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