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