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  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see [sun.com URL with a
  20  * copy of GPLv2].
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <obd_cksum.h>
  57 #include <lustre_cache.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) ffz(~(n))
  70 #endif
  71
  72 static inline void ll_pglist_fini(struct ll_sb_info *sbi)
  73 {
  74         struct page *page;
  75         int i;
  76
  77         if (sbi->ll_pglist == NULL)
  78                 return;
  79
  80         for_each_possible_cpu(i) {
  81                 page = sbi->ll_pglist[i]->llpd_page;
  82                 if (page) {
  83                         sbi->ll_pglist[i] = NULL;
  84                         __free_page(page);
  85                 }
  86         }
  87
  88         OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
  89         sbi->ll_pglist = NULL;
  90 }
  91
  92 static inline int ll_pglist_init(struct ll_sb_info *sbi)
  93 {
  94         struct ll_pglist_data *pd;
  95         unsigned long budget;
  96         int i, color = 0;
  97         ENTRY;
  98
  99         OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
 100         if (sbi->ll_pglist == NULL)
 101                 RETURN(-ENOMEM);
 102
 103         budget = sbi->ll_async_page_max / num_online_cpus();
 104         for_each_possible_cpu(i) {
 105                 struct page *page = alloc_pages_node(cpu_to_node(i),
 106                                                     GFP_KERNEL, 0);
 107                 if (page == NULL) {
 108                         ll_pglist_fini(sbi);
 109                         RETURN(-ENOMEM);
 110                 }
 111
 112                 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
 113                         color = 0;
 114
 115                 pd = (struct ll_pglist_data *)(page_address(page) + color);
 116                 memset(pd, 0, sizeof(*pd));
 117                 spin_lock_init(&pd->llpd_lock);
 118                 INIT_LIST_HEAD(&pd->llpd_list);
 119                 if (cpu_online(i))
 120                         pd->llpd_budget = budget;
 121                 pd->llpd_cpu = i;
 122                 pd->llpd_page = page;
 123                 atomic_set(&pd->llpd_sample_count, 0);
 124                 sbi->ll_pglist[i] = pd;
 125                 color += L1_CACHE_ALIGN(sizeof(*pd));
 126         }
 127
 128         RETURN(0);
 129 }
 130
 131 static struct ll_sb_info *ll_init_sbi(void)
 132 {
 133         struct ll_sb_info *sbi = NULL;
 134         unsigned long pages;
 135         struct sysinfo si;
 136         class_uuid_t uuid;
 137         int i;
 138         ENTRY;
 139
 140         OBD_ALLOC(sbi, sizeof(*sbi));
 141         if (!sbi)
 142                 RETURN(NULL);
 143
 144         OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
 145         if (sbi->ll_async_page_sample == NULL)
 146                 GOTO(out, 0);
 147
 148         spin_lock_init(&sbi->ll_lock);
 149         spin_lock_init(&sbi->ll_lco.lco_lock);
 150         spin_lock_init(&sbi->ll_pp_extent_lock);
 151         spin_lock_init(&sbi->ll_process_lock);
 152         sbi->ll_rw_stats_on = 0;
 153
 154         si_meminfo(&si);
 155         pages = si.totalram - si.totalhigh;
 156         if (pages >> (20 - CFS_PAGE_SHIFT) < 512)
 157                 sbi->ll_async_page_max = pages / 2;
 158         else
 159                 sbi->ll_async_page_max = (pages / 4) * 3;
 160
 161         lcounter_init(&sbi->ll_async_page_count);
 162         spin_lock_init(&sbi->ll_async_page_reblnc_lock);
 163         sbi->ll_async_page_sample_max = 64 * num_online_cpus();
 164         sbi->ll_async_page_reblnc_count = 0;
 165         sbi->ll_async_page_clock_hand = 0;
 166         if (ll_pglist_init(sbi))
 167                 GOTO(out, 0);
 168
 169         sbi->ll_ra_info.ra_max_pages = min(pages / 32,
 170                                            SBI_DEFAULT_READAHEAD_MAX);
 171         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 172                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 173         sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
 174         sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
 175         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 176         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 177
 178         ll_generate_random_uuid(uuid);
 179         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 180         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 181
 182         spin_lock(&ll_sb_lock);
 183         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 184         spin_unlock(&ll_sb_lock);
 185
 186 #ifdef ENABLE_LLITE_CHECKSUM
 187         sbi->ll_flags |= LL_SBI_CHECKSUM;
 188 #endif
 189
 190 #ifdef HAVE_LRU_RESIZE_SUPPORT
 191         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 192 #endif
 193
 194 #ifdef HAVE_EXPORT___IGET
 195         INIT_LIST_HEAD(&sbi->ll_deathrow);
 196         spin_lock_init(&sbi->ll_deathrow_lock);
 197 #endif
 198         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 199                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 200                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 201         }
 202
 203         /* metadata statahead is enabled by default */
 204         sbi->ll_sa_max = LL_SA_RPC_DEF;
 205
 206         RETURN(sbi);
 207
 208 out:
 209         if (sbi->ll_async_page_sample)
 210                 OBD_FREE(sbi->ll_async_page_sample,
 211                          sizeof(long) * num_possible_cpus());
 212         ll_pglist_fini(sbi);
 213         OBD_FREE(sbi, sizeof(*sbi));
 214         RETURN(NULL);
 215 }
 216
 217 void ll_free_sbi(struct super_block *sb)
 218 {
 219         struct ll_sb_info *sbi = ll_s2sbi(sb);
 220         ENTRY;
 221
 222         if (sbi != NULL) {
 223                 ll_pglist_fini(sbi);
 224                 spin_lock(&ll_sb_lock);
 225                 list_del(&sbi->ll_list);
 226                 spin_unlock(&ll_sb_lock);
 227                 lcounter_destroy(&sbi->ll_async_page_count);
 228                 OBD_FREE(sbi->ll_async_page_sample,
 229                          sizeof(long) * num_possible_cpus());
 230                 OBD_FREE(sbi, sizeof(*sbi));
 231         }
 232         EXIT;
 233 }
 234
 235 static struct dentry_operations ll_d_root_ops = {
 236 #ifdef DCACHE_LUSTRE_INVALID
 237         .d_compare = ll_dcompare,
 238 #endif
 239 };
 240
 241 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 242  * us to make MDS RPCs with large enough reply buffers to hold the
 243  * maximum-sized (= maximum striped) EA and cookie without having to
 244  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 245 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 246 {
 247         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 248         __u32 valsize = sizeof(struct lov_desc);
 249         int rc, easize, def_easize, cookiesize;
 250         struct lov_desc desc;
 251         __u32 stripes;
 252         ENTRY;
 253
 254         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
 255                           &valsize, &desc, NULL);
 256         if (rc)
 257                 RETURN(rc);
 258
 259         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 260         lsm.lsm_stripe_count = stripes;
 261         easize = obd_size_diskmd(dt_exp, &lsm);
 262
 263         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 264         def_easize = obd_size_diskmd(dt_exp, &lsm);
 265
 266         cookiesize = stripes * sizeof(struct llog_cookie);
 267
 268         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 269                easize, cookiesize);
 270
 271         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 272         RETURN(rc);
 273 }
 274
 275 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 276 {
 277         struct inode *root = 0;
 278         struct ll_sb_info *sbi = ll_s2sbi(sb);
 279         struct obd_device *obd;
 280         struct lu_fid rootfid;
 281         struct obd_capa *oc = NULL;
 282         struct obd_statfs osfs;
 283         struct ptlrpc_request *request = NULL;
 284         struct lustre_handle dt_conn = {0, };
 285         struct lustre_handle md_conn = {0, };
 286         struct obd_connect_data *data = NULL;
 287         struct lustre_md lmd;
 288         obd_valid valid;
 289         int size, err, checksum;
 290         ENTRY;
 291
 292         obd = class_name2obd(md);
 293         if (!obd) {
 294                 CERROR("MD %s: not setup or attached\n", md);
 295                 RETURN(-EINVAL);
 296         }
 297
 298         OBD_ALLOC_PTR(data);
 299         if (data == NULL)
 300                 RETURN(-ENOMEM);
 301
 302         if (proc_lustre_fs_root) {
 303                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 304                                                   dt, md);
 305                 if (err < 0)
 306                         CERROR("could not register mount in /proc/fs/lustre\n");
 307         }
 308
 309         /* indicate the features supported by this client */
 310         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 311                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 312                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 313                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 314                                   OBD_CONNECT_FID      | OBD_CONNECT_AT;
 315
 316 #ifdef HAVE_LRU_RESIZE_SUPPORT
 317         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 318                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 319 #endif
 320 #ifdef CONFIG_FS_POSIX_ACL
 321         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 322 #endif
 323         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 324         data->ocd_version = LUSTRE_VERSION_CODE;
 325
 326         if (sb->s_flags & MS_RDONLY)
 327                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 328         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 329                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 330
 331 #ifdef HAVE_MS_FLOCK_LOCK
 332         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 333         sb->s_flags |= MS_FLOCK_LOCK;
 334 #endif
 335
 336         if (sbi->ll_flags & LL_SBI_FLOCK)
 337                 sbi->ll_fop = &ll_file_operations_flock;
 338         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 339                 sbi->ll_fop = &ll_file_operations;
 340         else
 341                 sbi->ll_fop = &ll_file_operations_noflock;
 342
 343         /* real client */
 344         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 345         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 346                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 347                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 348         } else {
 349                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 350                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 351         }
 352
 353         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 354         if (err == -EBUSY) {
 355                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 356                                    "recovery, of which this client is not a "
 357                                    "part. Please wait for recovery to complete,"
 358                                    " abort, or time out.\n", md);
 359                 GOTO(out, err);
 360         } else if (err) {
 361                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 362                 GOTO(out, err);
 363         }
 364         sbi->ll_md_exp = class_conn2export(&md_conn);
 365
 366         err = obd_fid_init(sbi->ll_md_exp);
 367         if (err) {
 368                 CERROR("Can't init metadata layer FID infrastructure, "
 369                        "rc %d\n", err);
 370                 GOTO(out_md, err);
 371         }
 372
 373         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 374         if (err)
 375                 GOTO(out_md_fid, err);
 376
 377         size = sizeof(*data);
 378         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 379                            KEY_CONN_DATA,  &size, data, NULL);
 380         if (err) {
 381                 CERROR("Get connect data failed: %d \n", err);
 382                 GOTO(out_md, err);
 383         }
 384
 385         LASSERT(osfs.os_bsize);
 386         sb->s_blocksize = osfs.os_bsize;
 387         sb->s_blocksize_bits = log2(osfs.os_bsize);
 388         sb->s_magic = LL_SUPER_MAGIC;
 389
 390         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 391          *         retval = in_file->f_op->sendfile(...);
 392          *         if (*ppos > max)
 393          *                 retval = -EOVERFLOW;
 394          *
 395          * it will check if *ppos is greater than max. However, max equals to
 396          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 397          * has been defined as a signed long long ineger in linux kernel. */
 398 #if BITS_PER_LONG == 64
 399         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 400 #else
 401         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 402 #endif
 403         sbi->ll_namelen = osfs.os_namelen;
 404         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 405
 406         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 407             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 408                 LCONSOLE_INFO("Disabling user_xattr feature because "
 409                               "it is not supported on the server\n");
 410                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 411         }
 412
 413         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 414 #ifdef MS_POSIXACL
 415                 sb->s_flags |= MS_POSIXACL;
 416 #endif
 417                 sbi->ll_flags |= LL_SBI_ACL;
 418         } else {
 419                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 420 #ifdef MS_POSIXACL
 421                 sb->s_flags &= ~MS_POSIXACL;
 422 #endif
 423                 sbi->ll_flags &= ~LL_SBI_ACL;
 424         }
 425
 426         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 427                 sbi->ll_flags |= LL_SBI_JOIN;
 428
 429         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 430                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 431                         /* sometimes local client claims to be remote, but mdt
 432                          * will disagree when client gss not applied. */
 433                         LCONSOLE_INFO("client claims to be remote, but server "
 434                                       "rejected, forced to be local.\n");
 435                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 436                 }
 437         } else {
 438                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 439                         /* with gss applied, remote client can not claim to be
 440                          * local, so mdt maybe force client to be remote. */
 441                         LCONSOLE_INFO("client claims to be local, but server "
 442                                       "rejected, forced to be remote.\n");
 443                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 444                 }
 445         }
 446
 447         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 448                 LCONSOLE_INFO("client enabled MDS capability!\n");
 449                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 450         }
 451
 452         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 453                 LCONSOLE_INFO("client enabled OSS capability!\n");
 454                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 455         }
 456
 457         sbi->ll_sdev_orig = sb->s_dev;
 458 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 459         /* We set sb->s_dev equal on all lustre clients in order to support
 460          * NFS export clustering.  NFSD requires that the FSID be the same
 461          * on all clients. */
 462         /* s_dev is also used in lt_compare() to compare two fs, but that is
 463          * only a node-local comparison. */
 464
 465         /* XXX: this will not work with LMV */
 466         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 467                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 468 #endif
 469
 470         obd = class_name2obd(dt);
 471         if (!obd) {
 472                 CERROR("DT %s: not setup or attached\n", dt);
 473                 GOTO(out_md_fid, err = -ENODEV);
 474         }
 475
 476         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 477                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 478                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 479                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 480                                   OBD_CONNECT_AT;
 481         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 482                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 483
 484         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 485                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 486                  * disabled by default, because it can still be enabled on the
 487                  * fly via /proc. As a consequence, we still need to come to an
 488                  * agreement on the supported algorithms at connect time */
 489                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 490
 491                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 492                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 493                 else
 494                         /* send the list of supported checksum types */
 495                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 496         }
 497
 498 #ifdef HAVE_LRU_RESIZE_SUPPORT
 499         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 500 #endif
 501         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 502                "ocd_grant: %d\n", data->ocd_connect_flags,
 503                data->ocd_version, data->ocd_grant);
 504
 505         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 506         obd->obd_upcall.onu_upcall = ll_ocd_update;
 507         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 508
 509         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 510         if (err == -EBUSY) {
 511                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 512                                    "recovery, of which this client is not a "
 513                                    "part.  Please wait for recovery to "
 514                                    "complete, abort, or time out.\n", dt);
 515                 GOTO(out_md_fid, err);
 516         } else if (err) {
 517                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 518                 GOTO(out_md_fid, err);
 519         }
 520
 521         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 522
 523         err = obd_fid_init(sbi->ll_dt_exp);
 524         if (err) {
 525                 CERROR("Can't init data layer FID infrastructure, "
 526                        "rc %d\n", err);
 527                 GOTO(out_dt, err);
 528         }
 529
 530         spin_lock(&sbi->ll_lco.lco_lock);
 531         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 532         spin_unlock(&sbi->ll_lco.lco_lock);
 533
 534         err = obd_register_page_removal_cb(sbi->ll_dt_exp,
 535                                            ll_page_removal_cb,
 536                                            ll_pin_extent_cb);
 537         if (err) {
 538                 CERROR("cannot register page removal callback: rc = %d\n",err);
 539                 GOTO(out_dt, err);
 540         }
 541         err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
 542                                           ll_extent_lock_cancel_cb);
 543         if (err) {
 544                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
 545                 GOTO(out_page_rm_cb, err);
 546         }
 547
 548         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 549         if (err) {
 550                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 551                 GOTO(out_lock_cn_cb, err);
 552         }
 553
 554         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 555                                   0, NULL, NULL, NULL, 0, NULL);
 556         if (err < 0) {
 557                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 558                                    "filesystem. There must be at least one "
 559                                    "active OST for a client to start.\n");
 560                 GOTO(out_lock_cn_cb, err);
 561         }
 562
 563         if (!ll_async_page_slab) {
 564                 ll_async_page_slab_size =
 565                         size_round(sizeof(struct ll_async_page)) + err;
 566                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 567                                                           ll_async_page_slab_size,
 568                                                           0, 0);
 569                 if (!ll_async_page_slab)
 570                         GOTO(out_lock_cn_cb, err = -ENOMEM);
 571         }
 572
 573         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 574         if (err) {
 575                 CERROR("cannot mds_connect: rc = %d\n", err);
 576                 GOTO(out_lock_cn_cb, err);
 577         }
 578         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 579         sbi->ll_root_fid = rootfid;
 580
 581         sb->s_op = &lustre_super_operations;
 582         sb->s_export_op = &lustre_export_operations;
 583
 584         /* make root inode
 585          * XXX: move this to after cbd setup? */
 586         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 587         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 588                 valid |= OBD_MD_FLRMTPERM;
 589         else if (sbi->ll_flags & LL_SBI_ACL)
 590                 valid |= OBD_MD_FLACL;
 591
 592         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 593         if (oc)
 594                 free_capa(oc);
 595         if (err) {
 596                 CERROR("md_getattr failed for root: rc = %d\n", err);
 597                 GOTO(out_lock_cn_cb, err);
 598         }
 599         memset(&lmd, 0, sizeof(lmd));
 600         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 601                                sbi->ll_md_exp, &lmd);
 602         if (err) {
 603                 CERROR("failed to understand root inode md: rc = %d\n", err);
 604                 ptlrpc_req_finished (request);
 605                 GOTO(out_lock_cn_cb, err);
 606         }
 607
 608         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 609         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 610         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 611         ptlrpc_req_finished(request);
 612
 613         if (root == NULL || is_bad_inode(root)) {
 614                 if (lmd.lsm)
 615                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 616 #ifdef CONFIG_FS_POSIX_ACL
 617                 if (lmd.posix_acl) {
 618                         posix_acl_release(lmd.posix_acl);
 619                         lmd.posix_acl = NULL;
 620                 }
 621 #endif
 622                 CERROR("lustre_lite: bad iget4 for root\n");
 623                 GOTO(out_root, err = -EBADF);
 624         }
 625
 626         err = ll_close_thread_start(&sbi->ll_lcq);
 627         if (err) {
 628                 CERROR("cannot start close thread: rc %d\n", err);
 629                 GOTO(out_root, err);
 630         }
 631
 632 #ifdef CONFIG_FS_POSIX_ACL
 633         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 634                 rct_init(&sbi->ll_rct);
 635                 et_init(&sbi->ll_et);
 636         }
 637 #endif
 638
 639         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 640         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 641                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 642                                  NULL);
 643
 644         sb->s_root = d_alloc_root(root);
 645         if (data != NULL)
 646                 OBD_FREE(data, sizeof(*data));
 647         sb->s_root->d_op = &ll_d_root_ops;
 648         RETURN(err);
 649 out_root:
 650         if (root)
 651                 iput(root);
 652 out_lock_cn_cb:
 653         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
 654                                       ll_extent_lock_cancel_cb);
 655 out_page_rm_cb:
 656         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 657                                        ll_page_removal_cb);
 658         obd_fid_fini(sbi->ll_dt_exp);
 659 out_dt:
 660         obd_disconnect(sbi->ll_dt_exp);
 661         sbi->ll_dt_exp = NULL;
 662 out_md_fid:
 663         obd_fid_fini(sbi->ll_md_exp);
 664 out_md:
 665         obd_disconnect(sbi->ll_md_exp);
 666         sbi->ll_md_exp = NULL;
 667 out:
 668         if (data != NULL)
 669                 OBD_FREE_PTR(data);
 670         lprocfs_unregister_mountpoint(sbi);
 671         return err;
 672 }
 673
 674 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 675 {
 676         int size, rc;
 677
 678         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 679         size = sizeof(int);
 680         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 681                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 682         if (rc)
 683                 CERROR("Get max mdsize error rc %d \n", rc);
 684
 685         RETURN(rc);
 686 }
 687
 688 void ll_dump_inode(struct inode *inode)
 689 {
 690         struct list_head *tmp;
 691         int dentry_count = 0;
 692
 693         LASSERT(inode != NULL);
 694
 695         list_for_each(tmp, &inode->i_dentry)
 696                 dentry_count++;
 697
 698         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 699                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 700                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 701 }
 702
 703 void lustre_dump_dentry(struct dentry *dentry, int recur)
 704 {
 705         struct list_head *tmp;
 706         int subdirs = 0;
 707
 708         LASSERT(dentry != NULL);
 709
 710         list_for_each(tmp, &dentry->d_subdirs)
 711                 subdirs++;
 712
 713         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 714                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 715                dentry->d_name.len, dentry->d_name.name,
 716                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 717                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 718                dentry->d_flags, dentry->d_fsdata, subdirs);
 719         if (dentry->d_inode != NULL)
 720                 ll_dump_inode(dentry->d_inode);
 721
 722         if (recur == 0)
 723                 return;
 724
 725         list_for_each(tmp, &dentry->d_subdirs) {
 726                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 727                 lustre_dump_dentry(d, recur - 1);
 728         }
 729 }
 730
 731 #ifdef HAVE_EXPORT___IGET
 732 static void prune_dir_dentries(struct inode *inode)
 733 {
 734         struct dentry *dentry, *prev = NULL;
 735
 736         /* due to lustre specific logic, a directory
 737          * can have few dentries - a bug from VFS POV */
 738 restart:
 739         spin_lock(&dcache_lock);
 740         if (!list_empty(&inode->i_dentry)) {
 741                 dentry = list_entry(inode->i_dentry.prev,
 742                                     struct dentry, d_alias);
 743                 /* in order to prevent infinite loops we
 744                  * break if previous dentry is busy */
 745                 if (dentry != prev) {
 746                         prev = dentry;
 747                         dget_locked(dentry);
 748                         spin_unlock(&dcache_lock);
 749
 750                         /* try to kill all child dentries */
 751                         lock_dentry(dentry);
 752                         shrink_dcache_parent(dentry);
 753                         unlock_dentry(dentry);
 754                         dput(dentry);
 755
 756                         /* now try to get rid of current dentry */
 757                         d_prune_aliases(inode);
 758                         goto restart;
 759                 }
 760         }
 761         spin_unlock(&dcache_lock);
 762 }
 763
 764 static void prune_deathrow_one(struct ll_inode_info *lli)
 765 {
 766         struct inode *inode = ll_info2i(lli);
 767
 768         /* first, try to drop any dentries - they hold a ref on the inode */
 769         if (S_ISDIR(inode->i_mode))
 770                 prune_dir_dentries(inode);
 771         else
 772                 d_prune_aliases(inode);
 773
 774
 775         /* if somebody still uses it, leave it */
 776         LASSERT(atomic_read(&inode->i_count) > 0);
 777         if (atomic_read(&inode->i_count) > 1)
 778                 goto out;
 779
 780         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 781                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 782
 783         /* seems nobody uses it anymore */
 784         inode->i_nlink = 0;
 785
 786 out:
 787         iput(inode);
 788         return;
 789 }
 790
 791 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 792 {
 793         struct ll_inode_info *lli;
 794         int empty;
 795
 796         do {
 797                 if (need_resched() && try)
 798                         break;
 799
 800                 if (try) {
 801                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 802                                 break;
 803                 } else {
 804                         spin_lock(&sbi->ll_deathrow_lock);
 805                 }
 806
 807                 empty = 1;
 808                 lli = NULL;
 809                 if (!list_empty(&sbi->ll_deathrow)) {
 810                         lli = list_entry(sbi->ll_deathrow.next,
 811                                          struct ll_inode_info,
 812                                          lli_dead_list);
 813                         list_del_init(&lli->lli_dead_list);
 814                         if (!list_empty(&sbi->ll_deathrow))
 815                                 empty = 0;
 816                 }
 817                 spin_unlock(&sbi->ll_deathrow_lock);
 818
 819                 if (lli)
 820                         prune_deathrow_one(lli);
 821
 822         } while (empty == 0);
 823 }
 824 #else /* !HAVE_EXPORT___IGET */
 825 #define prune_deathrow(sbi, try) do {} while (0)
 826 #endif /* HAVE_EXPORT___IGET */
 827
 828 void client_common_put_super(struct super_block *sb)
 829 {
 830         struct ll_sb_info *sbi = ll_s2sbi(sb);
 831         ENTRY;
 832
 833 #ifdef CONFIG_FS_POSIX_ACL
 834         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 835                 et_fini(&sbi->ll_et);
 836                 rct_fini(&sbi->ll_rct);
 837         }
 838 #endif
 839
 840         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 841
 842         ll_close_thread_shutdown(sbi->ll_lcq);
 843
 844         /* destroy inodes in deathrow */
 845         prune_deathrow(sbi, 0);
 846
 847         list_del(&sbi->ll_conn_chain);
 848
 849         obd_fid_fini(sbi->ll_dt_exp);
 850         obd_disconnect(sbi->ll_dt_exp);
 851         sbi->ll_dt_exp = NULL;
 852
 853         lprocfs_unregister_mountpoint(sbi);
 854
 855         obd_fid_fini(sbi->ll_md_exp);
 856         obd_disconnect(sbi->ll_md_exp);
 857         sbi->ll_md_exp = NULL;
 858
 859         EXIT;
 860 }
 861
 862 void ll_kill_super(struct super_block *sb)
 863 {
 864         struct ll_sb_info *sbi;
 865
 866         ENTRY;
 867
 868         /* not init sb ?*/
 869         if (!(sb->s_flags & MS_ACTIVE))
 870                 return;
 871
 872         sbi = ll_s2sbi(sb);
 873         /* we need restore s_dev from changed for clustred NFS before put_super
 874          * because new kernels have cached s_dev and change sb->s_dev in
 875          * put_super not affected real removing devices */
 876         if (sbi)
 877                 sb->s_dev = sbi->ll_sdev_orig;
 878         EXIT;
 879 }
 880
 881 char *ll_read_opt(const char *opt, char *data)
 882 {
 883         char *value;
 884         char *retval;
 885         ENTRY;
 886
 887         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 888         if (strncmp(opt, data, strlen(opt)))
 889                 RETURN(NULL);
 890         if ((value = strchr(data, '=')) == NULL)
 891                 RETURN(NULL);
 892
 893         value++;
 894         OBD_ALLOC(retval, strlen(value) + 1);
 895         if (!retval) {
 896                 CERROR("out of memory!\n");
 897                 RETURN(NULL);
 898         }
 899
 900         memcpy(retval, value, strlen(value)+1);
 901         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 902         RETURN(retval);
 903 }
 904
 905 static inline int ll_set_opt(const char *opt, char *data, int fl)
 906 {
 907         if (strncmp(opt, data, strlen(opt)) != 0)
 908                 return(0);
 909         else
 910                 return(fl);
 911 }
 912
 913 /* non-client-specific mount options are parsed in lmd_parse */
 914 static int ll_options(char *options, int *flags)
 915 {
 916         int tmp;
 917         char *s1 = options, *s2;
 918         ENTRY;
 919
 920         if (!options)
 921                 RETURN(0);
 922
 923         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 924
 925         while (*s1) {
 926                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 927                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 928                 if (tmp) {
 929                         *flags |= tmp;
 930                         goto next;
 931                 }
 932                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 933                 if (tmp) {
 934                         *flags |= tmp;
 935                         goto next;
 936                 }
 937                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 938                 if (tmp) {
 939                         *flags |= tmp;
 940                         goto next;
 941                 }
 942                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 943                 if (tmp) {
 944                         *flags &= ~tmp;
 945                         goto next;
 946                 }
 947                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 948                 if (tmp) {
 949                         *flags |= tmp;
 950                         goto next;
 951                 }
 952                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 953                 if (tmp) {
 954                         *flags &= ~tmp;
 955                         goto next;
 956                 }
 957                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 958                 if (tmp) {
 959                         /* Ignore deprecated mount option.  The client will
 960                          * always try to mount with ACL support, whether this
 961                          * is used depends on whether server supports it. */
 962                         goto next;
 963                 }
 964                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 965                 if (tmp) {
 966                         goto next;
 967                 }
 968                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 969                 if (tmp) {
 970                         *flags |= tmp;
 971                         goto next;
 972                 }
 973
 974                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 975                 if (tmp) {
 976                         *flags |= tmp;
 977                         goto next;
 978                 }
 979                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 980                 if (tmp) {
 981                         *flags &= ~tmp;
 982                         goto next;
 983                 }
 984                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 985                 if (tmp) {
 986                         *flags |= tmp;
 987                         goto next;
 988                 }
 989                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 990                 if (tmp) {
 991                         *flags &= ~tmp;
 992                         goto next;
 993                 }
 994
 995                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 996                                    s1);
 997                 RETURN(-EINVAL);
 998
 999 next:
1000                 /* Find next opt */
1001                 s2 = strchr(s1, ',');
1002                 if (s2 == NULL)
1003                         break;
1004                 s1 = s2 + 1;
1005         }
1006         RETURN(0);
1007 }
1008
1009 void ll_lli_init(struct ll_inode_info *lli)
1010 {
1011         lli->lli_inode_magic = LLI_INODE_MAGIC;
1012         sema_init(&lli->lli_size_sem, 1);
1013         sema_init(&lli->lli_write_sem, 1);
1014         lli->lli_flags = 0;
1015         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1016         spin_lock_init(&lli->lli_lock);
1017         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1018         INIT_LIST_HEAD(&lli->lli_close_list);
1019         lli->lli_inode_magic = LLI_INODE_MAGIC;
1020         sema_init(&lli->lli_och_sem, 1);
1021         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1022         lli->lli_mds_exec_och = NULL;
1023         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1024         lli->lli_open_fd_exec_count = 0;
1025         INIT_LIST_HEAD(&lli->lli_dead_list);
1026         lli->lli_remote_perms = NULL;
1027         lli->lli_rmtperm_utime = 0;
1028         sema_init(&lli->lli_rmtperm_sem, 1);
1029         INIT_LIST_HEAD(&lli->lli_oss_capas);
1030 }
1031
1032 int ll_fill_super(struct super_block *sb)
1033 {
1034         struct lustre_profile *lprof;
1035         struct lustre_sb_info *lsi = s2lsi(sb);
1036         struct ll_sb_info *sbi;
1037         char  *dt = NULL, *md = NULL;
1038         char  *profilenm = get_profile_name(sb);
1039         struct config_llog_instance cfg = {0, };
1040         char   ll_instance[sizeof(sb) * 2 + 1];
1041         int    err;
1042         ENTRY;
1043
1044         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1045
1046         cfs_module_get();
1047
1048         /* client additional sb info */
1049         lsi->lsi_llsbi = sbi = ll_init_sbi();
1050         if (!sbi) {
1051                 cfs_module_put();
1052                 RETURN(-ENOMEM);
1053         }
1054
1055         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1056         if (err)
1057                 GOTO(out_free, err);
1058
1059         /* Generate a string unique to this super, in case some joker tries
1060            to mount the same fs at two mount points.
1061            Use the address of the super itself.*/
1062         sprintf(ll_instance, "%p", sb);
1063         cfg.cfg_instance = ll_instance;
1064         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1065
1066         /* set up client obds */
1067         err = lustre_process_log(sb, profilenm, &cfg);
1068         if (err < 0) {
1069                 CERROR("Unable to process log: %d\n", err);
1070                 GOTO(out_free, err);
1071         }
1072
1073         lprof = class_get_profile(profilenm);
1074         if (lprof == NULL) {
1075                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1076                                    " read from the MGS.  Does that filesystem "
1077                                    "exist?\n", profilenm);
1078                 GOTO(out_free, err = -EINVAL);
1079         }
1080         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1081                lprof->lp_md, lprof->lp_dt);
1082
1083         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1084                   strlen(ll_instance) + 2);
1085         if (!dt)
1086                 GOTO(out_free, err = -ENOMEM);
1087         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1088
1089         OBD_ALLOC(md, strlen(lprof->lp_md) +
1090                   strlen(ll_instance) + 2);
1091         if (!md)
1092                 GOTO(out_free, err = -ENOMEM);
1093         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1094
1095         /* connections, registrations, sb setup */
1096         err = client_common_fill_super(sb, md, dt);
1097
1098 out_free:
1099         if (md)
1100                 OBD_FREE(md, strlen(md) + 1);
1101         if (dt)
1102                 OBD_FREE(dt, strlen(dt) + 1);
1103         if (err)
1104                 ll_put_super(sb);
1105         else
1106                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1107
1108         RETURN(err);
1109 } /* ll_fill_super */
1110
1111
1112 void ll_put_super(struct super_block *sb)
1113 {
1114         struct config_llog_instance cfg;
1115         char   ll_instance[sizeof(sb) * 2 + 1];
1116         struct obd_device *obd;
1117         struct lustre_sb_info *lsi = s2lsi(sb);
1118         struct ll_sb_info *sbi = ll_s2sbi(sb);
1119         char *profilenm = get_profile_name(sb);
1120         int force = 1, next;
1121         ENTRY;
1122
1123         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1124
1125         ll_print_capa_stat(sbi);
1126
1127         sprintf(ll_instance, "%p", sb);
1128         cfg.cfg_instance = ll_instance;
1129         lustre_end_log(sb, NULL, &cfg);
1130
1131         if (sbi->ll_md_exp) {
1132                 obd = class_exp2obd(sbi->ll_md_exp);
1133                 if (obd)
1134                         force = obd->obd_force;
1135         }
1136
1137         /* We need to set force before the lov_disconnect in
1138            lustre_common_put_super, since l_d cleans up osc's as well. */
1139         if (force) {
1140                 next = 0;
1141                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1142                                                      &next)) != NULL) {
1143                         obd->obd_force = force;
1144                 }
1145         }
1146
1147         if (sbi->ll_lcq) {
1148                 /* Only if client_common_fill_super succeeded */
1149                 client_common_put_super(sb);
1150         }
1151         next = 0;
1152         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1153                 class_manual_cleanup(obd);
1154         }
1155
1156         if (profilenm)
1157                 class_del_profile(profilenm);
1158
1159         ll_free_sbi(sb);
1160         lsi->lsi_llsbi = NULL;
1161
1162         lustre_common_put_super(sb);
1163
1164         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1165
1166         cfs_module_put();
1167
1168         EXIT;
1169 } /* client_put_super */
1170
1171 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1172
1173 #if defined(HAVE_CACHE_RETURN_INT)
1174 static int
1175 #else
1176 static void
1177 #endif
1178 ll_shrink_cache(int priority, unsigned int gfp_mask)
1179 {
1180         struct ll_sb_info *sbi;
1181         int count = 0;
1182
1183         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1184                 count += llap_shrink_cache(sbi, priority);
1185
1186 #if defined(HAVE_CACHE_RETURN_INT)
1187         return count;
1188 #endif
1189 }
1190
1191 struct cache_definition ll_cache_definition = {
1192         .name = "llap_cache",
1193         .shrink = ll_shrink_cache
1194 };
1195 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1196
1197 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1198 {
1199         struct inode *inode = NULL;
1200         /* NOTE: we depend on atomic igrab() -bzzz */
1201         lock_res_and_lock(lock);
1202         if (lock->l_ast_data) {
1203                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1204                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1205                         inode = igrab(lock->l_ast_data);
1206                 } else {
1207                         inode = lock->l_ast_data;
1208                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1209                                                 D_INFO : D_WARNING,
1210                                         lock, __FILE__, __func__, __LINE__,
1211                                         "l_ast_data %p is bogus: magic %08x",
1212                                         lock->l_ast_data, lli->lli_inode_magic);
1213                         inode = NULL;
1214                 }
1215         }
1216         unlock_res_and_lock(lock);
1217         return inode;
1218 }
1219
1220 static int null_if_equal(struct ldlm_lock *lock, void *data)
1221 {
1222         if (data == lock->l_ast_data) {
1223                 lock->l_ast_data = NULL;
1224
1225                 if (lock->l_req_mode != lock->l_granted_mode)
1226                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1227         }
1228
1229         return LDLM_ITER_CONTINUE;
1230 }
1231
1232 void ll_clear_inode(struct inode *inode)
1233 {
1234         struct ll_inode_info *lli = ll_i2info(inode);
1235         struct ll_sb_info *sbi = ll_i2sbi(inode);
1236         ENTRY;
1237
1238         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1239                inode->i_generation, inode);
1240
1241         if (S_ISDIR(inode->i_mode)) {
1242                 /* these should have been cleared in ll_file_release */
1243                 LASSERT(lli->lli_sai == NULL);
1244                 LASSERT(lli->lli_opendir_key == NULL);
1245                 LASSERT(lli->lli_opendir_pid == 0);
1246         }
1247
1248         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1249         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1250                          null_if_equal, inode);
1251
1252         LASSERT(!lli->lli_open_fd_write_count);
1253         LASSERT(!lli->lli_open_fd_read_count);
1254         LASSERT(!lli->lli_open_fd_exec_count);
1255
1256         if (lli->lli_mds_write_och)
1257                 ll_md_real_close(inode, FMODE_WRITE);
1258         if (lli->lli_mds_exec_och)
1259                 ll_md_real_close(inode, FMODE_EXEC);
1260         if (lli->lli_mds_read_och)
1261                 ll_md_real_close(inode, FMODE_READ);
1262
1263         if (lli->lli_smd) {
1264                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1265                                   null_if_equal, inode);
1266
1267                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1268                 lli->lli_smd = NULL;
1269         }
1270
1271         if (lli->lli_symlink_name) {
1272                 OBD_FREE(lli->lli_symlink_name,
1273                          strlen(lli->lli_symlink_name) + 1);
1274                 lli->lli_symlink_name = NULL;
1275         }
1276
1277         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1278                 LASSERT(lli->lli_posix_acl == NULL);
1279                 if (lli->lli_remote_perms) {
1280                         free_rmtperm_hash(lli->lli_remote_perms);
1281                         lli->lli_remote_perms = NULL;
1282                 }
1283         }
1284 #ifdef CONFIG_FS_POSIX_ACL
1285         else if (lli->lli_posix_acl) {
1286                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1287                 LASSERT(lli->lli_remote_perms == NULL);
1288                 posix_acl_release(lli->lli_posix_acl);
1289                 lli->lli_posix_acl = NULL;
1290         }
1291 #endif
1292         lli->lli_inode_magic = LLI_INODE_DEAD;
1293
1294 #ifdef HAVE_EXPORT___IGET
1295         spin_lock(&sbi->ll_deathrow_lock);
1296         list_del_init(&lli->lli_dead_list);
1297         spin_unlock(&sbi->ll_deathrow_lock);
1298 #endif
1299         ll_clear_inode_capas(inode);
1300
1301         EXIT;
1302 }
1303
1304 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1305                   struct md_open_data **mod)
1306 {
1307         struct lustre_md md;
1308         struct ll_sb_info *sbi = ll_i2sbi(inode);
1309         struct ptlrpc_request *request = NULL;
1310         int rc;
1311         ENTRY;
1312
1313         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1314                                      LUSTRE_OPC_ANY, NULL);
1315         if (IS_ERR(op_data))
1316                 RETURN(PTR_ERR(op_data));
1317
1318         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1319                         &request, mod);
1320         if (rc) {
1321                 ptlrpc_req_finished(request);
1322                 if (rc == -ENOENT) {
1323                         inode->i_nlink = 0;
1324                         /* Unlinked special device node? Or just a race?
1325                          * Pretend we done everything. */
1326                         if (!S_ISREG(inode->i_mode) &&
1327                             !S_ISDIR(inode->i_mode))
1328                                 rc = inode_setattr(inode, &op_data->op_attr);
1329                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1330                         CERROR("md_setattr fails: rc = %d\n", rc);
1331                 }
1332                 RETURN(rc);
1333         }
1334
1335         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1336                               sbi->ll_md_exp, &md);
1337         if (rc) {
1338                 ptlrpc_req_finished(request);
1339                 RETURN(rc);
1340         }
1341
1342         /* We call inode_setattr to adjust timestamps.
1343          * If there is at least some data in file, we cleared ATTR_SIZE
1344          * above to avoid invoking vmtruncate, otherwise it is important
1345          * to call vmtruncate in inode_setattr to update inode->i_size
1346          * (bug 6196) */
1347         rc = inode_setattr(inode, &op_data->op_attr);
1348
1349         /* Extract epoch data if obtained. */
1350         op_data->op_handle = md.body->handle;
1351         op_data->op_ioepoch = md.body->ioepoch;
1352
1353         ll_update_inode(inode, &md);
1354         ptlrpc_req_finished(request);
1355
1356         RETURN(rc);
1357 }
1358
1359 /* Close IO epoch and send Size-on-MDS attribute update. */
1360 static int ll_setattr_done_writing(struct inode *inode,
1361                                    struct md_op_data *op_data,
1362                                    struct md_open_data *mod)
1363 {
1364         struct ll_inode_info *lli = ll_i2info(inode);
1365         int rc = 0;
1366         ENTRY;
1367
1368         LASSERT(op_data != NULL);
1369         if (!S_ISREG(inode->i_mode))
1370                 RETURN(0);
1371
1372         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1373                op_data->op_ioepoch, PFID(&lli->lli_fid));
1374
1375         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1376         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1377         if (rc == -EAGAIN) {
1378                 /* MDS has instructed us to obtain Size-on-MDS attribute
1379                  * from OSTs and send setattr to back to MDS. */
1380                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1381                                          op_data->op_ioepoch);
1382         } else if (rc) {
1383                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1384                        inode->i_ino, rc);
1385         }
1386         RETURN(rc);
1387 }
1388
1389 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1390 {
1391         struct ll_sb_info *sbi = ll_i2sbi(inode);
1392         struct ll_inode_info *lli = ll_i2info(inode);
1393         struct lov_stripe_md *lsm = lli->lli_smd;
1394         int rc;
1395         ldlm_policy_data_t policy = { .l_extent = {new_size,
1396                                                    OBD_OBJECT_EOF } };
1397         struct lustre_handle lockh = { 0 };
1398         int local_lock = 0; /* 0 - no local lock;
1399                              * 1 - lock taken by lock_extent;
1400                              * 2 - by obd_match*/
1401         int ast_flags;
1402         int err;
1403         ENTRY;
1404
1405         UNLOCK_INODE_MUTEX(inode);
1406         UP_WRITE_I_ALLOC_SEM(inode);
1407
1408         if (sbi->ll_lockless_truncate_enable &&
1409             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1410                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1411                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1412                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1413                 if (rc > 0) {
1414                         local_lock = 2;
1415                         rc = 0;
1416                 } else if (rc == 0) {
1417                         rc = ll_file_punch(inode, new_size, 1);
1418                 }
1419         } else {
1420                 /* XXX when we fix the AST intents to pass the discard-range
1421                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1422                  * XXX here. */
1423                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1424                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1425                                     &lockh, ast_flags);
1426                 if (likely(rc == 0))
1427                         local_lock = 1;
1428         }
1429
1430         LOCK_INODE_MUTEX(inode);
1431         DOWN_WRITE_I_ALLOC_SEM(inode);
1432
1433         if (likely(rc == 0)) {
1434                 /* Only ll_inode_size_lock is taken at this level.
1435                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1436                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1437                  * ll_truncate dropped ll_inode_size_lock() */
1438                 ll_inode_size_lock(inode, 0);
1439                 if (!local_lock) {
1440                         spin_lock(&lli->lli_lock);
1441                         lli->lli_flags |= LLIF_SRVLOCK;
1442                         spin_unlock(&lli->lli_lock);
1443                 }
1444                 rc = vmtruncate(inode, new_size);
1445                 if (!local_lock) {
1446                         spin_lock(&lli->lli_lock);
1447                         lli->lli_flags &= ~LLIF_SRVLOCK;
1448                         spin_unlock(&lli->lli_lock);
1449                 }
1450                 if (rc != 0) {
1451                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1452                         ll_inode_size_unlock(inode, 0);
1453                 }
1454         }
1455
1456         if (local_lock) {
1457                 if (local_lock == 2)
1458                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1459                 else
1460                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1461                 if (unlikely(err != 0)){
1462                         CERROR("extent unlock failed: err=%d,"
1463                                " unlock method =%d\n", err, local_lock);
1464                         if (rc == 0)
1465                                 rc = err;
1466                 }
1467         }
1468         RETURN(rc);
1469 }
1470
1471 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1472  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1473  * keep these values until such a time that objects are allocated for it.
1474  * We do the MDS operations first, as it is checking permissions for us.
1475  * We don't to the MDS RPC if there is nothing that we want to store there,
1476  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1477  * going to do an RPC anyways.
1478  *
1479  * If we are doing a truncate, we will send the mtime and ctime updates
1480  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1481  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1482  * at the same time.
1483  */
1484 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1485 {
1486         struct ll_inode_info *lli = ll_i2info(inode);
1487         struct lov_stripe_md *lsm = lli->lli_smd;
1488         struct ll_sb_info *sbi = ll_i2sbi(inode);
1489         struct md_op_data *op_data = NULL;
1490         struct md_open_data *mod = NULL;
1491         int ia_valid = attr->ia_valid;
1492         int rc = 0, rc1 = 0;
1493         ENTRY;
1494
1495         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1496                attr->ia_valid);
1497         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1498
1499         if (ia_valid & ATTR_SIZE) {
1500                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1501                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1502                                attr->ia_size, ll_file_maxbytes(inode));
1503                         RETURN(-EFBIG);
1504                 }
1505
1506                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1507         }
1508
1509         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1510         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1511                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1512                         RETURN(-EPERM);
1513         }
1514
1515         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1516         if (attr->ia_valid & ATTR_CTIME) {
1517                 attr->ia_ctime = CURRENT_TIME;
1518                 attr->ia_valid |= ATTR_CTIME_SET;
1519         }
1520         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1521                 attr->ia_atime = CURRENT_TIME;
1522                 attr->ia_valid |= ATTR_ATIME_SET;
1523         }
1524         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1525                 attr->ia_mtime = CURRENT_TIME;
1526                 attr->ia_valid |= ATTR_MTIME_SET;
1527         }
1528         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1529                 /* To avoid stale mtime on mds, obtain it from ost and send
1530                    to mds. */
1531                 rc = ll_glimpse_size(inode, 0);
1532                 if (rc)
1533                         RETURN(rc);
1534
1535                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1536                 attr->ia_mtime = inode->i_mtime;
1537         }
1538
1539         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1540                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1541                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1542                        cfs_time_current_sec());
1543
1544         /* NB: ATTR_SIZE will only be set after this point if the size
1545          * resides on the MDS, ie, this file has no objects. */
1546         if (lsm)
1547                 attr->ia_valid &= ~ATTR_SIZE;
1548
1549         /* We always do an MDS RPC, even if we're only changing the size;
1550          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1551
1552         OBD_ALLOC_PTR(op_data);
1553         if (op_data == NULL)
1554                 RETURN(-ENOMEM);
1555
1556         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1557
1558         /* Open epoch for truncate. */
1559         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1560             (ia_valid & ATTR_SIZE))
1561                 op_data->op_flags = MF_EPOCH_OPEN;
1562
1563         rc = ll_md_setattr(inode, op_data, &mod);
1564         if (rc)
1565                 GOTO(out, rc);
1566
1567         if (op_data->op_ioepoch)
1568                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1569                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1570
1571         if (!lsm || !S_ISREG(inode->i_mode)) {
1572                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1573                 GOTO(out, rc = 0);
1574         }
1575
1576         /* We really need to get our PW lock before we change inode->i_size.
1577          * If we don't we can race with other i_size updaters on our node, like
1578          * ll_file_read.  We can also race with i_size propogation to other
1579          * nodes through dirtying and writeback of final cached pages.  This
1580          * last one is especially bad for racing o_append users on other
1581          * nodes. */
1582         if (ia_valid & ATTR_SIZE) {
1583                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1584         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1585                 obd_flag flags;
1586                 struct obd_info oinfo = { { { 0 } } };
1587                 struct obdo *oa;
1588
1589                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1590                        inode->i_ino, LTIME_S(attr->ia_mtime));
1591
1592                 OBDO_ALLOC(oa);
1593                 if (oa) {
1594                         oa->o_id = lsm->lsm_object_id;
1595                         oa->o_gr = lsm->lsm_object_gr;
1596                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1597
1598                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1599                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1600                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1601                                 OBD_MD_FLGROUP;
1602
1603                         obdo_from_inode(oa, inode, flags);
1604
1605                         oinfo.oi_oa = oa;
1606                         oinfo.oi_md = lsm;
1607                         oinfo.oi_capa = ll_mdscapa_get(inode);
1608
1609                         /* XXX: this looks unnecessary now. */
1610                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1611                         capa_put(oinfo.oi_capa);
1612                         if (rc)
1613                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1614                         OBDO_FREE(oa);
1615                 } else {
1616                         rc = -ENOMEM;
1617                 }
1618         }
1619         EXIT;
1620 out:
1621         if (op_data) {
1622                 if (op_data->op_ioepoch)
1623                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1624                 ll_finish_md_op_data(op_data);
1625         }
1626         return rc ? rc : rc1;
1627 }
1628
1629 int ll_setattr(struct dentry *de, struct iattr *attr)
1630 {
1631         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1632             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1633                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1634
1635         return ll_setattr_raw(de->d_inode, attr);
1636 }
1637
1638 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1639                        __u64 max_age, __u32 flags)
1640 {
1641         struct ll_sb_info *sbi = ll_s2sbi(sb);
1642         struct obd_statfs obd_osfs;
1643         int rc;
1644         ENTRY;
1645
1646         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1647         if (rc) {
1648                 CERROR("md_statfs fails: rc = %d\n", rc);
1649                 RETURN(rc);
1650         }
1651
1652         osfs->os_type = sb->s_magic;
1653
1654         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1655                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1656
1657         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1658                               &obd_osfs, max_age, flags);
1659         if (rc) {
1660                 CERROR("obd_statfs fails: rc = %d\n", rc);
1661                 RETURN(rc);
1662         }
1663
1664         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1665                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1666                obd_osfs.os_files);
1667
1668         osfs->os_bsize = obd_osfs.os_bsize;
1669         osfs->os_blocks = obd_osfs.os_blocks;
1670         osfs->os_bfree = obd_osfs.os_bfree;
1671         osfs->os_bavail = obd_osfs.os_bavail;
1672
1673         /* If we don't have as many objects free on the OST as inodes
1674          * on the MDS, we reduce the total number of inodes to
1675          * compensate, so that the "inodes in use" number is correct.
1676          */
1677         if (obd_osfs.os_ffree < osfs->os_ffree) {
1678                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1679                         obd_osfs.os_ffree;
1680                 osfs->os_ffree = obd_osfs.os_ffree;
1681         }
1682
1683         RETURN(rc);
1684 }
1685 #ifndef HAVE_STATFS_DENTRY_PARAM
1686 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1687 {
1688 #else
1689 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1690 {
1691         struct super_block *sb = de->d_sb;
1692 #endif
1693         struct obd_statfs osfs;
1694         int rc;
1695
1696         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1697         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1698
1699         /* For now we will always get up-to-date statfs values, but in the
1700          * future we may allow some amount of caching on the client (e.g.
1701          * from QOS or lprocfs updates). */
1702         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1703         if (rc)
1704                 return rc;
1705
1706         statfs_unpack(sfs, &osfs);
1707
1708         /* We need to downshift for all 32-bit kernels, because we can't
1709          * tell if the kernel is being called via sys_statfs64() or not.
1710          * Stop before overflowing f_bsize - in which case it is better
1711          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1712         if (sizeof(long) < 8) {
1713                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1714                         sfs->f_bsize <<= 1;
1715
1716                         osfs.os_blocks >>= 1;
1717                         osfs.os_bfree >>= 1;
1718                         osfs.os_bavail >>= 1;
1719                 }
1720         }
1721
1722         sfs->f_blocks = osfs.os_blocks;
1723         sfs->f_bfree = osfs.os_bfree;
1724         sfs->f_bavail = osfs.os_bavail;
1725
1726         return 0;
1727 }
1728
1729 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1730 {
1731         struct ll_inode_info *lli;
1732         struct lov_stripe_md *lsm;
1733
1734         lli = ll_i2info(inode);
1735         LASSERT(lli->lli_size_sem_owner != current);
1736         down(&lli->lli_size_sem);
1737         LASSERT(lli->lli_size_sem_owner == NULL);
1738         lli->lli_size_sem_owner = current;
1739         lsm = lli->lli_smd;
1740         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1741                  lsm, lock_lsm);
1742         if (lock_lsm)
1743                 lov_stripe_lock(lsm);
1744 }
1745
1746 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1747 {
1748         struct ll_inode_info *lli;
1749         struct lov_stripe_md *lsm;
1750
1751         lli = ll_i2info(inode);
1752         lsm = lli->lli_smd;
1753         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1754                  lsm, unlock_lsm);
1755         if (unlock_lsm)
1756                 lov_stripe_unlock(lsm);
1757         LASSERT(lli->lli_size_sem_owner == current);
1758         lli->lli_size_sem_owner = NULL;
1759         up(&lli->lli_size_sem);
1760 }
1761
1762 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1763 {
1764         struct ll_inode_info *lli = ll_i2info(inode);
1765
1766         dump_lsm(D_INODE, lsm);
1767         dump_lsm(D_INODE, lli->lli_smd);
1768         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1769                  "lsm must be joined lsm %p\n", lsm);
1770         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1771         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1772                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1773         lli->lli_smd = lsm;
1774         lli->lli_maxbytes = lsm->lsm_maxbytes;
1775         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1776                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1777 }
1778
1779 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1780 {
1781         struct ll_inode_info *lli = ll_i2info(inode);
1782         struct mdt_body *body = md->body;
1783         struct lov_stripe_md *lsm = md->lsm;
1784         struct ll_sb_info *sbi = ll_i2sbi(inode);
1785
1786         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1787         if (lsm != NULL) {
1788                 if (lli->lli_smd == NULL) {
1789                         if (lsm->lsm_magic != LOV_MAGIC &&
1790                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1791                                 dump_lsm(D_ERROR, lsm);
1792                                 LBUG();
1793                         }
1794                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1795                                lsm, inode->i_ino, inode->i_generation, inode);
1796                         /* ll_inode_size_lock() requires it is only called
1797                          * with lli_smd != NULL or lock_lsm == 0 or we can
1798                          * race between lock/unlock.  bug 9547 */
1799                         lli->lli_smd = lsm;
1800                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1801                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1802                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1803                 } else {
1804                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1805                              lli->lli_smd->lsm_stripe_count ==
1806                                         lsm->lsm_stripe_count) {
1807                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1808                                         CERROR("lsm mismatch for inode %ld\n",
1809                                                 inode->i_ino);
1810                                         CERROR("lli_smd:\n");
1811                                         dump_lsm(D_ERROR, lli->lli_smd);
1812                                         CERROR("lsm:\n");
1813                                         dump_lsm(D_ERROR, lsm);
1814                                         LBUG();
1815                                 }
1816                         } else
1817                                 ll_replace_lsm(inode, lsm);
1818                 }
1819                 if (lli->lli_smd != lsm)
1820                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1821         }
1822
1823         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1824                 if (body->valid & OBD_MD_FLRMTPERM)
1825                         ll_update_remote_perm(inode, md->remote_perm);
1826         }
1827 #ifdef CONFIG_FS_POSIX_ACL
1828         else if (body->valid & OBD_MD_FLACL) {
1829                 spin_lock(&lli->lli_lock);
1830                 if (lli->lli_posix_acl)
1831                         posix_acl_release(lli->lli_posix_acl);
1832                 lli->lli_posix_acl = md->posix_acl;
1833                 spin_unlock(&lli->lli_lock);
1834         }
1835 #endif
1836         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1837
1838         if (body->valid & OBD_MD_FLATIME &&
1839             body->atime > LTIME_S(inode->i_atime))
1840                 LTIME_S(inode->i_atime) = body->atime;
1841
1842         /* mtime is always updated with ctime, but can be set in past.
1843            As write and utime(2) may happen within 1 second, and utime's
1844            mtime has a priority over write's one, so take mtime from mds
1845            for the same ctimes. */
1846         if (body->valid & OBD_MD_FLCTIME &&
1847             body->ctime >= LTIME_S(inode->i_ctime)) {
1848                 LTIME_S(inode->i_ctime) = body->ctime;
1849                 if (body->valid & OBD_MD_FLMTIME) {
1850                         CDEBUG(D_INODE, "setting ino %lu mtime "
1851                                "from %lu to "LPU64"\n", inode->i_ino,
1852                                LTIME_S(inode->i_mtime), body->mtime);
1853                         LTIME_S(inode->i_mtime) = body->mtime;
1854                 }
1855         }
1856         if (body->valid & OBD_MD_FLMODE)
1857                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1858         if (body->valid & OBD_MD_FLTYPE)
1859                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1860         if (S_ISREG(inode->i_mode)) {
1861                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1862         } else {
1863                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1864         }
1865 #ifdef HAVE_INODE_BLKSIZE
1866         inode->i_blksize = 1<<inode->i_blkbits;
1867 #endif
1868         if (body->valid & OBD_MD_FLUID)
1869                 inode->i_uid = body->uid;
1870         if (body->valid & OBD_MD_FLGID)
1871                 inode->i_gid = body->gid;
1872         if (body->valid & OBD_MD_FLFLAGS)
1873                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1874         if (body->valid & OBD_MD_FLNLINK)
1875                 inode->i_nlink = body->nlink;
1876         if (body->valid & OBD_MD_FLRDEV)
1877                 inode->i_rdev = old_decode_dev(body->rdev);
1878
1879         if (body->valid & OBD_MD_FLID) {
1880                 /* FID shouldn't be changed! */
1881                 if (fid_is_sane(&lli->lli_fid)) {
1882                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1883                                  "Trying to change FID "DFID
1884                                  " to the "DFID", inode %lu/%u(%p)\n",
1885                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1886                                  inode->i_ino, inode->i_generation, inode);
1887                 } else
1888                         lli->lli_fid = body->fid1;
1889         }
1890
1891         LASSERT(fid_seq(&lli->lli_fid) != 0);
1892
1893         if (body->valid & OBD_MD_FLSIZE) {
1894                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1895                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1896                         struct lustre_handle lockh;
1897                         ldlm_mode_t mode;
1898
1899                         /* As it is possible a blocking ast has been processed
1900                          * by this time, we need to check there is an UPDATE
1901                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1902                          * it. */
1903                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1904                                                &lockh);
1905                         if (mode) {
1906                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1907                                                       LLIF_EPOCH_PENDING |
1908                                                       LLIF_SOM_DIRTY)) {
1909                                         CERROR("ino %lu flags %lu still has "
1910                                                "size authority! do not trust "
1911                                                "the size got from MDS\n",
1912                                                inode->i_ino, lli->lli_flags);
1913                                 } else {
1914                                         /* Use old size assignment to avoid
1915                                          * deadlock bz14138 & bz14326 */
1916                                         inode->i_size = body->size;
1917                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1918                                 }
1919                                 ldlm_lock_decref(&lockh, mode);
1920                         }
1921                 } else {
1922                         /* Use old size assignment to avoid
1923                          * deadlock bz14138 & bz14326 */
1924                         inode->i_size = body->size;
1925                 }
1926
1927                 if (body->valid & OBD_MD_FLBLOCKS)
1928                         inode->i_blocks = body->blocks;
1929         }
1930
1931         if (body->valid & OBD_MD_FLMDSCAPA) {
1932                 LASSERT(md->mds_capa);
1933                 ll_add_capa(inode, md->mds_capa);
1934         }
1935         if (body->valid & OBD_MD_FLOSSCAPA) {
1936                 LASSERT(md->oss_capa);
1937                 ll_add_capa(inode, md->oss_capa);
1938         }
1939 }
1940
1941 static struct backing_dev_info ll_backing_dev_info = {
1942         .ra_pages       = 0,    /* No readahead */
1943 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1944         .capabilities   = 0,    /* Does contribute to dirty memory */
1945 #else
1946         .memory_backed  = 0,    /* Does contribute to dirty memory */
1947 #endif
1948 };
1949
1950 void ll_read_inode2(struct inode *inode, void *opaque)
1951 {
1952         struct lustre_md *md = opaque;
1953         struct ll_inode_info *lli = ll_i2info(inode);
1954         ENTRY;
1955
1956         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1957                inode->i_ino, inode->i_generation, inode);
1958
1959         ll_lli_init(lli);
1960
1961         LASSERT(!lli->lli_smd);
1962
1963         /* Core attributes from the MDS first.  This is a new inode, and
1964          * the VFS doesn't zero times in the core inode so we have to do
1965          * it ourselves.  They will be overwritten by either MDS or OST
1966          * attributes - we just need to make sure they aren't newer. */
1967         LTIME_S(inode->i_mtime) = 0;
1968         LTIME_S(inode->i_atime) = 0;
1969         LTIME_S(inode->i_ctime) = 0;
1970         inode->i_rdev = 0;
1971         ll_update_inode(inode, md);
1972
1973         /* OIDEBUG(inode); */
1974
1975         if (S_ISREG(inode->i_mode)) {
1976                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1977                 inode->i_op = &ll_file_inode_operations;
1978                 inode->i_fop = sbi->ll_fop;
1979                 inode->i_mapping->a_ops = &ll_aops;
1980                 EXIT;
1981         } else if (S_ISDIR(inode->i_mode)) {
1982                 inode->i_op = &ll_dir_inode_operations;
1983                 inode->i_fop = &ll_dir_operations;
1984                 inode->i_mapping->a_ops = &ll_dir_aops;
1985                 EXIT;
1986         } else if (S_ISLNK(inode->i_mode)) {
1987                 inode->i_op = &ll_fast_symlink_inode_operations;
1988                 EXIT;
1989         } else {
1990                 inode->i_op = &ll_special_inode_operations;
1991
1992                 init_special_inode(inode, inode->i_mode,
1993                                    kdev_t_to_nr(inode->i_rdev));
1994
1995                 /* initializing backing dev info. */
1996                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1997
1998                 EXIT;
1999         }
2000 }
2001
2002 void ll_delete_inode(struct inode *inode)
2003 {
2004         struct ll_sb_info *sbi = ll_i2sbi(inode);
2005         int rc;
2006         ENTRY;
2007
2008         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2009         if (rc) {
2010                 CERROR("fid_delete() failed, rc %d\n", rc);
2011         }
2012         truncate_inode_pages(&inode->i_data, 0);
2013         clear_inode(inode);
2014
2015         EXIT;
2016 }
2017
2018 int ll_iocontrol(struct inode *inode, struct file *file,
2019                  unsigned int cmd, unsigned long arg)
2020 {
2021         struct ll_sb_info *sbi = ll_i2sbi(inode);
2022         struct ptlrpc_request *req = NULL;
2023         int rc, flags = 0;
2024         ENTRY;
2025
2026         switch(cmd) {
2027         case EXT3_IOC_GETFLAGS: {
2028                 struct mdt_body *body;
2029                 struct obd_capa *oc;
2030
2031                 oc = ll_mdscapa_get(inode);
2032                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2033                                 OBD_MD_FLFLAGS, 0, &req);
2034                 capa_put(oc);
2035                 if (rc) {
2036                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2037                         RETURN(-abs(rc));
2038                 }
2039
2040                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2041
2042                 flags = body->flags;
2043
2044                 ptlrpc_req_finished(req);
2045
2046                 RETURN(put_user(flags, (int *)arg));
2047         }
2048         case EXT3_IOC_SETFLAGS: {
2049                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2050                 struct obd_info oinfo = { { { 0 } } };
2051                 struct md_op_data *op_data;
2052
2053                 if (get_user(flags, (int *)arg))
2054                         RETURN(-EFAULT);
2055
2056                 oinfo.oi_md = lsm;
2057                 OBDO_ALLOC(oinfo.oi_oa);
2058                 if (!oinfo.oi_oa)
2059                         RETURN(-ENOMEM);
2060
2061                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2062                                              LUSTRE_OPC_ANY, NULL);
2063                 if (IS_ERR(op_data))
2064                         RETURN(PTR_ERR(op_data));
2065
2066                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2067                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2068                 rc = md_setattr(sbi->ll_md_exp, op_data,
2069                                 NULL, 0, NULL, 0, &req, NULL);
2070                 ll_finish_md_op_data(op_data);
2071                 ptlrpc_req_finished(req);
2072                 if (rc || lsm == NULL) {
2073                         OBDO_FREE(oinfo.oi_oa);
2074                         RETURN(rc);
2075                 }
2076
2077                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2078                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2079                 oinfo.oi_oa->o_flags = flags;
2080                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2081                                        OBD_MD_FLGROUP;
2082                 oinfo.oi_capa = ll_mdscapa_get(inode);
2083
2084                 obdo_from_inode(oinfo.oi_oa, inode,
2085                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2086                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2087                 capa_put(oinfo.oi_capa);
2088                 OBDO_FREE(oinfo.oi_oa);
2089                 if (rc) {
2090                         if (rc != -EPERM && rc != -EACCES)
2091                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2092                         RETURN(rc);
2093                 }
2094
2095                 inode->i_flags = ll_ext_to_inode_flags(flags |
2096                                                        MDS_BFLAG_EXT_FLAGS);
2097                 RETURN(0);
2098         }
2099         default:
2100                 RETURN(-ENOSYS);
2101         }
2102
2103         RETURN(0);
2104 }
2105
2106 int ll_flush_ctx(struct inode *inode)
2107 {
2108         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2109
2110         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2111
2112         obd_set_info_async(sbi->ll_md_exp,
2113                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2114                            0, NULL, NULL);
2115         obd_set_info_async(sbi->ll_dt_exp,
2116                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2117                            0, NULL, NULL);
2118         return 0;
2119 }
2120
2121 /* umount -f client means force down, don't save state */
2122 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2123 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2124 {
2125         struct super_block *sb = vfsmnt->mnt_sb;
2126 #else
2127 void ll_umount_begin(struct super_block *sb)
2128 {
2129 #endif
2130         struct lustre_sb_info *lsi = s2lsi(sb);
2131         struct ll_sb_info *sbi = ll_s2sbi(sb);
2132         struct obd_device *obd;
2133         struct obd_ioctl_data ioc_data = { 0 };
2134         ENTRY;
2135
2136 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2137         if (!(flags & MNT_FORCE)) {
2138                 EXIT;
2139                 return;
2140         }
2141 #endif
2142
2143         /* Tell the MGC we got umount -f */
2144         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2145
2146         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2147                sb->s_count, atomic_read(&sb->s_active));
2148
2149         obd = class_exp2obd(sbi->ll_md_exp);
2150         if (obd == NULL) {
2151                 CERROR("Invalid MDC connection handle "LPX64"\n",
2152                        sbi->ll_md_exp->exp_handle.h_cookie);
2153                 EXIT;
2154                 return;
2155         }
2156         obd->obd_force = 1;
2157         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2158                       &ioc_data, NULL);
2159
2160         obd = class_exp2obd(sbi->ll_dt_exp);
2161         if (obd == NULL) {
2162                 CERROR("Invalid LOV connection handle "LPX64"\n",
2163                        sbi->ll_dt_exp->exp_handle.h_cookie);
2164                 EXIT;
2165                 return;
2166         }
2167
2168         obd->obd_force = 1;
2169         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2170                       &ioc_data, NULL);
2171
2172         /* Really, we'd like to wait until there are no requests outstanding,
2173          * and then continue.  For now, we just invalidate the requests,
2174          * schedule, and hope.
2175          */
2176         schedule();
2177
2178         EXIT;
2179 }
2180
2181 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2182 {
2183         struct ll_sb_info *sbi = ll_s2sbi(sb);
2184         int err;
2185         __u32 read_only;
2186
2187         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2188                 read_only = *flags & MS_RDONLY;
2189                 err = obd_set_info_async(sbi->ll_md_exp,
2190                                          sizeof(KEY_READ_ONLY),
2191                                          KEY_READ_ONLY, sizeof(read_only),
2192                                          &read_only, NULL);
2193                 if (err) {
2194                         CERROR("Failed to change the read-only flag during "
2195                                "remount: %d\n", err);
2196                         return err;
2197                 }
2198
2199                 if (read_only)
2200                         sb->s_flags |= MS_RDONLY;
2201                 else
2202                         sb->s_flags &= ~MS_RDONLY;
2203         }
2204         return 0;
2205 }
2206
2207 int ll_prep_inode(struct inode **inode,
2208                   struct ptlrpc_request *req,
2209                   struct super_block *sb)
2210 {
2211         struct ll_sb_info *sbi = NULL;
2212         struct lustre_md md;
2213         int rc = 0;
2214         ENTRY;
2215
2216         LASSERT(*inode || sb);
2217         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2218         prune_deathrow(sbi, 1);
2219         memset(&md, 0, sizeof(struct lustre_md));
2220
2221         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2222                               sbi->ll_md_exp, &md);
2223         if (rc)
2224                 RETURN(rc);
2225
2226         if (*inode) {
2227                 ll_update_inode(*inode, &md);
2228         } else {
2229                 LASSERT(sb != NULL);
2230
2231                 /*
2232                  * At this point server returns to client's same fid as client
2233                  * generated for creating. So using ->fid1 is okay here.
2234                  */
2235                 LASSERT(fid_is_sane(&md.body->fid1));
2236
2237                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2238                 if (*inode == NULL || is_bad_inode(*inode)) {
2239                         if (md.lsm)
2240                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2241 #ifdef CONFIG_FS_POSIX_ACL
2242                         if (md.posix_acl) {
2243                                 posix_acl_release(md.posix_acl);
2244                                 md.posix_acl = NULL;
2245                         }
2246 #endif
2247                         rc = -ENOMEM;
2248                         CERROR("new_inode -fatal: rc %d\n", rc);
2249                         GOTO(out, rc);
2250                 }
2251         }
2252
2253         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2254                          ll_i2info(*inode)->lli_smd);
2255 out:
2256         md_free_lustre_md(sbi->ll_md_exp, &md);
2257         RETURN(rc);
2258 }
2259
2260 char *llap_origins[] = {
2261         [LLAP_ORIGIN_UNKNOWN] = "--",
2262         [LLAP_ORIGIN_READPAGE] = "rp",
2263         [LLAP_ORIGIN_READAHEAD] = "ra",
2264         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2265         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2266         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2267 };
2268
2269 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2270                                              struct list_head *list)
2271 {
2272         struct ll_async_page *llap;
2273         struct list_head *pos;
2274
2275         list_for_each(pos, list) {
2276                 if (pos == head)
2277                         return NULL;
2278                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2279                 if (llap->llap_page == NULL)
2280                         continue;
2281                 return llap;
2282         }
2283         LBUG();
2284         return NULL;
2285 }
2286
2287 int ll_obd_statfs(struct inode *inode, void *arg)
2288 {
2289         struct ll_sb_info *sbi = NULL;
2290         struct obd_export *exp;
2291         char *buf = NULL;
2292         struct obd_ioctl_data *data = NULL;
2293         __u32 type;
2294         int len = 0, rc;
2295
2296         if (!inode || !(sbi = ll_i2sbi(inode)))
2297                 GOTO(out_statfs, rc = -EINVAL);
2298
2299         rc = obd_ioctl_getdata(&buf, &len, arg);
2300         if (rc)
2301                 GOTO(out_statfs, rc);
2302
2303         data = (void*)buf;
2304         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2305             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2306                 GOTO(out_statfs, rc = -EINVAL);
2307
2308         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2309         if (type == LL_STATFS_MDC)
2310                 exp = sbi->ll_md_exp;
2311         else if (type == LL_STATFS_LOV)
2312                 exp = sbi->ll_dt_exp;
2313         else
2314                 GOTO(out_statfs, rc = -ENODEV);
2315
2316         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2317         if (rc)
2318                 GOTO(out_statfs, rc);
2319 out_statfs:
2320         if (buf)
2321                 obd_ioctl_freedata(buf, len);
2322         return rc;
2323 }
2324
2325 int ll_process_config(struct lustre_cfg *lcfg)
2326 {
2327         char *ptr;
2328         void *sb;
2329         struct lprocfs_static_vars lvars;
2330         unsigned long x;
2331         int rc = 0;
2332
2333         lprocfs_llite_init_vars(&lvars);
2334
2335         /* The instance name contains the sb: lustre-client-aacfe000 */
2336         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2337         if (!ptr || !*(++ptr))
2338                 return -EINVAL;
2339         if (sscanf(ptr, "%lx", &x) != 1)
2340                 return -EINVAL;
2341         sb = (void *)x;
2342         /* This better be a real Lustre superblock! */
2343         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2344
2345         /* Note we have not called client_common_fill_super yet, so
2346            proc fns must be able to handle that! */
2347         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2348                                       lcfg, sb);
2349         return(rc);
2350 }
2351
2352 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2353 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2354                                        struct inode *i1, struct inode *i2,
2355                                        const char *name, int namelen,
2356                                        int mode, __u32 opc, void *data)
2357 {
2358         LASSERT(i1 != NULL);
2359
2360         if (namelen > ll_i2sbi(i1)->ll_namelen)
2361                 return ERR_PTR(-ENAMETOOLONG);
2362
2363         if (op_data == NULL)
2364                 OBD_ALLOC_PTR(op_data);
2365
2366         if (op_data == NULL)
2367                 return ERR_PTR(-ENOMEM);
2368
2369         ll_i2gids(op_data->op_suppgids, i1, i2);
2370         op_data->op_fid1 = *ll_inode2fid(i1);
2371         op_data->op_capa1 = ll_mdscapa_get(i1);
2372
2373         if (i2) {
2374                 op_data->op_fid2 = *ll_inode2fid(i2);
2375                 op_data->op_capa2 = ll_mdscapa_get(i2);
2376         } else {
2377                 fid_zero(&op_data->op_fid2);
2378                 op_data->op_capa2 = NULL;
2379         }
2380
2381         op_data->op_name = name;
2382         op_data->op_namelen = namelen;
2383         op_data->op_mode = mode;
2384         op_data->op_mod_time = cfs_time_current_sec();
2385         op_data->op_fsuid = current->fsuid;
2386         op_data->op_fsgid = current->fsgid;
2387         op_data->op_cap = current->cap_effective;
2388         op_data->op_bias = MDS_CHECK_SPLIT;
2389         op_data->op_opc = opc;
2390         op_data->op_mds = 0;
2391         op_data->op_data = data;
2392
2393         return op_data;
2394 }
2395
2396 void ll_finish_md_op_data(struct md_op_data *op_data)
2397 {
2398         capa_put(op_data->op_capa1);
2399         capa_put(op_data->op_capa2);
2400         OBD_FREE_PTR(op_data);
2401 }