(inode->i_mtime) = 0;
1879         LTIME_S(inode->i_atime) = 0;
1880         LTIME_S(inode->i_ctime) = 0;
1881         inode->i_rdev = 0;
1882         ll_update_inode(inode, md);
1883
1884         /* OIDEBUG(inode); */
1885
1886         if (S_ISREG(inode->i_mode)) {
1887                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1888                 inode->i_op = &ll_file_inode_operations;
1889                 inode->i_fop = sbi->ll_fop;
1890                 inode->i_mapping->a_ops = &ll_aops;
1891                 EXIT;
1892         } else if (S_ISDIR(inode->i_mode)) {
1893                 inode->i_op = &ll_dir_inode_operations;
1894                 inode->i_fop = &ll_dir_operations;
1895                 inode->i_mapping->a_ops = &ll_dir_aops;
1896                 EXIT;
1897         } else if (S_ISLNK(inode->i_mode)) {
1898                 inode->i_op = &ll_fast_symlink_inode_operations;
1899                 EXIT;
1900         } else {
1901                 inode->i_op = &ll_special_inode_operations;
1902
1903                 init_special_inode(inode, inode->i_mode,
1904                                    kdev_t_to_nr(inode->i_rdev));
1905
1906                 /* initializing backing dev info. */
1907                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1908
1909                 EXIT;
1910         }
1911 }
1912
1913 void ll_delete_inode(struct inode *inode)
1914 {
1915         struct ll_sb_info *sbi = ll_i2sbi(inode);
1916         int rc;
1917         ENTRY;
1918
1919         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
1920         if (rc) {
1921                 CERROR("fid_delete() failed, rc %d\n", rc);
1922         }
1923         truncate_inode_pages(&inode->i_data, 0);
1924         clear_inode(inode);
1925
1926         EXIT;
1927 }
1928
1929 int ll_iocontrol(struct inode *inode, struct file *file,
1930                  unsigned int cmd, unsigned long arg)
1931 {
1932         struct ll_sb_info *sbi = ll_i2sbi(inode);
1933         struct ptlrpc_request *req = NULL;
1934         int rc, flags = 0;
1935         ENTRY;
1936
1937         switch(cmd) {
1938         case EXT3_IOC_GETFLAGS: {
1939                 struct mdt_body *body;
1940                 struct obd_capa *oc;
1941
1942                 oc = ll_mdscapa_get(inode);
1943                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
1944                                 OBD_MD_FLFLAGS, 0, &req);
1945                 capa_put(oc);
1946                 if (rc) {
1947                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1948                         RETURN(-abs(rc));
1949                 }
1950
1951                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1952
1953                 flags = body->flags;
1954
1955                 ptlrpc_req_finished(req);
1956
1957                 RETURN(put_user(flags, (int *)arg));
1958         }
1959         case EXT3_IOC_SETFLAGS: {
1960                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
1961                 struct obd_info oinfo = { { { 0 } } };
1962                 struct md_op_data *op_data;
1963
1964                 if (get_user(flags, (int *)arg))
1965                         RETURN(-EFAULT);
1966
1967                 oinfo.oi_md = lsm;
1968                 OBDO_ALLOC(oinfo.oi_oa);
1969                 if (!oinfo.oi_oa)
1970                         RETURN(-ENOMEM);
1971
1972                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1973                                              LUSTRE_OPC_ANY, NULL);
1974                 if (IS_ERR(op_data))
1975                         RETURN(PTR_ERR(op_data));
1976
1977                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1978                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1979                 rc = md_setattr(sbi->ll_md_exp, op_data,
1980                                 NULL, 0, NULL, 0, &req, NULL);
1981                 ll_finish_md_op_data(op_data);
1982                 ptlrpc_req_finished(req);
1983                 if (rc || lsm == NULL) {
1984                         OBDO_FREE(oinfo.oi_oa);
1985                         RETURN(rc);
1986                 }
1987
1988                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1989                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
1990                 oinfo.oi_oa->o_flags = flags;
1991                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1992                                        OBD_MD_FLGROUP;
1993                 oinfo.oi_capa = ll_mdscapa_get(inode);
1994
1995                 obdo_from_inode(oinfo.oi_oa, inode,
1996                                 OBD_MD_FLFID | OBD_MD_FLGENER);
1997                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1998                 capa_put(oinfo.oi_capa);
1999                 OBDO_FREE(oinfo.oi_oa);
2000                 if (rc) {
2001                         if (rc != -EPERM && rc != -EACCES)
2002                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2003                         RETURN(rc);
2004                 }
2005
2006                 inode->i_flags = ll_ext_to_inode_flags(flags |
2007                                                        MDS_BFLAG_EXT_FLAGS);
2008                 RETURN(0);
2009         }
2010         default:
2011                 RETURN(-ENOSYS);
2012         }
2013
2014         RETURN(0);
2015 }
2016
2017 int ll_flush_ctx(struct inode *inode)
2018 {
2019         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2020
2021         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2022
2023         obd_set_info_async(sbi->ll_md_exp,
2024                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
2025                            0, NULL, NULL);
2026         obd_set_info_async(sbi->ll_dt_exp,
2027                            sizeof(KEY_FLUSH_CTX) - 1, KEY_FLUSH_CTX,
2028                            0, NULL, NULL);
2029         return 0;
2030 }
2031
2032 /* umount -f client means force down, don't save state */
2033 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2034 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2035 {
2036         struct super_block *sb = vfsmnt->mnt_sb;
2037 #else
2038 void ll_umount_begin(struct super_block *sb)
2039 {
2040 #endif
2041         struct lustre_sb_info *lsi = s2lsi(sb);
2042         struct ll_sb_info *sbi = ll_s2sbi(sb);
2043         struct obd_device *obd;
2044         struct obd_ioctl_data ioc_data = { 0 };
2045         ENTRY;
2046
2047 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2048         if (!(flags & MNT_FORCE)) {
2049                 EXIT;
2050                 return;
2051         }
2052 #endif
2053
2054         /* Tell the MGC we got umount -f */
2055         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2056
2057         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2058                sb->s_count, atomic_read(&sb->s_active));
2059
2060         obd = class_exp2obd(sbi->ll_md_exp);
2061         if (obd == NULL) {
2062                 CERROR("Invalid MDC connection handle "LPX64"\n",
2063                        sbi->ll_md_exp->exp_handle.h_cookie);
2064                 EXIT;
2065                 return;
2066         }
2067         obd->obd_force = 1;
2068         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2069                       &ioc_data, NULL);
2070
2071         obd = class_exp2obd(sbi->ll_dt_exp);
2072         if (obd == NULL) {
2073                 CERROR("Invalid LOV connection handle "LPX64"\n",
2074                        sbi->ll_dt_exp->exp_handle.h_cookie);
2075                 EXIT;
2076                 return;
2077         }
2078
2079         obd->obd_force = 1;
2080         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2081                       &ioc_data, NULL);
2082
2083         /* Really, we'd like to wait until there are no requests outstanding,
2084          * and then continue.  For now, we just invalidate the requests,
2085          * schedule, and hope.
2086          */
2087         schedule();
2088
2089         EXIT;
2090 }
2091
2092 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2093 {
2094         struct ll_sb_info *sbi = ll_s2sbi(sb);
2095         int err;
2096         __u32 read_only;
2097
2098         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2099                 read_only = *flags & MS_RDONLY;
2100                 err = obd_set_info_async(sbi->ll_md_exp,
2101                                          sizeof(KEY_READ_ONLY) - 1,
2102                                          KEY_READ_ONLY, sizeof(read_only),
2103                                          &read_only, NULL);
2104                 if (err) {
2105                         CERROR("Failed to change the read-only flag during "
2106                                "remount: %d\n", err);
2107                         return err;
2108                 }
2109
2110                 if (read_only)
2111                         sb->s_flags |= MS_RDONLY;
2112                 else
2113                         sb->s_flags &= ~MS_RDONLY;
2114         }
2115         return 0;
2116 }
2117
2118 int ll_prep_inode(struct inode **inode,
2119                   struct ptlrpc_request *req,
2120                   struct super_block *sb)
2121 {
2122         struct ll_sb_info *sbi = NULL;
2123         struct lustre_md md;
2124         int rc = 0;
2125         ENTRY;
2126
2127         LASSERT(*inode || sb);
2128         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2129         prune_deathrow(sbi, 1);
2130         memset(&md, 0, sizeof(struct lustre_md));
2131
2132         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2133                               sbi->ll_md_exp, &md);
2134         if (rc)
2135                 RETURN(rc);
2136
2137         if (*inode) {
2138                 ll_update_inode(*inode, &md);
2139         } else {
2140                 LASSERT(sb != NULL);
2141
2142                 /*
2143                  * At this point server returns to client's same fid as client
2144                  * generated for creating. So using ->fid1 is okay here.
2145                  */
2146                 LASSERT(fid_is_sane(&md.body->fid1));
2147
2148                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2149                 if (*inode == NULL || is_bad_inode(*inode)) {
2150                         if (md.lsm)
2151                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2152 #ifdef CONFIG_FS_POSIX_ACL
2153                         if (md.posix_acl) {
2154                                 posix_acl_release(md.posix_acl);
2155                                 md.posix_acl = NULL;
2156                         }
2157 #endif
2158                         rc = -ENOMEM;
2159                         CERROR("new_inode -fatal: rc %d\n", rc);
2160                         GOTO(out, rc);
2161                 }
2162         }
2163
2164         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2165                          ll_i2info(*inode)->lli_smd);
2166 out:
2167         md_free_lustre_md(sbi->ll_md_exp, &md);
2168         RETURN(rc);
2169 }
2170
2171 char *llap_origins[] = {
2172         [LLAP_ORIGIN_UNKNOWN] = "--",
2173         [LLAP_ORIGIN_READPAGE] = "rp",
2174         [LLAP_ORIGIN_READAHEAD] = "ra",
2175         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2176         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2177         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2178 };
2179
2180 struct ll_async_page *llite_pglist_next_llap(struct ll_sb_info *sbi,
2181                                              struct list_head *list)
2182 {
2183         struct ll_async_page *llap;
2184         struct list_head *pos;
2185
2186         list_for_each(pos, list) {
2187                 if (pos == &sbi->ll_pglist)
2188                         return NULL;
2189                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2190                 if (llap->llap_page == NULL)
2191                         continue;
2192                 return llap;
2193         }
2194         LBUG();
2195         return NULL;
2196 }
2197
2198 int ll_obd_statfs(struct inode *inode, void *arg)
2199 {
2200         struct ll_sb_info *sbi = NULL;
2201         struct obd_export *exp;
2202         char *buf = NULL;
2203         struct obd_ioctl_data *data = NULL;
2204         __u32 type;
2205         int len = 0, rc;
2206
2207         if (!inode || !(sbi = ll_i2sbi(inode)))
2208                 GOTO(out_statfs, rc = -EINVAL);
2209
2210         rc = obd_ioctl_getdata(&buf, &len, arg);
2211         if (rc)
2212                 GOTO(out_statfs, rc);
2213
2214         data = (void*)buf;
2215         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2216             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2217                 GOTO(out_statfs, rc = -EINVAL);
2218
2219         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2220         if (type == LL_STATFS_MDC)
2221                 exp = sbi->ll_md_exp;
2222         else if (type == LL_STATFS_LOV)
2223                 exp = sbi->ll_dt_exp;
2224         else
2225                 GOTO(out_statfs, rc = -ENODEV);
2226
2227         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2228         if (rc)
2229                 GOTO(out_statfs, rc);
2230 out_statfs:
2231         if (buf)
2232                 obd_ioctl_freedata(buf, len);
2233         return rc;
2234 }
2235
2236 int ll_process_config(struct lustre_cfg *lcfg)
2237 {
2238         char *ptr;
2239         void *sb;
2240         struct lprocfs_static_vars lvars;
2241         unsigned long x;
2242         int rc = 0;
2243
2244         lprocfs_llite_init_vars(&lvars);
2245
2246         /* The instance name contains the sb: lustre-client-aacfe000 */
2247         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2248         if (!ptr || !*(++ptr))
2249                 return -EINVAL;
2250         if (sscanf(ptr, "%lx", &x) != 1)
2251                 return -EINVAL;
2252         sb = (void *)x;
2253         /* This better be a real Lustre superblock! */
2254         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2255
2256         /* Note we have not called client_common_fill_super yet, so
2257            proc fns must be able to handle that! */
2258         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2259                                       lcfg, sb);
2260         return(rc);
2261 }
2262
2263 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2264 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2265                                        struct inode *i1, struct inode *i2,
2266                                        const char *name, int namelen,
2267                                        int mode, __u32 opc, void *data)
2268 {
2269         LASSERT(i1 != NULL);
2270
2271         if (namelen > ll_i2sbi(i1)->ll_namelen)
2272                 return ERR_PTR(-ENAMETOOLONG);
2273
2274         if (op_data == NULL)
2275                 OBD_ALLOC_PTR(op_data);
2276
2277         if (op_data == NULL)
2278                 return ERR_PTR(-ENOMEM);
2279
2280         ll_i2gids(op_data->op_suppgids, i1, i2);
2281         op_data->op_fid1 = *ll_inode2fid(i1);
2282         op_data->op_capa1 = ll_mdscapa_get(i1);
2283
2284         if (i2) {
2285                 op_data->op_fid2 = *ll_inode2fid(i2);
2286                 op_data->op_capa2 = ll_mdscapa_get(i2);
2287         } else {
2288                 fid_zero(&op_data->op_fid2);
2289                 op_data->op_capa2 = NULL;
2290         }
2291
2292         op_data->op_name = name;
2293         op_data->op_namelen = namelen;
2294         op_data->op_mode = mode;
2295         op_data->op_mod_time = cfs_time_current_sec();
2296         op_data->op_fsuid = current->fsuid;
2297         op_data->op_fsgid = current->fsgid;
2298         op_data->op_cap = current->cap_effective;
2299         op_data->op_bias = MDS_CHECK_SPLIT;
2300         op_data->op_opc = opc;
2301         op_data->op_mds = 0;
2302         op_data->op_data = data;
2303
2304         return op_data;
2305 }
2306
2307 void ll_finish_md_op_data(struct md_op_data *op_data)
2308 {
2309         capa_put(op_data->op_capa1);
2310         capa_put(op_data->op_capa2);
2311         OBD_FREE_PTR(op_data);
2312 }