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 #ifdef HAVE_BGL_SUPPORT
 158                 sbi->ll_async_page_max = pages / 4;
 159 #else
 160                 sbi->ll_async_page_max = pages / 2;
 161 #endif
 162         } else {
 163                 sbi->ll_async_page_max = (pages / 4) * 3;
 164         }
 165         lcounter_init(&sbi->ll_async_page_count);
 166         spin_lock_init(&sbi->ll_async_page_reblnc_lock);
 167         sbi->ll_async_page_sample_max = 64 * num_online_cpus();
 168         sbi->ll_async_page_reblnc_count = 0;
 169         sbi->ll_async_page_clock_hand = 0;
 170         if (ll_pglist_init(sbi))
 171                 GOTO(out, 0);
 172
 173         sbi->ll_ra_info.ra_max_pages = min(pages / 32,
 174                                            SBI_DEFAULT_READAHEAD_MAX);
 175         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 176                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 177         sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
 178         sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
 179         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 180         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 181
 182         ll_generate_random_uuid(uuid);
 183         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 184         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 185
 186         spin_lock(&ll_sb_lock);
 187         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 188         spin_unlock(&ll_sb_lock);
 189
 190 #ifdef ENABLE_LLITE_CHECKSUM
 191         sbi->ll_flags |= LL_SBI_CHECKSUM;
 192 #endif
 193
 194 #ifdef HAVE_LRU_RESIZE_SUPPORT
 195         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 196 #endif
 197
 198 #ifdef HAVE_EXPORT___IGET
 199         INIT_LIST_HEAD(&sbi->ll_deathrow);
 200         spin_lock_init(&sbi->ll_deathrow_lock);
 201 #endif
 202         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 203                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 204                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 205         }
 206
 207         /* metadata statahead is enabled by default */
 208         sbi->ll_sa_max = LL_SA_RPC_DEF;
 209
 210         RETURN(sbi);
 211
 212 out:
 213         if (sbi->ll_async_page_sample)
 214                 OBD_FREE(sbi->ll_async_page_sample,
 215                          sizeof(long) * num_possible_cpus());
 216         ll_pglist_fini(sbi);
 217         OBD_FREE(sbi, sizeof(*sbi));
 218         RETURN(NULL);
 219 }
 220
 221 void ll_free_sbi(struct super_block *sb)
 222 {
 223         struct ll_sb_info *sbi = ll_s2sbi(sb);
 224         ENTRY;
 225
 226         if (sbi != NULL) {
 227                 ll_pglist_fini(sbi);
 228                 spin_lock(&ll_sb_lock);
 229                 list_del(&sbi->ll_list);
 230                 spin_unlock(&ll_sb_lock);
 231                 lcounter_destroy(&sbi->ll_async_page_count);
 232                 OBD_FREE(sbi->ll_async_page_sample,
 233                          sizeof(long) * num_possible_cpus());
 234                 OBD_FREE(sbi, sizeof(*sbi));
 235         }
 236         EXIT;
 237 }
 238
 239 static struct dentry_operations ll_d_root_ops = {
 240 #ifdef DCACHE_LUSTRE_INVALID
 241         .d_compare = ll_dcompare,
 242 #endif
 243 };
 244
 245 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 246  * us to make MDS RPCs with large enough reply buffers to hold the
 247  * maximum-sized (= maximum striped) EA and cookie without having to
 248  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 249 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 250 {
 251         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 252         __u32 valsize = sizeof(struct lov_desc);
 253         int rc, easize, def_easize, cookiesize;
 254         struct lov_desc desc;
 255         __u32 stripes;
 256         ENTRY;
 257
 258         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
 259                           &valsize, &desc, NULL);
 260         if (rc)
 261                 RETURN(rc);
 262
 263         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 264         lsm.lsm_stripe_count = stripes;
 265         easize = obd_size_diskmd(dt_exp, &lsm);
 266
 267         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 268         def_easize = obd_size_diskmd(dt_exp, &lsm);
 269
 270         cookiesize = stripes * sizeof(struct llog_cookie);
 271
 272         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 273                easize, cookiesize);
 274
 275         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 276         RETURN(rc);
 277 }
 278
 279 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 280 {
 281         struct inode *root = 0;
 282         struct ll_sb_info *sbi = ll_s2sbi(sb);
 283         struct obd_device *obd;
 284         struct lu_fid rootfid;
 285         struct obd_capa *oc = NULL;
 286         struct obd_statfs osfs;
 287         struct ptlrpc_request *request = NULL;
 288         struct lustre_handle dt_conn = {0, };
 289         struct lustre_handle md_conn = {0, };
 290         struct obd_connect_data *data = NULL;
 291         struct lustre_md lmd;
 292         obd_valid valid;
 293         int size, err, checksum;
 294         ENTRY;
 295
 296         obd = class_name2obd(md);
 297         if (!obd) {
 298                 CERROR("MD %s: not setup or attached\n", md);
 299                 RETURN(-EINVAL);
 300         }
 301
 302         OBD_ALLOC_PTR(data);
 303         if (data == NULL)
 304                 RETURN(-ENOMEM);
 305
 306         if (proc_lustre_fs_root) {
 307                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 308                                                   dt, md);
 309                 if (err < 0)
 310                         CERROR("could not register mount in /proc/fs/lustre\n");
 311         }
 312
 313         /* indicate the features supported by this client */
 314         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 315                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 316                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 317                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 318                                   OBD_CONNECT_FID      | OBD_CONNECT_AT;
 319
 320 #ifdef HAVE_LRU_RESIZE_SUPPORT
 321         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 322                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 323 #endif
 324 #ifdef CONFIG_FS_POSIX_ACL
 325         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 326 #endif
 327         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 328         data->ocd_version = LUSTRE_VERSION_CODE;
 329
 330         if (sb->s_flags & MS_RDONLY)
 331                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 332         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 333                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 334
 335 #ifdef HAVE_MS_FLOCK_LOCK
 336         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 337         sb->s_flags |= MS_FLOCK_LOCK;
 338 #endif
 339
 340         if (sbi->ll_flags & LL_SBI_FLOCK)
 341                 sbi->ll_fop = &ll_file_operations_flock;
 342         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 343                 sbi->ll_fop = &ll_file_operations;
 344         else
 345                 sbi->ll_fop = &ll_file_operations_noflock;
 346
 347         /* real client */
 348         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 349         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 350                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 351                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 352         } else {
 353                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 354                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 355         }
 356
 357         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 358         if (err == -EBUSY) {
 359                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 360                                    "recovery, of which this client is not a "
 361                                    "part. Please wait for recovery to complete,"
 362                                    " abort, or time out.\n", md);
 363                 GOTO(out, err);
 364         } else if (err) {
 365                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 366                 GOTO(out, err);
 367         }
 368         sbi->ll_md_exp = class_conn2export(&md_conn);
 369
 370         err = obd_fid_init(sbi->ll_md_exp);
 371         if (err) {
 372                 CERROR("Can't init metadata layer FID infrastructure, "
 373                        "rc %d\n", err);
 374                 GOTO(out_md, err);
 375         }
 376
 377         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 378         if (err)
 379                 GOTO(out_md_fid, err);
 380
 381         size = sizeof(*data);
 382         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 383                            KEY_CONN_DATA,  &size, data, NULL);
 384         if (err) {
 385                 CERROR("Get connect data failed: %d \n", err);
 386                 GOTO(out_md, err);
 387         }
 388
 389         LASSERT(osfs.os_bsize);
 390         sb->s_blocksize = osfs.os_bsize;
 391         sb->s_blocksize_bits = log2(osfs.os_bsize);
 392         sb->s_magic = LL_SUPER_MAGIC;
 393
 394         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 395          *         retval = in_file->f_op->sendfile(...);
 396          *         if (*ppos > max)
 397          *                 retval = -EOVERFLOW;
 398          *
 399          * it will check if *ppos is greater than max. However, max equals to
 400          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 401          * has been defined as a signed long long ineger in linux kernel. */
 402 #if BITS_PER_LONG == 64
 403         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 404 #else
 405         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 406 #endif
 407         sbi->ll_namelen = osfs.os_namelen;
 408         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 409
 410         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 411             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 412                 LCONSOLE_INFO("Disabling user_xattr feature because "
 413                               "it is not supported on the server\n");
 414                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 415         }
 416
 417         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 418 #ifdef MS_POSIXACL
 419                 sb->s_flags |= MS_POSIXACL;
 420 #endif
 421                 sbi->ll_flags |= LL_SBI_ACL;
 422         } else {
 423                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 424 #ifdef MS_POSIXACL
 425                 sb->s_flags &= ~MS_POSIXACL;
 426 #endif
 427                 sbi->ll_flags &= ~LL_SBI_ACL;
 428         }
 429
 430         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 431                 sbi->ll_flags |= LL_SBI_JOIN;
 432
 433         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 434                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 435                         /* sometimes local client claims to be remote, but mdt
 436                          * will disagree when client gss not applied. */
 437                         LCONSOLE_INFO("client claims to be remote, but server "
 438                                       "rejected, forced to be local.\n");
 439                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 440                 }
 441         } else {
 442                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 443                         /* with gss applied, remote client can not claim to be
 444                          * local, so mdt maybe force client to be remote. */
 445                         LCONSOLE_INFO("client claims to be local, but server "
 446                                       "rejected, forced to be remote.\n");
 447                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 448                 }
 449         }
 450
 451         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 452                 LCONSOLE_INFO("client enabled MDS capability!\n");
 453                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 454         }
 455
 456         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 457                 LCONSOLE_INFO("client enabled OSS capability!\n");
 458                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 459         }
 460
 461         sbi->ll_sdev_orig = sb->s_dev;
 462 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,0))
 463         /* We set sb->s_dev equal on all lustre clients in order to support
 464          * NFS export clustering.  NFSD requires that the FSID be the same
 465          * on all clients. */
 466         /* s_dev is also used in lt_compare() to compare two fs, but that is
 467          * only a node-local comparison. */
 468
 469         /* XXX: this will not work with LMV */
 470         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 471                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 472 #endif
 473
 474         obd = class_name2obd(dt);
 475         if (!obd) {
 476                 CERROR("DT %s: not setup or attached\n", dt);
 477                 GOTO(out_md_fid, err = -ENODEV);
 478         }
 479
 480         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 481                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 482                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 483                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 484                                   OBD_CONNECT_AT;
 485         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 486                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 487
 488         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 489                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 490                  * disabled by default, because it can still be enabled on the
 491                  * fly via /proc. As a consequence, we still need to come to an
 492                  * agreement on the supported algorithms at connect time */
 493                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 494
 495                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 496                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 497                 else
 498                         /* send the list of supported checksum types */
 499                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 500         }
 501
 502 #ifdef HAVE_LRU_RESIZE_SUPPORT
 503         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 504 #endif
 505         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 506                "ocd_grant: %d\n", data->ocd_connect_flags,
 507                data->ocd_version, data->ocd_grant);
 508
 509         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 510         obd->obd_upcall.onu_upcall = ll_ocd_update;
 511         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 512
 513         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 514         if (err == -EBUSY) {
 515                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 516                                    "recovery, of which this client is not a "
 517                                    "part.  Please wait for recovery to "
 518                                    "complete, abort, or time out.\n", dt);
 519                 GOTO(out_md_fid, err);
 520         } else if (err) {
 521                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 522                 GOTO(out_md_fid, err);
 523         }
 524
 525         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 526
 527         err = obd_fid_init(sbi->ll_dt_exp);
 528         if (err) {
 529                 CERROR("Can't init data layer FID infrastructure, "
 530                        "rc %d\n", err);
 531                 GOTO(out_dt, err);
 532         }
 533
 534         spin_lock(&sbi->ll_lco.lco_lock);
 535         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 536         spin_unlock(&sbi->ll_lco.lco_lock);
 537
 538         err = obd_register_page_removal_cb(sbi->ll_dt_exp,
 539                                            ll_page_removal_cb,
 540                                            ll_pin_extent_cb);
 541         if (err) {
 542                 CERROR("cannot register page removal callback: rc = %d\n",err);
 543                 GOTO(out_dt, err);
 544         }
 545         err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
 546                                           ll_extent_lock_cancel_cb);
 547         if (err) {
 548                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
 549                 GOTO(out_page_rm_cb, err);
 550         }
 551
 552         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 553         if (err) {
 554                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 555                 GOTO(out_lock_cn_cb, err);
 556         }
 557
 558         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 559                                   0, NULL, NULL, NULL, 0, NULL);
 560         if (err < 0) {
 561                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 562                                    "filesystem. There must be at least one "
 563                                    "active OST for a client to start.\n");
 564                 GOTO(out_lock_cn_cb, err);
 565         }
 566
 567         if (!ll_async_page_slab) {
 568                 ll_async_page_slab_size =
 569                         size_round(sizeof(struct ll_async_page)) + err;
 570                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 571                                                           ll_async_page_slab_size,
 572                                                           0, 0);
 573                 if (!ll_async_page_slab)
 574                         GOTO(out_lock_cn_cb, err = -ENOMEM);
 575         }
 576
 577         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 578         if (err) {
 579                 CERROR("cannot mds_connect: rc = %d\n", err);
 580                 GOTO(out_lock_cn_cb, err);
 581         }
 582         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 583         sbi->ll_root_fid = rootfid;
 584
 585         sb->s_op = &lustre_super_operations;
 586         sb->s_export_op = &lustre_export_operations;
 587
 588         /* make root inode
 589          * XXX: move this to after cbd setup? */
 590         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 591         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 592                 valid |= OBD_MD_FLRMTPERM;
 593         else if (sbi->ll_flags & LL_SBI_ACL)
 594                 valid |= OBD_MD_FLACL;
 595
 596         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 597         if (oc)
 598                 free_capa(oc);
 599         if (err) {
 600                 CERROR("md_getattr failed for root: rc = %d\n", err);
 601                 GOTO(out_lock_cn_cb, err);
 602         }
 603         memset(&lmd, 0, sizeof(lmd));
 604         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 605                                sbi->ll_md_exp, &lmd);
 606         if (err) {
 607                 CERROR("failed to understand root inode md: rc = %d\n", err);
 608                 ptlrpc_req_finished (request);
 609                 GOTO(out_lock_cn_cb, err);
 610         }
 611
 612         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 613         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 614         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 615         ptlrpc_req_finished(request);
 616
 617         if (root == NULL || is_bad_inode(root)) {
 618                 if (lmd.lsm)
 619                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 620 #ifdef CONFIG_FS_POSIX_ACL
 621                 if (lmd.posix_acl) {
 622                         posix_acl_release(lmd.posix_acl);
 623                         lmd.posix_acl = NULL;
 624                 }
 625 #endif
 626                 CERROR("lustre_lite: bad iget4 for root\n");
 627                 GOTO(out_root, err = -EBADF);
 628         }
 629
 630         err = ll_close_thread_start(&sbi->ll_lcq);
 631         if (err) {
 632                 CERROR("cannot start close thread: rc %d\n", err);
 633                 GOTO(out_root, err);
 634         }
 635
 636 #ifdef CONFIG_FS_POSIX_ACL
 637         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 638                 rct_init(&sbi->ll_rct);
 639                 et_init(&sbi->ll_et);
 640         }
 641 #endif
 642
 643         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 644         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 645                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 646                                  NULL);
 647
 648         sb->s_root = d_alloc_root(root);
 649         if (data != NULL)
 650                 OBD_FREE(data, sizeof(*data));
 651         sb->s_root->d_op = &ll_d_root_ops;
 652         RETURN(err);
 653 out_root:
 654         if (root)
 655                 iput(root);
 656 out_lock_cn_cb:
 657         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
 658                                       ll_extent_lock_cancel_cb);
 659 out_page_rm_cb:
 660         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 661                                        ll_page_removal_cb);
 662         obd_fid_fini(sbi->ll_dt_exp);
 663 out_dt:
 664         obd_disconnect(sbi->ll_dt_exp);
 665         sbi->ll_dt_exp = NULL;
 666 out_md_fid:
 667         obd_fid_fini(sbi->ll_md_exp);
 668 out_md:
 669         obd_disconnect(sbi->ll_md_exp);
 670         sbi->ll_md_exp = NULL;
 671 out:
 672         if (data != NULL)
 673                 OBD_FREE_PTR(data);
 674         lprocfs_unregister_mountpoint(sbi);
 675         return err;
 676 }
 677
 678 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 679 {
 680         int size, rc;
 681
 682         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 683         size = sizeof(int);
 684         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 685                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 686         if (rc)
 687                 CERROR("Get max mdsize error rc %d \n", rc);
 688
 689         RETURN(rc);
 690 }
 691
 692 void ll_dump_inode(struct inode *inode)
 693 {
 694         struct list_head *tmp;
 695         int dentry_count = 0;
 696
 697         LASSERT(inode != NULL);
 698
 699         list_for_each(tmp, &inode->i_dentry)
 700                 dentry_count++;
 701
 702         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 703                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 704                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 705 }
 706
 707 void lustre_dump_dentry(struct dentry *dentry, int recur)
 708 {
 709         struct list_head *tmp;
 710         int subdirs = 0;
 711
 712         LASSERT(dentry != NULL);
 713
 714         list_for_each(tmp, &dentry->d_subdirs)
 715                 subdirs++;
 716
 717         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 718                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 719                dentry->d_name.len, dentry->d_name.name,
 720                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 721                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 722                dentry->d_flags, dentry->d_fsdata, subdirs);
 723         if (dentry->d_inode != NULL)
 724                 ll_dump_inode(dentry->d_inode);
 725
 726         if (recur == 0)
 727                 return;
 728
 729         list_for_each(tmp, &dentry->d_subdirs) {
 730                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 731                 lustre_dump_dentry(d, recur - 1);
 732         }
 733 }
 734
 735 #ifdef HAVE_EXPORT___IGET
 736 static void prune_dir_dentries(struct inode *inode)
 737 {
 738         struct dentry *dentry, *prev = NULL;
 739
 740         /* due to lustre specific logic, a directory
 741          * can have few dentries - a bug from VFS POV */
 742 restart:
 743         spin_lock(&dcache_lock);
 744         if (!list_empty(&inode->i_dentry)) {
 745                 dentry = list_entry(inode->i_dentry.prev,
 746                                     struct dentry, d_alias);
 747                 /* in order to prevent infinite loops we
 748                  * break if previous dentry is busy */
 749                 if (dentry != prev) {
 750                         prev = dentry;
 751                         dget_locked(dentry);
 752                         spin_unlock(&dcache_lock);
 753
 754                         /* try to kill all child dentries */
 755                         lock_dentry(dentry);
 756                         shrink_dcache_parent(dentry);
 757                         unlock_dentry(dentry);
 758                         dput(dentry);
 759
 760                         /* now try to get rid of current dentry */
 761                         d_prune_aliases(inode);
 762                         goto restart;
 763                 }
 764         }
 765         spin_unlock(&dcache_lock);
 766 }
 767
 768 static void prune_deathrow_one(struct ll_inode_info *lli)
 769 {
 770         struct inode *inode = ll_info2i(lli);
 771
 772         /* first, try to drop any dentries - they hold a ref on the inode */
 773         if (S_ISDIR(inode->i_mode))
 774                 prune_dir_dentries(inode);
 775         else
 776                 d_prune_aliases(inode);
 777
 778
 779         /* if somebody still uses it, leave it */
 780         LASSERT(atomic_read(&inode->i_count) > 0);
 781         if (atomic_read(&inode->i_count) > 1)
 782                 goto out;
 783
 784         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 785                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 786
 787         /* seems nobody uses it anymore */
 788         inode->i_nlink = 0;
 789
 790 out:
 791         iput(inode);
 792         return;
 793 }
 794
 795 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 796 {
 797         struct ll_inode_info *lli;
 798         int empty;
 799
 800         do {
 801                 if (need_resched() && try)
 802                         break;
 803
 804                 if (try) {
 805                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 806                                 break;
 807                 } else {
 808                         spin_lock(&sbi->ll_deathrow_lock);
 809                 }
 810
 811                 empty = 1;
 812                 lli = NULL;
 813                 if (!list_empty(&sbi->ll_deathrow)) {
 814                         lli = list_entry(sbi->ll_deathrow.next,
 815                                          struct ll_inode_info,
 816                                          lli_dead_list);
 817                         list_del_init(&lli->lli_dead_list);
 818                         if (!list_empty(&sbi->ll_deathrow))
 819                                 empty = 0;
 820                 }
 821                 spin_unlock(&sbi->ll_deathrow_lock);
 822
 823                 if (lli)
 824                         prune_deathrow_one(lli);
 825
 826         } while (empty == 0);
 827 }
 828 #else /* !HAVE_EXPORT___IGET */
 829 #define prune_deathrow(sbi, try) do {} while (0)
 830 #endif /* HAVE_EXPORT___IGET */
 831
 832 void client_common_put_super(struct super_block *sb)
 833 {
 834         struct ll_sb_info *sbi = ll_s2sbi(sb);
 835         ENTRY;
 836
 837 #ifdef CONFIG_FS_POSIX_ACL
 838         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 839                 et_fini(&sbi->ll_et);
 840                 rct_fini(&sbi->ll_rct);
 841         }
 842 #endif
 843
 844         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 845
 846         ll_close_thread_shutdown(sbi->ll_lcq);
 847
 848         /* destroy inodes in deathrow */
 849         prune_deathrow(sbi, 0);
 850
 851         list_del(&sbi->ll_conn_chain);
 852
 853         obd_fid_fini(sbi->ll_dt_exp);
 854         obd_disconnect(sbi->ll_dt_exp);
 855         sbi->ll_dt_exp = NULL;
 856
 857         lprocfs_unregister_mountpoint(sbi);
 858
 859         obd_fid_fini(sbi->ll_md_exp);
 860         obd_disconnect(sbi->ll_md_exp);
 861         sbi->ll_md_exp = NULL;
 862
 863         EXIT;
 864 }
 865
 866 void ll_kill_super(struct super_block *sb)
 867 {
 868         struct ll_sb_info *sbi;
 869
 870         ENTRY;
 871
 872         /* not init sb ?*/
 873         if (!(sb->s_flags & MS_ACTIVE))
 874                 return;
 875
 876         sbi = ll_s2sbi(sb);
 877         /* we need restore s_dev from changed for clustred NFS before put_super
 878          * because new kernels have cached s_dev and change sb->s_dev in
 879          * put_super not affected real removing devices */
 880         if (sbi)
 881                 sb->s_dev = sbi->ll_sdev_orig;
 882         EXIT;
 883 }
 884
 885 char *ll_read_opt(const char *opt, char *data)
 886 {
 887         char *value;
 888         char *retval;
 889         ENTRY;
 890
 891         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 892         if (strncmp(opt, data, strlen(opt)))
 893                 RETURN(NULL);
 894         if ((value = strchr(data, '=')) == NULL)
 895                 RETURN(NULL);
 896
 897         value++;
 898         OBD_ALLOC(retval, strlen(value) + 1);
 899         if (!retval) {
 900                 CERROR("out of memory!\n");
 901                 RETURN(NULL);
 902         }
 903
 904         memcpy(retval, value, strlen(value)+1);
 905         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 906         RETURN(retval);
 907 }
 908
 909 static inline int ll_set_opt(const char *opt, char *data, int fl)
 910 {
 911         if (strncmp(opt, data, strlen(opt)) != 0)
 912                 return(0);
 913         else
 914                 return(fl);
 915 }
 916
 917 /* non-client-specific mount options are parsed in lmd_parse */
 918 static int ll_options(char *options, int *flags)
 919 {
 920         int tmp;
 921         char *s1 = options, *s2;
 922         ENTRY;
 923
 924         if (!options)
 925                 RETURN(0);
 926
 927         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 928
 929         while (*s1) {
 930                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 931                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 932                 if (tmp) {
 933                         *flags |= tmp;
 934                         goto next;
 935                 }
 936                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 937                 if (tmp) {
 938                         *flags |= tmp;
 939                         goto next;
 940                 }
 941                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 942                 if (tmp) {
 943                         *flags |= tmp;
 944                         goto next;
 945                 }
 946                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 947                 if (tmp) {
 948                         *flags &= ~tmp;
 949                         goto next;
 950                 }
 951                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 952                 if (tmp) {
 953                         *flags |= tmp;
 954                         goto next;
 955                 }
 956                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 957                 if (tmp) {
 958                         *flags &= ~tmp;
 959                         goto next;
 960                 }
 961                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 962                 if (tmp) {
 963                         /* Ignore deprecated mount option.  The client will
 964                          * always try to mount with ACL support, whether this
 965                          * is used depends on whether server supports it. */
 966                         goto next;
 967                 }
 968                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 969                 if (tmp) {
 970                         goto next;
 971                 }
 972                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 973                 if (tmp) {
 974                         *flags |= tmp;
 975                         goto next;
 976                 }
 977
 978                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 979                 if (tmp) {
 980                         *flags |= tmp;
 981                         goto next;
 982                 }
 983                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 984                 if (tmp) {
 985                         *flags &= ~tmp;
 986                         goto next;
 987                 }
 988                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 989                 if (tmp) {
 990                         *flags |= tmp;
 991                         goto next;
 992                 }
 993                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 994                 if (tmp) {
 995                         *flags &= ~tmp;
 996                         goto next;
 997                 }
 998
 999                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1000                                    s1);
1001                 RETURN(-EINVAL);
1002
1003 next:
1004                 /* Find next opt */
1005                 s2 = strchr(s1, ',');
1006                 if (s2 == NULL)
1007                         break;
1008                 s1 = s2 + 1;
1009         }
1010         RETURN(0);
1011 }
1012
1013 void ll_lli_init(struct ll_inode_info *lli)
1014 {
1015         lli->lli_inode_magic = LLI_INODE_MAGIC;
1016         sema_init(&lli->lli_size_sem, 1);
1017         sema_init(&lli->lli_write_sem, 1);
1018         lli->lli_flags = 0;
1019         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1020         spin_lock_init(&lli->lli_lock);
1021         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1022         INIT_LIST_HEAD(&lli->lli_close_list);
1023         lli->lli_inode_magic = LLI_INODE_MAGIC;
1024         sema_init(&lli->lli_och_sem, 1);
1025         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1026         lli->lli_mds_exec_och = NULL;
1027         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1028         lli->lli_open_fd_exec_count = 0;
1029         INIT_LIST_HEAD(&lli->lli_dead_list);
1030         lli->lli_remote_perms = NULL;
1031         lli->lli_rmtperm_utime = 0;
1032         sema_init(&lli->lli_rmtperm_sem, 1);
1033         INIT_LIST_HEAD(&lli->lli_oss_capas);
1034 }
1035
1036 int ll_fill_super(struct super_block *sb)
1037 {
1038         struct lustre_profile *lprof;
1039         struct lustre_sb_info *lsi = s2lsi(sb);
1040         struct ll_sb_info *sbi;
1041         char  *dt = NULL, *md = NULL;
1042         char  *profilenm = get_profile_name(sb);
1043         struct config_llog_instance cfg = {0, };
1044         char   ll_instance[sizeof(sb) * 2 + 1];
1045         int    err;
1046         ENTRY;
1047
1048         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1049
1050         cfs_module_get();
1051
1052         /* client additional sb info */
1053         lsi->lsi_llsbi = sbi = ll_init_sbi();
1054         if (!sbi) {
1055                 cfs_module_put();
1056                 RETURN(-ENOMEM);
1057         }
1058
1059         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1060         if (err)
1061                 GOTO(out_free, err);
1062
1063         /* Generate a string unique to this super, in case some joker tries
1064            to mount the same fs at two mount points.
1065            Use the address of the super itself.*/
1066         sprintf(ll_instance, "%p", sb);
1067         cfg.cfg_instance = ll_instance;
1068         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1069
1070         /* set up client obds */
1071         err = lustre_process_log(sb, profilenm, &cfg);
1072         if (err < 0) {
1073                 CERROR("Unable to process log: %d\n", err);
1074                 GOTO(out_free, err);
1075         }
1076
1077         lprof = class_get_profile(profilenm);
1078         if (lprof == NULL) {
1079                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1080                                    " read from the MGS.  Does that filesystem "
1081                                    "exist?\n", profilenm);
1082                 GOTO(out_free, err = -EINVAL);
1083         }
1084         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1085                lprof->lp_md, lprof->lp_dt);
1086
1087         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1088                   strlen(ll_instance) + 2);
1089         if (!dt)
1090                 GOTO(out_free, err = -ENOMEM);
1091         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1092
1093         OBD_ALLOC(md, strlen(lprof->lp_md) +
1094                   strlen(ll_instance) + 2);
1095         if (!md)
1096                 GOTO(out_free, err = -ENOMEM);
1097         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1098
1099         /* connections, registrations, sb setup */
1100         err = client_common_fill_super(sb, md, dt);
1101
1102 out_free:
1103         if (md)
1104                 OBD_FREE(md, strlen(md) + 1);
1105         if (dt)
1106                 OBD_FREE(dt, strlen(dt) + 1);
1107         if (err)
1108                 ll_put_super(sb);
1109         else
1110                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1111
1112         RETURN(err);
1113 } /* ll_fill_super */
1114
1115
1116 void ll_put_super(struct super_block *sb)
1117 {
1118         struct config_llog_instance cfg;
1119         char   ll_instance[sizeof(sb) * 2 + 1];
1120         struct obd_device *obd;
1121         struct lustre_sb_info *lsi = s2lsi(sb);
1122         struct ll_sb_info *sbi = ll_s2sbi(sb);
1123         char *profilenm = get_profile_name(sb);
1124         int force = 1, next;
1125         ENTRY;
1126
1127         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1128
1129         ll_print_capa_stat(sbi);
1130
1131         sprintf(ll_instance, "%p", sb);
1132         cfg.cfg_instance = ll_instance;
1133         lustre_end_log(sb, NULL, &cfg);
1134
1135         if (sbi->ll_md_exp) {
1136                 obd = class_exp2obd(sbi->ll_md_exp);
1137                 if (obd)
1138                         force = obd->obd_force;
1139         }
1140
1141         /* We need to set force before the lov_disconnect in
1142            lustre_common_put_super, since l_d cleans up osc's as well. */
1143         if (force) {
1144                 next = 0;
1145                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1146                                                      &next)) != NULL) {
1147                         obd->obd_force = force;
1148                 }
1149         }
1150
1151         if (sbi->ll_lcq) {
1152                 /* Only if client_common_fill_super succeeded */
1153                 client_common_put_super(sb);
1154         }
1155         next = 0;
1156         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1157                 class_manual_cleanup(obd);
1158         }
1159
1160         if (profilenm)
1161                 class_del_profile(profilenm);
1162
1163         ll_free_sbi(sb);
1164         lsi->lsi_llsbi = NULL;
1165
1166         lustre_common_put_super(sb);
1167
1168         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1169
1170         cfs_module_put();
1171
1172         EXIT;
1173 } /* client_put_super */
1174
1175 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1176
1177 #if defined(HAVE_CACHE_RETURN_INT)
1178 static int
1179 #else
1180 static void
1181 #endif
1182 ll_shrink_cache(int priority, unsigned int gfp_mask)
1183 {
1184         struct ll_sb_info *sbi;
1185         int count = 0;
1186
1187         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1188                 count += llap_shrink_cache(sbi, priority);
1189
1190 #if defined(HAVE_CACHE_RETURN_INT)
1191         return count;
1192 #endif
1193 }
1194
1195 struct cache_definition ll_cache_definition = {
1196         .name = "llap_cache",
1197         .shrink = ll_shrink_cache
1198 };
1199 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1200
1201 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1202 {
1203         struct inode *inode = NULL;
1204         /* NOTE: we depend on atomic igrab() -bzzz */
1205         lock_res_and_lock(lock);
1206         if (lock->l_ast_data) {
1207                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1208                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1209                         inode = igrab(lock->l_ast_data);
1210                 } else {
1211                         inode = lock->l_ast_data;
1212                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1213                                                 D_INFO : D_WARNING,
1214                                         lock, __FILE__, __func__, __LINE__,
1215                                         "l_ast_data %p is bogus: magic %08x",
1216                                         lock->l_ast_data, lli->lli_inode_magic);
1217                         inode = NULL;
1218                 }
1219         }
1220         unlock_res_and_lock(lock);
1221         return inode;
1222 }
1223
1224 static int null_if_equal(struct ldlm_lock *lock, void *data)
1225 {
1226         if (data == lock->l_ast_data) {
1227                 lock->l_ast_data = NULL;
1228
1229                 if (lock->l_req_mode != lock->l_granted_mode)
1230                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1231         }
1232
1233         return LDLM_ITER_CONTINUE;
1234 }
1235
1236 void ll_clear_inode(struct inode *inode)
1237 {
1238         struct ll_inode_info *lli = ll_i2info(inode);
1239         struct ll_sb_info *sbi = ll_i2sbi(inode);
1240         ENTRY;
1241
1242         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1243                inode->i_generation, inode);
1244
1245         if (S_ISDIR(inode->i_mode)) {
1246                 /* these should have been cleared in ll_file_release */
1247                 LASSERT(lli->lli_sai == NULL);
1248                 LASSERT(lli->lli_opendir_key == NULL);
1249                 LASSERT(lli->lli_opendir_pid == 0);
1250         }
1251
1252         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1253         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1254                          null_if_equal, inode);
1255
1256         LASSERT(!lli->lli_open_fd_write_count);
1257         LASSERT(!lli->lli_open_fd_read_count);
1258         LASSERT(!lli->lli_open_fd_exec_count);
1259
1260         if (lli->lli_mds_write_och)
1261                 ll_md_real_close(inode, FMODE_WRITE);
1262         if (lli->lli_mds_exec_och)
1263                 ll_md_real_close(inode, FMODE_EXEC);
1264         if (lli->lli_mds_read_och)
1265                 ll_md_real_close(inode, FMODE_READ);
1266
1267         if (lli->lli_smd) {
1268                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1269                                   null_if_equal, inode);
1270
1271                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1272                 lli->lli_smd = NULL;
1273         }
1274
1275         if (lli->lli_symlink_name) {
1276                 OBD_FREE(lli->lli_symlink_name,
1277                          strlen(lli->lli_symlink_name) + 1);
1278                 lli->lli_symlink_name = NULL;
1279         }
1280
1281         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1282                 LASSERT(lli->lli_posix_acl == NULL);
1283                 if (lli->lli_remote_perms) {
1284                         free_rmtperm_hash(lli->lli_remote_perms);
1285                         lli->lli_remote_perms = NULL;
1286                 }
1287         }
1288 #ifdef CONFIG_FS_POSIX_ACL
1289         else if (lli->lli_posix_acl) {
1290                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1291                 LASSERT(lli->lli_remote_perms == NULL);
1292                 posix_acl_release(lli->lli_posix_acl);
1293                 lli->lli_posix_acl = NULL;
1294         }
1295 #endif
1296         lli->lli_inode_magic = LLI_INODE_DEAD;
1297
1298 #ifdef HAVE_EXPORT___IGET
1299         spin_lock(&sbi->ll_deathrow_lock);
1300         list_del_init(&lli->lli_dead_list);
1301         spin_unlock(&sbi->ll_deathrow_lock);
1302 #endif
1303         ll_clear_inode_capas(inode);
1304
1305         EXIT;
1306 }
1307
1308 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1309                   struct md_open_data **mod)
1310 {
1311         struct lustre_md md;
1312         struct ll_sb_info *sbi = ll_i2sbi(inode);
1313         struct ptlrpc_request *request = NULL;
1314         int rc;
1315         ENTRY;
1316
1317         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1318                                      LUSTRE_OPC_ANY, NULL);
1319         if (IS_ERR(op_data))
1320                 RETURN(PTR_ERR(op_data));
1321
1322         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1323                         &request, mod);
1324         if (rc) {
1325                 ptlrpc_req_finished(request);
1326                 if (rc == -ENOENT) {
1327                         inode->i_nlink = 0;
1328                         /* Unlinked special device node? Or just a race?
1329                          * Pretend we done everything. */
1330                         if (!S_ISREG(inode->i_mode) &&
1331                             !S_ISDIR(inode->i_mode))
1332                                 rc = inode_setattr(inode, &op_data->op_attr);
1333                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1334                         CERROR("md_setattr fails: rc = %d\n", rc);
1335                 }
1336                 RETURN(rc);
1337         }
1338
1339         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1340                               sbi->ll_md_exp, &md);
1341         if (rc) {
1342                 ptlrpc_req_finished(request);
1343                 RETURN(rc);
1344         }
1345
1346         /* We call inode_setattr to adjust timestamps.
1347          * If there is at least some data in file, we cleared ATTR_SIZE
1348          * above to avoid invoking vmtruncate, otherwise it is important
1349          * to call vmtruncate in inode_setattr to update inode->i_size
1350          * (bug 6196) */
1351         rc = inode_setattr(inode, &op_data->op_attr);
1352
1353         /* Extract epoch data if obtained. */
1354         op_data->op_handle = md.body->handle;
1355         op_data->op_ioepoch = md.body->ioepoch;
1356
1357         ll_update_inode(inode, &md);
1358         ptlrpc_req_finished(request);
1359
1360         RETURN(rc);
1361 }
1362
1363 /* Close IO epoch and send Size-on-MDS attribute update. */
1364 static int ll_setattr_done_writing(struct inode *inode,
1365                                    struct md_op_data *op_data,
1366                                    struct md_open_data *mod)
1367 {
1368         struct ll_inode_info *lli = ll_i2info(inode);
1369         int rc = 0;
1370         ENTRY;
1371
1372         LASSERT(op_data != NULL);
1373         if (!S_ISREG(inode->i_mode))
1374                 RETURN(0);
1375
1376         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1377                op_data->op_ioepoch, PFID(&lli->lli_fid));
1378
1379         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1380         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1381         if (rc == -EAGAIN) {
1382                 /* MDS has instructed us to obtain Size-on-MDS attribute
1383                  * from OSTs and send setattr to back to MDS. */
1384                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1385                                          op_data->op_ioepoch);
1386         } else if (rc) {
1387                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1388                        inode->i_ino, rc);
1389         }
1390         RETURN(rc);
1391 }
1392
1393 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1394 {
1395         struct ll_sb_info *sbi = ll_i2sbi(inode);
1396         struct ll_inode_info *lli = ll_i2info(inode);
1397         struct lov_stripe_md *lsm = lli->lli_smd;
1398         int rc;
1399         ldlm_policy_data_t policy = { .l_extent = {new_size,
1400                                                    OBD_OBJECT_EOF } };
1401         struct lustre_handle lockh = { 0 };
1402         int local_lock = 0; /* 0 - no local lock;
1403                              * 1 - lock taken by lock_extent;
1404                              * 2 - by obd_match*/
1405         int ast_flags;
1406         int err;
1407         ENTRY;
1408
1409         UNLOCK_INODE_MUTEX(inode);
1410         UP_WRITE_I_ALLOC_SEM(inode);
1411
1412         if (sbi->ll_lockless_truncate_enable &&
1413             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1414                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1415                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1416                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1417                 if (rc > 0) {
1418                         local_lock = 2;
1419                         rc = 0;
1420                 } else if (rc == 0) {
1421                         rc = ll_file_punch(inode, new_size, 1);
1422                 }
1423         } else {
1424                 /* XXX when we fix the AST intents to pass the discard-range
1425                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1426                  * XXX here. */
1427                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1428                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1429                                     &lockh, ast_flags);
1430                 if (likely(rc == 0))
1431                         local_lock = 1;
1432         }
1433
1434         LOCK_INODE_MUTEX(inode);
1435         DOWN_WRITE_I_ALLOC_SEM(inode);
1436
1437         if (likely(rc == 0)) {
1438                 /* Only ll_inode_size_lock is taken at this level.
1439                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1440                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1441                  * ll_truncate dropped ll_inode_size_lock() */
1442                 ll_inode_size_lock(inode, 0);
1443                 if (!local_lock) {
1444                         spin_lock(&lli->lli_lock);
1445                         lli->lli_flags |= LLIF_SRVLOCK;
1446                         spin_unlock(&lli->lli_lock);
1447                 }
1448                 rc = vmtruncate(inode, new_size);
1449                 if (!local_lock) {
1450                         spin_lock(&lli->lli_lock);
1451                         lli->lli_flags &= ~LLIF_SRVLOCK;
1452                         spin_unlock(&lli->lli_lock);
1453                 }
1454                 if (rc != 0) {
1455                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1456                         ll_inode_size_unlock(inode, 0);
1457                 }
1458         }
1459
1460         if (local_lock) {
1461                 if (local_lock == 2)
1462                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1463                 else
1464                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1465                 if (unlikely(err != 0)){
1466                         CERROR("extent unlock failed: err=%d,"
1467                                " unlock method =%d\n", err, local_lock);
1468                         if (rc == 0)
1469                                 rc = err;
1470                 }
1471         }
1472         RETURN(rc);
1473 }
1474
1475 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1476  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1477  * keep these values until such a time that objects are allocated for it.
1478  * We do the MDS operations first, as it is checking permissions for us.
1479  * We don't to the MDS RPC if there is nothing that we want to store there,
1480  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1481  * going to do an RPC anyways.
1482  *
1483  * If we are doing a truncate, we will send the mtime and ctime updates
1484  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1485  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1486  * at the same time.
1487  */
1488 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1489 {
1490         struct ll_inode_info *lli = ll_i2info(inode);
1491         struct lov_stripe_md *lsm = lli->lli_smd;
1492         struct ll_sb_info *sbi = ll_i2sbi(inode);
1493         struct md_op_data *op_data = NULL;
1494         struct md_open_data *mod = NULL;
1495         int ia_valid = attr->ia_valid;
1496         int rc = 0, rc1 = 0;
1497         ENTRY;
1498
1499         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1500                attr->ia_valid);
1501         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1502
1503         if (ia_valid & ATTR_SIZE) {
1504                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1505                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1506                                attr->ia_size, ll_file_maxbytes(inode));
1507                         RETURN(-EFBIG);
1508                 }
1509
1510                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1511         }
1512
1513         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1514         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1515                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1516                         RETURN(-EPERM);
1517         }
1518
1519         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1520         if (attr->ia_valid & ATTR_CTIME) {
1521                 attr->ia_ctime = CURRENT_TIME;
1522                 attr->ia_valid |= ATTR_CTIME_SET;
1523         }
1524         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1525                 attr->ia_atime = CURRENT_TIME;
1526                 attr->ia_valid |= ATTR_ATIME_SET;
1527         }
1528         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1529                 attr->ia_mtime = CURRENT_TIME;
1530                 attr->ia_valid |= ATTR_MTIME_SET;
1531         }
1532         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1533                 /* To avoid stale mtime on mds, obtain it from ost and send
1534                    to mds. */
1535                 rc = ll_glimpse_size(inode, 0);
1536                 if (rc)
1537                         RETURN(rc);
1538
1539                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1540                 attr->ia_mtime = inode->i_mtime;
1541         }
1542
1543         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1544                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1545                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1546                        cfs_time_current_sec());
1547
1548         /* NB: ATTR_SIZE will only be set after this point if the size
1549          * resides on the MDS, ie, this file has no objects. */
1550         if (lsm)
1551                 attr->ia_valid &= ~ATTR_SIZE;
1552
1553         /* We always do an MDS RPC, even if we're only changing the size;
1554          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1555
1556         OBD_ALLOC_PTR(op_data);
1557         if (op_data == NULL)
1558                 RETURN(-ENOMEM);
1559
1560         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1561
1562         /* Open epoch for truncate. */
1563         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1564             (ia_valid & ATTR_SIZE))
1565                 op_data->op_flags = MF_EPOCH_OPEN;
1566
1567         rc = ll_md_setattr(inode, op_data, &mod);
1568         if (rc)
1569                 GOTO(out, rc);
1570
1571         if (op_data->op_ioepoch)
1572                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1573                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1574
1575         if (!lsm || !S_ISREG(inode->i_mode)) {
1576                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1577                 GOTO(out, rc = 0);
1578         }
1579
1580         /* We really need to get our PW lock before we change inode->i_size.
1581          * If we don't we can race with other i_size updaters on our node, like
1582          * ll_file_read.  We can also race with i_size propogation to other
1583          * nodes through dirtying and writeback of final cached pages.  This
1584          * last one is especially bad for racing o_append users on other
1585          * nodes. */
1586         if (ia_valid & ATTR_SIZE) {
1587                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1588         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1589                 obd_flag flags;
1590                 struct obd_info oinfo = { { { 0 } } };
1591                 struct obdo *oa;
1592
1593                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1594                        inode->i_ino, LTIME_S(attr->ia_mtime));
1595
1596                 OBDO_ALLOC(oa);
1597                 if (oa) {
1598                         oa->o_id = lsm->lsm_object_id;
1599                         oa->o_gr = lsm->lsm_object_gr;
1600                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1601
1602                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1603                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1604                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1605                                 OBD_MD_FLGROUP;
1606
1607                         obdo_from_inode(oa, inode, flags);
1608
1609                         oinfo.oi_oa = oa;
1610                         oinfo.oi_md = lsm;
1611                         oinfo.oi_capa = ll_mdscapa_get(inode);
1612
1613                         /* XXX: this looks unnecessary now. */
1614                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1615                         capa_put(oinfo.oi_capa);
1616                         if (rc)
1617                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1618                         OBDO_FREE(oa);
1619                 } else {
1620                         rc = -ENOMEM;
1621                 }
1622         }
1623         EXIT;
1624 out:
1625         if (op_data) {
1626                 if (op_data->op_ioepoch)
1627                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1628                 ll_finish_md_op_data(op_data);
1629         }
1630         return rc ? rc : rc1;
1631 }
1632
1633 int ll_setattr(struct dentry *de, struct iattr *attr)
1634 {
1635         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1636             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1637                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1638
1639         return ll_setattr_raw(de->d_inode, attr);
1640 }
1641
1642 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1643                        __u64 max_age, __u32 flags)
1644 {
1645         struct ll_sb_info *sbi = ll_s2sbi(sb);
1646         struct obd_statfs obd_osfs;
1647         int rc;
1648         ENTRY;
1649
1650         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1651         if (rc) {
1652                 CERROR("md_statfs fails: rc = %d\n", rc);
1653                 RETURN(rc);
1654         }
1655
1656         osfs->os_type = sb->s_magic;
1657
1658         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1659                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1660
1661         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1662                               &obd_osfs, max_age, flags);
1663         if (rc) {
1664                 CERROR("obd_statfs fails: rc = %d\n", rc);
1665                 RETURN(rc);
1666         }
1667
1668         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1669                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1670                obd_osfs.os_files);
1671
1672         osfs->os_bsize = obd_osfs.os_bsize;
1673         osfs->os_blocks = obd_osfs.os_blocks;
1674         osfs->os_bfree = obd_osfs.os_bfree;
1675         osfs->os_bavail = obd_osfs.os_bavail;
1676
1677         /* If we don't have as many objects free on the OST as inodes
1678          * on the MDS, we reduce the total number of inodes to
1679          * compensate, so that the "inodes in use" number is correct.
1680          */
1681         if (obd_osfs.os_ffree < osfs->os_ffree) {
1682                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1683                         obd_osfs.os_ffree;
1684                 osfs->os_ffree = obd_osfs.os_ffree;
1685         }
1686
1687         RETURN(rc);
1688 }
1689 #ifndef HAVE_STATFS_DENTRY_PARAM
1690 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1691 {
1692 #else
1693 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1694 {
1695         struct super_block *sb = de->d_sb;
1696 #endif
1697         struct obd_statfs osfs;
1698         int rc;
1699
1700         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1701         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1702
1703         /* For now we will always get up-to-date statfs values, but in the
1704          * future we may allow some amount of caching on the client (e.g.
1705          * from QOS or lprocfs updates). */
1706         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1707         if (rc)
1708                 return rc;
1709
1710         statfs_unpack(sfs, &osfs);
1711
1712         /* We need to downshift for all 32-bit kernels, because we can't
1713          * tell if the kernel is being called via sys_statfs64() or not.
1714          * Stop before overflowing f_bsize - in which case it is better
1715          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1716         if (sizeof(long) < 8) {
1717                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1718                         sfs->f_bsize <<= 1;
1719
1720                         osfs.os_blocks >>= 1;
1721                         osfs.os_bfree >>= 1;
1722                         osfs.os_bavail >>= 1;
1723                 }
1724         }
1725
1726         sfs->f_blocks = osfs.os_blocks;
1727         sfs->f_bfree = osfs.os_bfree;
1728         sfs->f_bavail = osfs.os_bavail;
1729
1730         return 0;
1731 }
1732
1733 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1734 {
1735         struct ll_inode_info *lli;
1736         struct lov_stripe_md *lsm;
1737
1738         lli = ll_i2info(inode);
1739         LASSERT(lli->lli_size_sem_owner != current);
1740         down(&lli->lli_size_sem);
1741         LASSERT(lli->lli_size_sem_owner == NULL);
1742         lli->lli_size_sem_owner = current;
1743         lsm = lli->lli_smd;
1744         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1745                  lsm, lock_lsm);
1746         if (lock_lsm)
1747                 lov_stripe_lock(lsm);
1748 }
1749
1750 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1751 {
1752         struct ll_inode_info *lli;
1753         struct lov_stripe_md *lsm;
1754
1755         lli = ll_i2info(inode);
1756         lsm = lli->lli_smd;
1757         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1758                  lsm, unlock_lsm);
1759         if (unlock_lsm)
1760                 lov_stripe_unlock(lsm);
1761         LASSERT(lli->lli_size_sem_owner == current);
1762         lli->lli_size_sem_owner = NULL;
1763         up(&lli->lli_size_sem);
1764 }
1765
1766 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1767 {
1768         struct ll_inode_info *lli = ll_i2info(inode);
1769
1770         dump_lsm(D_INODE, lsm);
1771         dump_lsm(D_INODE, lli->lli_smd);
1772         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1773                  "lsm must be joined lsm %p\n", lsm);
1774         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1775         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1776                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1777         lli->lli_smd = lsm;
1778         lli->lli_maxbytes = lsm->lsm_maxbytes;
1779         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1780                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1781 }
1782
1783 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1784 {
1785         struct ll_inode_info *lli = ll_i2info(inode);
1786         struct mdt_body *body = md->body;
1787         struct lov_stripe_md *lsm = md->lsm;
1788         struct ll_sb_info *sbi = ll_i2sbi(inode);
1789
1790         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1791         if (lsm != NULL) {
1792                 if (lli->lli_smd == NULL) {
1793                         if (lsm->lsm_magic != LOV_MAGIC &&
1794                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1795                                 dump_lsm(D_ERROR, lsm);
1796                                 LBUG();
1797                         }
1798                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1799                                lsm, inode->i_ino, inode->i_generation, inode);
1800                         /* ll_inode_size_lock() requires it is only called
1801                          * with lli_smd != NULL or lock_lsm == 0 or we can
1802                          * race between lock/unlock.  bug 9547 */
1803                         lli->lli_smd = lsm;
1804                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1805                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1806                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1807                 } else {
1808                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1809                              lli->lli_smd->lsm_stripe_count ==
1810                                         lsm->lsm_stripe_count) {
1811                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1812                                         CERROR("lsm mismatch for inode %ld\n",
1813                                                 inode->i_ino);
1814                                         CERROR("lli_smd:\n");
1815                                         dump_lsm(D_ERROR, lli->lli_smd);
1816                                         CERROR("lsm:\n");
1817                                         dump_lsm(D_ERROR, lsm);
1818                                         LBUG();
1819                                 }
1820                         } else
1821                                 ll_replace_lsm(inode, lsm);
1822                 }
1823                 if (lli->lli_smd != lsm)
1824                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1825         }
1826
1827         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1828                 if (body->valid & OBD_MD_FLRMTPERM)
1829                         ll_update_remote_perm(inode, md->remote_perm);
1830         }
1831 #ifdef CONFIG_FS_POSIX_ACL
1832         else if (body->valid & OBD_MD_FLACL) {
1833                 spin_lock(&lli->lli_lock);
1834                 if (lli->lli_posix_acl)
1835                         posix_acl_release(lli->lli_posix_acl);
1836                 lli->lli_posix_acl = md->posix_acl;
1837                 spin_unlock(&lli->lli_lock);
1838         }
1839 #endif
1840         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1841
1842         if (body->valid & OBD_MD_FLATIME &&
1843             body->atime > LTIME_S(inode->i_atime))
1844                 LTIME_S(inode->i_atime) = body->atime;
1845
1846         /* mtime is always updated with ctime, but can be set in past.
1847            As write and utime(2) may happen within 1 second, and utime's
1848            mtime has a priority over write's one, so take mtime from mds
1849            for the same ctimes. */
1850         if (body->valid & OBD_MD_FLCTIME &&
1851             body->ctime >= LTIME_S(inode->i_ctime)) {
1852                 LTIME_S(inode->i_ctime) = body->ctime;
1853                 if (body->valid & OBD_MD_FLMTIME) {
1854                         CDEBUG(D_INODE, "setting ino %lu mtime "
1855                                "from %lu to "LPU64"\n", inode->i_ino,
1856                                LTIME_S(inode->i_mtime), body->mtime);
1857                         LTIME_S(inode->i_mtime) = body->mtime;
1858                 }
1859         }
1860         if (body->valid & OBD_MD_FLMODE)
1861                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1862         if (body->valid & OBD_MD_FLTYPE)
1863                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1864         if (S_ISREG(inode->i_mode)) {
1865                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1866         } else {
1867                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1868         }
1869 #ifdef HAVE_INODE_BLKSIZE
1870         inode->i_blksize = 1<<inode->i_blkbits;
1871 #endif
1872         if (body->valid & OBD_MD_FLUID)
1873                 inode->i_uid = body->uid;
1874         if (body->valid & OBD_MD_FLGID)
1875                 inode->i_gid = body->gid;
1876         if (body->valid & OBD_MD_FLFLAGS)
1877                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1878         if (body->valid & OBD_MD_FLNLINK)
1879                 inode->i_nlink = body->nlink;
1880         if (body->valid & OBD_MD_FLRDEV)
1881                 inode->i_rdev = old_decode_dev(body->rdev);
1882
1883         if (body->valid & OBD_MD_FLID) {
1884                 /* FID shouldn't be changed! */
1885                 if (fid_is_sane(&lli->lli_fid)) {
1886                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1887                                  "Trying to change FID "DFID
1888                                  " to the "DFID", inode %lu/%u(%p)\n",
1889                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1890                                  inode->i_ino, inode->i_generation, inode);
1891                 } else
1892                         lli->lli_fid = body->fid1;
1893         }
1894
1895         LASSERT(fid_seq(&lli->lli_fid) != 0);
1896
1897         if (body->valid & OBD_MD_FLSIZE) {
1898                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1899                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1900                         struct lustre_handle lockh;
1901                         ldlm_mode_t mode;
1902
1903                         /* As it is possible a blocking ast has been processed
1904                          * by this time, we need to check there is an UPDATE
1905                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1906                          * it. */
1907                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1908                                                &lockh);
1909                         if (mode) {
1910                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1911                                                       LLIF_EPOCH_PENDING |
1912                                                       LLIF_SOM_DIRTY)) {
1913                                         CERROR("ino %lu flags %lu still has "
1914                                                "size authority! do not trust "
1915                                                "the size got from MDS\n",
1916                                                inode->i_ino, lli->lli_flags);
1917                                 } else {
1918                                         /* Use old size assignment to avoid
1919                                          * deadlock bz14138 & bz14326 */
1920                                         inode->i_size = body->size;
1921                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1922                                 }
1923                                 ldlm_lock_decref(&lockh, mode);
1924                         }
1925                 } else {
1926                         /* Use old size assignment to avoid
1927                          * deadlock bz14138 & bz14326 */
1928                         inode->i_size = body->size;
1929                 }
1930
1931                 if (body->valid & OBD_MD_FLBLOCKS)
1932                         inode->i_blocks = body->blocks;
1933         }
1934
1935         if (body->valid & OBD_MD_FLMDSCAPA) {
1936                 LASSERT(md->mds_capa);
1937                 ll_add_capa(inode, md->mds_capa);
1938         }
1939         if (body->valid & OBD_MD_FLOSSCAPA) {
1940                 LASSERT(md->oss_capa);
1941                 ll_add_capa(inode, md->oss_capa);
1942         }
1943 }
1944
1945 static struct backing_dev_info ll_backing_dev_info = {
1946         .ra_pages       = 0,    /* No readahead */
1947 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1948         .capabilities   = 0,    /* Does contribute to dirty memory */
1949 #else
1950         .memory_backed  = 0,    /* Does contribute to dirty memory */
1951 #endif
1952 };
1953
1954 void ll_read_inode2(struct inode *inode, void *opaque)
1955 {
1956         struct lustre_md *md = opaque;
1957         struct ll_inode_info *lli = ll_i2info(inode);
1958         ENTRY;
1959
1960         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1961                inode->i_ino, inode->i_generation, inode);
1962
1963         ll_lli_init(lli);
1964
1965         LASSERT(!lli->lli_smd);
1966
1967         /* Core attributes from the MDS first.  This is a new inode, and
1968          * the VFS doesn't zero times in the core inode so we have to do
1969          * it ourselves.  They will be overwritten by either MDS or OST
1970          * attributes - we just need to make sure they aren't newer. */
1971         LTIME_S(inode->i_mtime) = 0;
1972         LTIME_S(inode->i_atime) = 0;
1973         LTIME_S(inode->i_ctime) = 0;
1974         inode->i_rdev = 0;
1975         ll_update_inode(inode, md);
1976
1977         /* OIDEBUG(inode); */
1978
1979         if (S_ISREG(inode->i_mode)) {
1980                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1981                 inode->i_op = &ll_file_inode_operations;
1982                 inode->i_fop = sbi->ll_fop;
1983                 inode->i_mapping->a_ops = &ll_aops;
1984                 EXIT;
1985         } else if (S_ISDIR(inode->i_mode)) {
1986                 inode->i_op = &ll_dir_inode_operations;
1987                 inode->i_fop = &ll_dir_operations;
1988                 inode->i_mapping->a_ops = &ll_dir_aops;
1989                 EXIT;
1990         } else if (S_ISLNK(inode->i_mode)) {
1991                 inode->i_op = &ll_fast_symlink_inode_operations;
1992                 EXIT;
1993         } else {
1994                 inode->i_op = &ll_special_inode_operations;
1995
1996                 init_special_inode(inode, inode->i_mode,
1997                                    kdev_t_to_nr(inode->i_rdev));
1998
1999                 /* initializing backing dev info. */
2000                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2001
2002                 EXIT;
2003         }
2004 }
2005
2006 void ll_delete_inode(struct inode *inode)
2007 {
2008         struct ll_sb_info *sbi = ll_i2sbi(inode);
2009         int rc;
2010         ENTRY;
2011
2012         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2013         if (rc) {
2014                 CERROR("fid_delete() failed, rc %d\n", rc);
2015         }
2016         truncate_inode_pages(&inode->i_data, 0);
2017         clear_inode(inode);
2018
2019         EXIT;
2020 }
2021
2022 int ll_iocontrol(struct inode *inode, struct file *file,
2023                  unsigned int cmd, unsigned long arg)
2024 {
2025         struct ll_sb_info *sbi = ll_i2sbi(inode);
2026         struct ptlrpc_request *req = NULL;
2027         int rc, flags = 0;
2028         ENTRY;
2029
2030         switch(cmd) {
2031         case EXT3_IOC_GETFLAGS: {
2032                 struct mdt_body *body;
2033                 struct obd_capa *oc;
2034
2035                 oc = ll_mdscapa_get(inode);
2036                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2037                                 OBD_MD_FLFLAGS, 0, &req);
2038                 capa_put(oc);
2039                 if (rc) {
2040                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2041                         RETURN(-abs(rc));
2042                 }
2043
2044                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2045
2046                 flags = body->flags;
2047
2048                 ptlrpc_req_finished(req);
2049
2050                 RETURN(put_user(flags, (int *)arg));
2051         }
2052         case EXT3_IOC_SETFLAGS: {
2053                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2054                 struct obd_info oinfo = { { { 0 } } };
2055                 struct md_op_data *op_data;
2056
2057                 if (get_user(flags, (int *)arg))
2058                         RETURN(-EFAULT);
2059
2060                 oinfo.oi_md = lsm;
2061                 OBDO_ALLOC(oinfo.oi_oa);
2062                 if (!oinfo.oi_oa)
2063                         RETURN(-ENOMEM);
2064
2065                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2066                                              LUSTRE_OPC_ANY, NULL);
2067                 if (IS_ERR(op_data))
2068                         RETURN(PTR_ERR(op_data));
2069
2070                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2071                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2072                 rc = md_setattr(sbi->ll_md_exp, op_data,
2073                                 NULL, 0, NULL, 0, &req, NULL);
2074                 ll_finish_md_op_data(op_data);
2075                 ptlrpc_req_finished(req);
2076                 if (rc || lsm == NULL) {
2077                         OBDO_FREE(oinfo.oi_oa);
2078                         RETURN(rc);
2079                 }
2080
2081                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2082                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2083                 oinfo.oi_oa->o_flags = flags;
2084                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2085                                        OBD_MD_FLGROUP;
2086                 oinfo.oi_capa = ll_mdscapa_get(inode);
2087
2088                 obdo_from_inode(oinfo.oi_oa, inode,
2089                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2090                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2091                 capa_put(oinfo.oi_capa);
2092                 OBDO_FREE(oinfo.oi_oa);
2093                 if (rc) {
2094                         if (rc != -EPERM && rc != -EACCES)
2095                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2096                         RETURN(rc);
2097                 }
2098
2099                 inode->i_flags = ll_ext_to_inode_flags(flags |
2100                                                        MDS_BFLAG_EXT_FLAGS);
2101                 RETURN(0);
2102         }
2103         default:
2104                 RETURN(-ENOSYS);
2105         }
2106
2107         RETURN(0);
2108 }
2109
2110 int ll_flush_ctx(struct inode *inode)
2111 {
2112         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2113
2114         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2115
2116         obd_set_info_async(sbi->ll_md_exp,
2117                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2118                            0, NULL, NULL);
2119         obd_set_info_async(sbi->ll_dt_exp,
2120                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2121                            0, NULL, NULL);
2122         return 0;
2123 }
2124
2125 /* umount -f client means force down, don't save state */
2126 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2127 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2128 {
2129         struct super_block *sb = vfsmnt->mnt_sb;
2130 #else
2131 void ll_umount_begin(struct super_block *sb)
2132 {
2133 #endif
2134         struct lustre_sb_info *lsi = s2lsi(sb);
2135         struct ll_sb_info *sbi = ll_s2sbi(sb);
2136         struct obd_device *obd;
2137         struct obd_ioctl_data ioc_data = { 0 };
2138         ENTRY;
2139
2140 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2141         if (!(flags & MNT_FORCE)) {
2142                 EXIT;
2143                 return;
2144         }
2145 #endif
2146
2147         /* Tell the MGC we got umount -f */
2148         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2149
2150         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2151                sb->s_count, atomic_read(&sb->s_active));
2152
2153         obd = class_exp2obd(sbi->ll_md_exp);
2154         if (obd == NULL) {
2155                 CERROR("Invalid MDC connection handle "LPX64"\n",
2156                        sbi->ll_md_exp->exp_handle.h_cookie);
2157                 EXIT;
2158                 return;
2159         }
2160         obd->obd_force = 1;
2161         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2162                       &ioc_data, NULL);
2163
2164         obd = class_exp2obd(sbi->ll_dt_exp);
2165         if (obd == NULL) {
2166                 CERROR("Invalid LOV connection handle "LPX64"\n",
2167                        sbi->ll_dt_exp->exp_handle.h_cookie);
2168                 EXIT;
2169                 return;
2170         }
2171
2172         obd->obd_force = 1;
2173         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2174                       &ioc_data, NULL);
2175
2176         /* Really, we'd like to wait until there are no requests outstanding,
2177          * and then continue.  For now, we just invalidate the requests,
2178          * schedule, and hope.
2179          */
2180         schedule();
2181
2182         EXIT;
2183 }
2184
2185 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2186 {
2187         struct ll_sb_info *sbi = ll_s2sbi(sb);
2188         int err;
2189         __u32 read_only;
2190
2191         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2192                 read_only = *flags & MS_RDONLY;
2193                 err = obd_set_info_async(sbi->ll_md_exp,
2194                                          sizeof(KEY_READ_ONLY),
2195                                          KEY_READ_ONLY, sizeof(read_only),
2196                                          &read_only, NULL);
2197                 if (err) {
2198                         CERROR("Failed to change the read-only flag during "
2199                                "remount: %d\n", err);
2200                         return err;
2201                 }
2202
2203                 if (read_only)
2204                         sb->s_flags |= MS_RDONLY;
2205                 else
2206                         sb->s_flags &= ~MS_RDONLY;
2207         }
2208         return 0;
2209 }
2210
2211 int ll_prep_inode(struct inode **inode,
2212                   struct ptlrpc_request *req,
2213                   struct super_block *sb)
2214 {
2215         struct ll_sb_info *sbi = NULL;
2216         struct lustre_md md;
2217         int rc = 0;
2218         ENTRY;
2219
2220         LASSERT(*inode || sb);
2221         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2222         prune_deathrow(sbi, 1);
2223         memset(&md, 0, sizeof(struct lustre_md));
2224
2225         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2226                               sbi->ll_md_exp, &md);
2227         if (rc)
2228                 RETURN(rc);
2229
2230         if (*inode) {
2231                 ll_update_inode(*inode, &md);
2232         } else {
2233                 LASSERT(sb != NULL);
2234
2235                 /*
2236                  * At this point server returns to client's same fid as client
2237                  * generated for creating. So using ->fid1 is okay here.
2238                  */
2239                 LASSERT(fid_is_sane(&md.body->fid1));
2240
2241                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2242                 if (*inode == NULL || is_bad_inode(*inode)) {
2243                         if (md.lsm)
2244                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2245 #ifdef CONFIG_FS_POSIX_ACL
2246                         if (md.posix_acl) {
2247                                 posix_acl_release(md.posix_acl);
2248                                 md.posix_acl = NULL;
2249                         }
2250 #endif
2251                         rc = -ENOMEM;
2252                         CERROR("new_inode -fatal: rc %d\n", rc);
2253                         GOTO(out, rc);
2254                 }
2255         }
2256
2257         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2258                          ll_i2info(*inode)->lli_smd);
2259 out:
2260         md_free_lustre_md(sbi->ll_md_exp, &md);
2261         RETURN(rc);
2262 }
2263
2264 char *llap_origins[] = {
2265         [LLAP_ORIGIN_UNKNOWN] = "--",
2266         [LLAP_ORIGIN_READPAGE] = "rp",
2267         [LLAP_ORIGIN_READAHEAD] = "ra",
2268         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2269         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2270         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2271 };
2272
2273 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2274                                              struct list_head *list)
2275 {
2276         struct ll_async_page *llap;
2277         struct list_head *pos;
2278
2279         list_for_each(pos, list) {
2280                 if (pos == head)
2281                         return NULL;
2282                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2283                 if (llap->llap_page == NULL)
2284                         continue;
2285                 return llap;
2286         }
2287         LBUG();
2288         return NULL;
2289 }
2290
2291 int ll_obd_statfs(struct inode *inode, void *arg)
2292 {
2293         struct ll_sb_info *sbi = NULL;
2294         struct obd_export *exp;
2295         char *buf = NULL;
2296         struct obd_ioctl_data *data = NULL;
2297         __u32 type;
2298         int len = 0, rc;
2299
2300         if (!inode || !(sbi = ll_i2sbi(inode)))
2301                 GOTO(out_statfs, rc = -EINVAL);
2302
2303         rc = obd_ioctl_getdata(&buf, &len, arg);
2304         if (rc)
2305                 GOTO(out_statfs, rc);
2306
2307         data = (void*)buf;
2308         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2309             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2310                 GOTO(out_statfs, rc = -EINVAL);
2311
2312         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2313         if (type == LL_STATFS_MDC)
2314                 exp = sbi->ll_md_exp;
2315         else if (type == LL_STATFS_LOV)
2316                 exp = sbi->ll_dt_exp;
2317         else
2318                 GOTO(out_statfs, rc = -ENODEV);
2319
2320         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2321         if (rc)
2322                 GOTO(out_statfs, rc);
2323 out_statfs:
2324         if (buf)
2325                 obd_ioctl_freedata(buf, len);
2326         return rc;
2327 }
2328
2329 int ll_process_config(struct lustre_cfg *lcfg)
2330 {
2331         char *ptr;
2332         void *sb;
2333         struct lprocfs_static_vars lvars;
2334         unsigned long x;
2335         int rc = 0;
2336
2337         lprocfs_llite_init_vars(&lvars);
2338
2339         /* The instance name contains the sb: lustre-client-aacfe000 */
2340         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2341         if (!ptr || !*(++ptr))
2342                 return -EINVAL;
2343         if (sscanf(ptr, "%lx", &x) != 1)
2344                 return -EINVAL;
2345         sb = (void *)x;
2346         /* This better be a real Lustre superblock! */
2347         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2348
2349         /* Note we have not called client_common_fill_super yet, so
2350            proc fns must be able to handle that! */
2351         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2352                                       lcfg, sb);
2353         return(rc);
2354 }
2355
2356 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2357 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2358                                        struct inode *i1, struct inode *i2,
2359                                        const char *name, int namelen,
2360                                        int mode, __u32 opc, void *data)
2361 {
2362         LASSERT(i1 != NULL);
2363
2364         if (namelen > ll_i2sbi(i1)->ll_namelen)
2365                 return ERR_PTR(-ENAMETOOLONG);
2366
2367         if (op_data == NULL)
2368                 OBD_ALLOC_PTR(op_data);
2369
2370         if (op_data == NULL)
2371                 return ERR_PTR(-ENOMEM);
2372
2373         ll_i2gids(op_data->op_suppgids, i1, i2);
2374         op_data->op_fid1 = *ll_inode2fid(i1);
2375         op_data->op_capa1 = ll_mdscapa_get(i1);
2376
2377         if (i2) {
2378                 op_data->op_fid2 = *ll_inode2fid(i2);
2379                 op_data->op_capa2 = ll_mdscapa_get(i2);
2380         } else {
2381                 fid_zero(&op_data->op_fid2);
2382                 op_data->op_capa2 = NULL;
2383         }
2384
2385         op_data->op_name = name;
2386         op_data->op_namelen = namelen;
2387         op_data->op_mode = mode;
2388         op_data->op_mod_time = cfs_time_current_sec();
2389         op_data->op_fsuid = current->fsuid;
2390         op_data->op_fsgid = current->fsgid;
2391         op_data->op_cap = current->cap_effective;
2392         op_data->op_bias = MDS_CHECK_SPLIT;
2393         op_data->op_opc = opc;
2394         op_data->op_mds = 0;
2395         op_data->op_data = data;
2396
2397         return op_data;
2398 }
2399
2400 void ll_finish_md_op_data(struct md_op_data *op_data)
2401 {
2402         capa_put(op_data->op_capa1);
2403         capa_put(op_data->op_capa2);
2404         OBD_FREE_PTR(op_data);
2405 }