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
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  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 obd_uuid *uuid;
 292         struct lustre_md lmd;
 293         obd_valid valid;
 294         int size, err, checksum;
 295         ENTRY;
 296
 297         obd = class_name2obd(md);
 298         if (!obd) {
 299                 CERROR("MD %s: not setup or attached\n", md);
 300                 RETURN(-EINVAL);
 301         }
 302
 303         OBD_ALLOC_PTR(data);
 304         if (data == NULL)
 305                 RETURN(-ENOMEM);
 306
 307         if (proc_lustre_fs_root) {
 308                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 309                                                   dt, md);
 310                 if (err < 0)
 311                         CERROR("could not register mount in /proc/fs/lustre\n");
 312         }
 313
 314         /* indicate the features supported by this client */
 315         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 316                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 317                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 318                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 319                                   OBD_CONNECT_FID      | OBD_CONNECT_AT;
 320
 321 #ifdef HAVE_LRU_RESIZE_SUPPORT
 322         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 323                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 324 #endif
 325 #ifdef CONFIG_FS_POSIX_ACL
 326         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 327 #endif
 328         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 329         data->ocd_version = LUSTRE_VERSION_CODE;
 330
 331         if (sb->s_flags & MS_RDONLY)
 332                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 333         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 334                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 335
 336 #ifdef HAVE_MS_FLOCK_LOCK
 337         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 338         sb->s_flags |= MS_FLOCK_LOCK;
 339 #endif
 340
 341         if (sbi->ll_flags & LL_SBI_FLOCK)
 342                 sbi->ll_fop = &ll_file_operations_flock;
 343         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 344                 sbi->ll_fop = &ll_file_operations;
 345         else
 346                 sbi->ll_fop = &ll_file_operations_noflock;
 347
 348         /* real client */
 349         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 350         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 351                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 352                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 353         } else {
 354                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 355                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 356         }
 357
 358         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 359         if (err == -EBUSY) {
 360                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 361                                    "recovery, of which this client is not a "
 362                                    "part. Please wait for recovery to complete,"
 363                                    " abort, or time out.\n", md);
 364                 GOTO(out, err);
 365         } else if (err) {
 366                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 367                 GOTO(out, err);
 368         }
 369         sbi->ll_md_exp = class_conn2export(&md_conn);
 370
 371         err = obd_fid_init(sbi->ll_md_exp);
 372         if (err) {
 373                 CERROR("Can't init metadata layer FID infrastructure, "
 374                        "rc %d\n", err);
 375                 GOTO(out_md, err);
 376         }
 377
 378         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 379         if (err)
 380                 GOTO(out_md_fid, err);
 381
 382         size = sizeof(*data);
 383         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 384                            KEY_CONN_DATA,  &size, data, NULL);
 385         if (err) {
 386                 CERROR("Get connect data failed: %d \n", err);
 387                 GOTO(out_md, err);
 388         }
 389
 390         LASSERT(osfs.os_bsize);
 391         sb->s_blocksize = osfs.os_bsize;
 392         sb->s_blocksize_bits = log2(osfs.os_bsize);
 393         sb->s_magic = LL_SUPER_MAGIC;
 394
 395         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 396          *         retval = in_file->f_op->sendfile(...);
 397          *         if (*ppos > max)
 398          *                 retval = -EOVERFLOW;
 399          *
 400          * it will check if *ppos is greater than max. However, max equals to
 401          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 402          * has been defined as a signed long long ineger in linux kernel. */
 403 #if BITS_PER_LONG == 64
 404         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 405 #else
 406         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 407 #endif
 408         sbi->ll_namelen = osfs.os_namelen;
 409         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 410
 411         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 412             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 413                 LCONSOLE_INFO("Disabling user_xattr feature because "
 414                               "it is not supported on the server\n");
 415                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 416         }
 417
 418         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 419 #ifdef MS_POSIXACL
 420                 sb->s_flags |= MS_POSIXACL;
 421 #endif
 422                 sbi->ll_flags |= LL_SBI_ACL;
 423         } else {
 424                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 425 #ifdef MS_POSIXACL
 426                 sb->s_flags &= ~MS_POSIXACL;
 427 #endif
 428                 sbi->ll_flags &= ~LL_SBI_ACL;
 429         }
 430
 431         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 432                 sbi->ll_flags |= LL_SBI_JOIN;
 433
 434         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 435                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 436                         /* sometimes local client claims to be remote, but mdt
 437                          * will disagree when client gss not applied. */
 438                         LCONSOLE_INFO("client claims to be remote, but server "
 439                                       "rejected, forced to be local.\n");
 440                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 441                 }
 442         } else {
 443                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 444                         /* with gss applied, remote client can not claim to be
 445                          * local, so mdt maybe force client to be remote. */
 446                         LCONSOLE_INFO("client claims to be local, but server "
 447                                       "rejected, forced to be remote.\n");
 448                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 449                 }
 450         }
 451
 452         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 453                 LCONSOLE_INFO("client enabled MDS capability!\n");
 454                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 455         }
 456
 457         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 458                 LCONSOLE_INFO("client enabled OSS capability!\n");
 459                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 460         }
 461
 462         obd = class_name2obd(dt);
 463         if (!obd) {
 464                 CERROR("DT %s: not setup or attached\n", dt);
 465                 GOTO(out_md_fid, err = -ENODEV);
 466         }
 467
 468         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 469                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 470                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 471                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 472                                   OBD_CONNECT_AT;
 473         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 474                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 475
 476         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 477                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 478                  * disabled by default, because it can still be enabled on the
 479                  * fly via /proc. As a consequence, we still need to come to an
 480                  * agreement on the supported algorithms at connect time */
 481                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 482
 483                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 484                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 485                 else
 486                         /* send the list of supported checksum types */
 487                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 488         }
 489
 490 #ifdef HAVE_LRU_RESIZE_SUPPORT
 491         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 492 #endif
 493         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 494                "ocd_grant: %d\n", data->ocd_connect_flags,
 495                data->ocd_version, data->ocd_grant);
 496
 497         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 498         obd->obd_upcall.onu_upcall = ll_ocd_update;
 499         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 500
 501         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 502         if (err == -EBUSY) {
 503                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 504                                    "recovery, of which this client is not a "
 505                                    "part.  Please wait for recovery to "
 506                                    "complete, abort, or time out.\n", dt);
 507                 GOTO(out_md_fid, err);
 508         } else if (err) {
 509                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 510                 GOTO(out_md_fid, err);
 511         }
 512
 513         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 514
 515         err = obd_fid_init(sbi->ll_dt_exp);
 516         if (err) {
 517                 CERROR("Can't init data layer FID infrastructure, "
 518                        "rc %d\n", err);
 519                 GOTO(out_dt, err);
 520         }
 521
 522         spin_lock(&sbi->ll_lco.lco_lock);
 523         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 524         spin_unlock(&sbi->ll_lco.lco_lock);
 525
 526         err = obd_register_page_removal_cb(sbi->ll_dt_exp,
 527                                            ll_page_removal_cb,
 528                                            ll_pin_extent_cb);
 529         if (err) {
 530                 CERROR("cannot register page removal callback: rc = %d\n",err);
 531                 GOTO(out_dt, err);
 532         }
 533         err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
 534                                           ll_extent_lock_cancel_cb);
 535         if (err) {
 536                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
 537                 GOTO(out_page_rm_cb, err);
 538         }
 539
 540         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 541         if (err) {
 542                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 543                 GOTO(out_lock_cn_cb, err);
 544         }
 545
 546         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 547                                   0, NULL, NULL, NULL, 0, NULL);
 548         if (err < 0) {
 549                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 550                                    "filesystem. There must be at least one "
 551                                    "active OST for a client to start.\n");
 552                 GOTO(out_lock_cn_cb, err);
 553         }
 554
 555         if (!ll_async_page_slab) {
 556                 ll_async_page_slab_size =
 557                         size_round(sizeof(struct ll_async_page)) + err;
 558                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 559                                                           ll_async_page_slab_size,
 560                                                           0, 0);
 561                 if (!ll_async_page_slab)
 562                         GOTO(out_lock_cn_cb, err = -ENOMEM);
 563         }
 564
 565         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 566         if (err) {
 567                 CERROR("cannot mds_connect: rc = %d\n", err);
 568                 GOTO(out_lock_cn_cb, err);
 569         }
 570         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 571         sbi->ll_root_fid = rootfid;
 572
 573         sb->s_op = &lustre_super_operations;
 574         sb->s_export_op = &lustre_export_operations;
 575
 576         /* make root inode
 577          * XXX: move this to after cbd setup? */
 578         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 579         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 580                 valid |= OBD_MD_FLRMTPERM;
 581         else if (sbi->ll_flags & LL_SBI_ACL)
 582                 valid |= OBD_MD_FLACL;
 583
 584         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 585         if (oc)
 586                 free_capa(oc);
 587         if (err) {
 588                 CERROR("md_getattr failed for root: rc = %d\n", err);
 589                 GOTO(out_lock_cn_cb, err);
 590         }
 591         memset(&lmd, 0, sizeof(lmd));
 592         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 593                                sbi->ll_md_exp, &lmd);
 594         if (err) {
 595                 CERROR("failed to understand root inode md: rc = %d\n", err);
 596                 ptlrpc_req_finished (request);
 597                 GOTO(out_lock_cn_cb, err);
 598         }
 599
 600         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 601         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 602         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 603         ptlrpc_req_finished(request);
 604
 605         if (root == NULL || is_bad_inode(root)) {
 606                 if (lmd.lsm)
 607                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 608 #ifdef CONFIG_FS_POSIX_ACL
 609                 if (lmd.posix_acl) {
 610                         posix_acl_release(lmd.posix_acl);
 611                         lmd.posix_acl = NULL;
 612                 }
 613 #endif
 614                 CERROR("lustre_lite: bad iget4 for root\n");
 615                 GOTO(out_root, err = -EBADF);
 616         }
 617
 618         err = ll_close_thread_start(&sbi->ll_lcq);
 619         if (err) {
 620                 CERROR("cannot start close thread: rc %d\n", err);
 621                 GOTO(out_root, err);
 622         }
 623
 624 #ifdef CONFIG_FS_POSIX_ACL
 625         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 626                 rct_init(&sbi->ll_rct);
 627                 et_init(&sbi->ll_et);
 628         }
 629 #endif
 630
 631         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 632         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 633                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 634                                  NULL);
 635
 636         sb->s_root = d_alloc_root(root);
 637         if (data != NULL)
 638                 OBD_FREE(data, sizeof(*data));
 639
 640         sb->s_root->d_op = &ll_d_root_ops;
 641
 642         sbi->ll_sdev_orig = sb->s_dev;
 643
 644         /* We set sb->s_dev equal on all lustre clients in order to support
 645          * NFS export clustering.  NFSD requires that the FSID be the same
 646          * on all clients. */
 647         /* s_dev is also used in lt_compare() to compare two fs, but that is
 648          * only a node-local comparison. */
 649         uuid = obd_get_uuid(sbi->ll_md_exp);
 650         if (uuid != NULL)
 651                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 652
 653         RETURN(err);
 654 out_root:
 655         if (root)
 656                 iput(root);
 657 out_lock_cn_cb:
 658         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
 659                                       ll_extent_lock_cancel_cb);
 660 out_page_rm_cb:
 661         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 662                                        ll_page_removal_cb);
 663         obd_fid_fini(sbi->ll_dt_exp);
 664 out_dt:
 665         obd_disconnect(sbi->ll_dt_exp);
 666         sbi->ll_dt_exp = NULL;
 667 out_md_fid:
 668         obd_fid_fini(sbi->ll_md_exp);
 669 out_md:
 670         obd_disconnect(sbi->ll_md_exp);
 671         sbi->ll_md_exp = NULL;
 672 out:
 673         if (data != NULL)
 674                 OBD_FREE_PTR(data);
 675         lprocfs_unregister_mountpoint(sbi);
 676         return err;
 677 }
 678
 679 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 680 {
 681         int size, rc;
 682
 683         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 684         size = sizeof(int);
 685         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 686                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 687         if (rc)
 688                 CERROR("Get max mdsize error rc %d \n", rc);
 689
 690         RETURN(rc);
 691 }
 692
 693 void ll_dump_inode(struct inode *inode)
 694 {
 695         struct list_head *tmp;
 696         int dentry_count = 0;
 697
 698         LASSERT(inode != NULL);
 699
 700         list_for_each(tmp, &inode->i_dentry)
 701                 dentry_count++;
 702
 703         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 704                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 705                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 706 }
 707
 708 void lustre_dump_dentry(struct dentry *dentry, int recur)
 709 {
 710         struct list_head *tmp;
 711         int subdirs = 0;
 712
 713         LASSERT(dentry != NULL);
 714
 715         list_for_each(tmp, &dentry->d_subdirs)
 716                 subdirs++;
 717
 718         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 719                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 720                dentry->d_name.len, dentry->d_name.name,
 721                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 722                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 723                dentry->d_flags, dentry->d_fsdata, subdirs);
 724         if (dentry->d_inode != NULL)
 725                 ll_dump_inode(dentry->d_inode);
 726
 727         if (recur == 0)
 728                 return;
 729
 730         list_for_each(tmp, &dentry->d_subdirs) {
 731                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 732                 lustre_dump_dentry(d, recur - 1);
 733         }
 734 }
 735
 736 #ifdef HAVE_EXPORT___IGET
 737 static void prune_dir_dentries(struct inode *inode)
 738 {
 739         struct dentry *dentry, *prev = NULL;
 740
 741         /* due to lustre specific logic, a directory
 742          * can have few dentries - a bug from VFS POV */
 743 restart:
 744         spin_lock(&dcache_lock);
 745         if (!list_empty(&inode->i_dentry)) {
 746                 dentry = list_entry(inode->i_dentry.prev,
 747                                     struct dentry, d_alias);
 748                 /* in order to prevent infinite loops we
 749                  * break if previous dentry is busy */
 750                 if (dentry != prev) {
 751                         prev = dentry;
 752                         dget_locked(dentry);
 753                         spin_unlock(&dcache_lock);
 754
 755                         /* try to kill all child dentries */
 756                         lock_dentry(dentry);
 757                         shrink_dcache_parent(dentry);
 758                         unlock_dentry(dentry);
 759                         dput(dentry);
 760
 761                         /* now try to get rid of current dentry */
 762                         d_prune_aliases(inode);
 763                         goto restart;
 764                 }
 765         }
 766         spin_unlock(&dcache_lock);
 767 }
 768
 769 static void prune_deathrow_one(struct ll_inode_info *lli)
 770 {
 771         struct inode *inode = ll_info2i(lli);
 772
 773         /* first, try to drop any dentries - they hold a ref on the inode */
 774         if (S_ISDIR(inode->i_mode))
 775                 prune_dir_dentries(inode);
 776         else
 777                 d_prune_aliases(inode);
 778
 779
 780         /* if somebody still uses it, leave it */
 781         LASSERT(atomic_read(&inode->i_count) > 0);
 782         if (atomic_read(&inode->i_count) > 1)
 783                 goto out;
 784
 785         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 786                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 787
 788         /* seems nobody uses it anymore */
 789         inode->i_nlink = 0;
 790
 791 out:
 792         iput(inode);
 793         return;
 794 }
 795
 796 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 797 {
 798         struct ll_inode_info *lli;
 799         int empty;
 800
 801         do {
 802                 if (need_resched() && try)
 803                         break;
 804
 805                 if (try) {
 806                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 807                                 break;
 808                 } else {
 809                         spin_lock(&sbi->ll_deathrow_lock);
 810                 }
 811
 812                 empty = 1;
 813                 lli = NULL;
 814                 if (!list_empty(&sbi->ll_deathrow)) {
 815                         lli = list_entry(sbi->ll_deathrow.next,
 816                                          struct ll_inode_info,
 817                                          lli_dead_list);
 818                         list_del_init(&lli->lli_dead_list);
 819                         if (!list_empty(&sbi->ll_deathrow))
 820                                 empty = 0;
 821                 }
 822                 spin_unlock(&sbi->ll_deathrow_lock);
 823
 824                 if (lli)
 825                         prune_deathrow_one(lli);
 826
 827         } while (empty == 0);
 828 }
 829 #else /* !HAVE_EXPORT___IGET */
 830 #define prune_deathrow(sbi, try) do {} while (0)
 831 #endif /* HAVE_EXPORT___IGET */
 832
 833 void client_common_put_super(struct super_block *sb)
 834 {
 835         struct ll_sb_info *sbi = ll_s2sbi(sb);
 836         ENTRY;
 837
 838 #ifdef CONFIG_FS_POSIX_ACL
 839         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 840                 et_fini(&sbi->ll_et);
 841                 rct_fini(&sbi->ll_rct);
 842         }
 843 #endif
 844
 845         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 846
 847         ll_close_thread_shutdown(sbi->ll_lcq);
 848
 849         /* destroy inodes in deathrow */
 850         prune_deathrow(sbi, 0);
 851
 852         list_del(&sbi->ll_conn_chain);
 853
 854         obd_fid_fini(sbi->ll_dt_exp);
 855         obd_disconnect(sbi->ll_dt_exp);
 856         sbi->ll_dt_exp = NULL;
 857
 858         lprocfs_unregister_mountpoint(sbi);
 859
 860         obd_fid_fini(sbi->ll_md_exp);
 861         obd_disconnect(sbi->ll_md_exp);
 862         sbi->ll_md_exp = NULL;
 863
 864         EXIT;
 865 }
 866
 867 void ll_kill_super(struct super_block *sb)
 868 {
 869         struct ll_sb_info *sbi;
 870
 871         ENTRY;
 872
 873         /* not init sb ?*/
 874         if (!(sb->s_flags & MS_ACTIVE))
 875                 return;
 876
 877         sbi = ll_s2sbi(sb);
 878         /* we need restore s_dev from changed for clustred NFS before put_super
 879          * because new kernels have cached s_dev and change sb->s_dev in
 880          * put_super not affected real removing devices */
 881         if (sbi)
 882                 sb->s_dev = sbi->ll_sdev_orig;
 883         EXIT;
 884 }
 885
 886 char *ll_read_opt(const char *opt, char *data)
 887 {
 888         char *value;
 889         char *retval;
 890         ENTRY;
 891
 892         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 893         if (strncmp(opt, data, strlen(opt)))
 894                 RETURN(NULL);
 895         if ((value = strchr(data, '=')) == NULL)
 896                 RETURN(NULL);
 897
 898         value++;
 899         OBD_ALLOC(retval, strlen(value) + 1);
 900         if (!retval) {
 901                 CERROR("out of memory!\n");
 902                 RETURN(NULL);
 903         }
 904
 905         memcpy(retval, value, strlen(value)+1);
 906         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 907         RETURN(retval);
 908 }
 909
 910 static inline int ll_set_opt(const char *opt, char *data, int fl)
 911 {
 912         if (strncmp(opt, data, strlen(opt)) != 0)
 913                 return(0);
 914         else
 915                 return(fl);
 916 }
 917
 918 /* non-client-specific mount options are parsed in lmd_parse */
 919 static int ll_options(char *options, int *flags)
 920 {
 921         int tmp;
 922         char *s1 = options, *s2;
 923         ENTRY;
 924
 925         if (!options)
 926                 RETURN(0);
 927
 928         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 929
 930         while (*s1) {
 931                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 932                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 933                 if (tmp) {
 934                         *flags |= tmp;
 935                         goto next;
 936                 }
 937                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 938                 if (tmp) {
 939                         *flags |= tmp;
 940                         goto next;
 941                 }
 942                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 943                 if (tmp) {
 944                         *flags |= tmp;
 945                         goto next;
 946                 }
 947                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 948                 if (tmp) {
 949                         *flags &= ~tmp;
 950                         goto next;
 951                 }
 952                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 953                 if (tmp) {
 954                         *flags |= tmp;
 955                         goto next;
 956                 }
 957                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 958                 if (tmp) {
 959                         *flags &= ~tmp;
 960                         goto next;
 961                 }
 962                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 963                 if (tmp) {
 964                         /* Ignore deprecated mount option.  The client will
 965                          * always try to mount with ACL support, whether this
 966                          * is used depends on whether server supports it. */
 967                         goto next;
 968                 }
 969                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 970                 if (tmp) {
 971                         goto next;
 972                 }
 973                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 974                 if (tmp) {
 975                         *flags |= tmp;
 976                         goto next;
 977                 }
 978
 979                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 980                 if (tmp) {
 981                         *flags |= tmp;
 982                         goto next;
 983                 }
 984                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 985                 if (tmp) {
 986                         *flags &= ~tmp;
 987                         goto next;
 988                 }
 989                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 990                 if (tmp) {
 991                         *flags |= tmp;
 992                         goto next;
 993                 }
 994                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 995                 if (tmp) {
 996                         *flags &= ~tmp;
 997                         goto next;
 998                 }
 999
1000                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1001                                    s1);
1002                 RETURN(-EINVAL);
1003
1004 next:
1005                 /* Find next opt */
1006                 s2 = strchr(s1, ',');
1007                 if (s2 == NULL)
1008                         break;
1009                 s1 = s2 + 1;
1010         }
1011         RETURN(0);
1012 }
1013
1014 void ll_lli_init(struct ll_inode_info *lli)
1015 {
1016         lli->lli_inode_magic = LLI_INODE_MAGIC;
1017         sema_init(&lli->lli_size_sem, 1);
1018         sema_init(&lli->lli_write_sem, 1);
1019         lli->lli_flags = 0;
1020         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1021         spin_lock_init(&lli->lli_lock);
1022         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1023         INIT_LIST_HEAD(&lli->lli_close_list);
1024         lli->lli_inode_magic = LLI_INODE_MAGIC;
1025         sema_init(&lli->lli_och_sem, 1);
1026         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1027         lli->lli_mds_exec_och = NULL;
1028         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1029         lli->lli_open_fd_exec_count = 0;
1030         INIT_LIST_HEAD(&lli->lli_dead_list);
1031         lli->lli_remote_perms = NULL;
1032         lli->lli_rmtperm_utime = 0;
1033         sema_init(&lli->lli_rmtperm_sem, 1);
1034         INIT_LIST_HEAD(&lli->lli_oss_capas);
1035 }
1036
1037 int ll_fill_super(struct super_block *sb)
1038 {
1039         struct lustre_profile *lprof;
1040         struct lustre_sb_info *lsi = s2lsi(sb);
1041         struct ll_sb_info *sbi;
1042         char  *dt = NULL, *md = NULL;
1043         char  *profilenm = get_profile_name(sb);
1044         struct config_llog_instance cfg = {0, };
1045         char   ll_instance[sizeof(sb) * 2 + 1];
1046         int    err;
1047         ENTRY;
1048
1049         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1050
1051         cfs_module_get();
1052
1053         /* client additional sb info */
1054         lsi->lsi_llsbi = sbi = ll_init_sbi();
1055         if (!sbi) {
1056                 cfs_module_put();
1057                 RETURN(-ENOMEM);
1058         }
1059
1060         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1061         if (err)
1062                 GOTO(out_free, err);
1063
1064         /* Generate a string unique to this super, in case some joker tries
1065            to mount the same fs at two mount points.
1066            Use the address of the super itself.*/
1067         sprintf(ll_instance, "%p", sb);
1068         cfg.cfg_instance = ll_instance;
1069         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1070
1071         /* set up client obds */
1072         err = lustre_process_log(sb, profilenm, &cfg);
1073         if (err < 0) {
1074                 CERROR("Unable to process log: %d\n", err);
1075                 GOTO(out_free, err);
1076         }
1077
1078         lprof = class_get_profile(profilenm);
1079         if (lprof == NULL) {
1080                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1081                                    " read from the MGS.  Does that filesystem "
1082                                    "exist?\n", profilenm);
1083                 GOTO(out_free, err = -EINVAL);
1084         }
1085         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1086                lprof->lp_md, lprof->lp_dt);
1087
1088         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1089                   strlen(ll_instance) + 2);
1090         if (!dt)
1091                 GOTO(out_free, err = -ENOMEM);
1092         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1093
1094         OBD_ALLOC(md, strlen(lprof->lp_md) +
1095                   strlen(ll_instance) + 2);
1096         if (!md)
1097                 GOTO(out_free, err = -ENOMEM);
1098         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1099
1100         /* connections, registrations, sb setup */
1101         err = client_common_fill_super(sb, md, dt);
1102
1103 out_free:
1104         if (md)
1105                 OBD_FREE(md, strlen(md) + 1);
1106         if (dt)
1107                 OBD_FREE(dt, strlen(dt) + 1);
1108         if (err)
1109                 ll_put_super(sb);
1110         else
1111                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1112
1113         RETURN(err);
1114 } /* ll_fill_super */
1115
1116
1117 void ll_put_super(struct super_block *sb)
1118 {
1119         struct config_llog_instance cfg;
1120         char   ll_instance[sizeof(sb) * 2 + 1];
1121         struct obd_device *obd;
1122         struct lustre_sb_info *lsi = s2lsi(sb);
1123         struct ll_sb_info *sbi = ll_s2sbi(sb);
1124         char *profilenm = get_profile_name(sb);
1125         int force = 1, next;
1126         ENTRY;
1127
1128         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1129
1130         ll_print_capa_stat(sbi);
1131
1132         sprintf(ll_instance, "%p", sb);
1133         cfg.cfg_instance = ll_instance;
1134         lustre_end_log(sb, NULL, &cfg);
1135
1136         if (sbi->ll_md_exp) {
1137                 obd = class_exp2obd(sbi->ll_md_exp);
1138                 if (obd)
1139                         force = obd->obd_force;
1140         }
1141
1142         /* We need to set force before the lov_disconnect in
1143            lustre_common_put_super, since l_d cleans up osc's as well. */
1144         if (force) {
1145                 next = 0;
1146                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1147                                                      &next)) != NULL) {
1148                         obd->obd_force = force;
1149                 }
1150         }
1151
1152         if (sbi->ll_lcq) {
1153                 /* Only if client_common_fill_super succeeded */
1154                 client_common_put_super(sb);
1155         }
1156         next = 0;
1157         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1158                 class_manual_cleanup(obd);
1159         }
1160
1161         if (profilenm)
1162                 class_del_profile(profilenm);
1163
1164         ll_free_sbi(sb);
1165         lsi->lsi_llsbi = NULL;
1166
1167         lustre_common_put_super(sb);
1168
1169         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1170
1171         cfs_module_put();
1172
1173         EXIT;
1174 } /* client_put_super */
1175
1176 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1177
1178 #if defined(HAVE_CACHE_RETURN_INT)
1179 static int
1180 #else
1181 static void
1182 #endif
1183 ll_shrink_cache(int priority, unsigned int gfp_mask)
1184 {
1185         struct ll_sb_info *sbi;
1186         int count = 0;
1187
1188         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1189                 count += llap_shrink_cache(sbi, priority);
1190
1191 #if defined(HAVE_CACHE_RETURN_INT)
1192         return count;
1193 #endif
1194 }
1195
1196 struct cache_definition ll_cache_definition = {
1197         .name = "llap_cache",
1198         .shrink = ll_shrink_cache
1199 };
1200 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1201
1202 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1203 {
1204         struct inode *inode = NULL;
1205         /* NOTE: we depend on atomic igrab() -bzzz */
1206         lock_res_and_lock(lock);
1207         if (lock->l_ast_data) {
1208                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1209                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1210                         inode = igrab(lock->l_ast_data);
1211                 } else {
1212                         inode = lock->l_ast_data;
1213                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1214                                                 D_INFO : D_WARNING,
1215                                         lock, __FILE__, __func__, __LINE__,
1216                                         "l_ast_data %p is bogus: magic %08x",
1217                                         lock->l_ast_data, lli->lli_inode_magic);
1218                         inode = NULL;
1219                 }
1220         }
1221         unlock_res_and_lock(lock);
1222         return inode;
1223 }
1224
1225 static int null_if_equal(struct ldlm_lock *lock, void *data)
1226 {
1227         if (data == lock->l_ast_data) {
1228                 lock->l_ast_data = NULL;
1229
1230                 if (lock->l_req_mode != lock->l_granted_mode)
1231                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1232         }
1233
1234         return LDLM_ITER_CONTINUE;
1235 }
1236
1237 void ll_clear_inode(struct inode *inode)
1238 {
1239         struct ll_inode_info *lli = ll_i2info(inode);
1240         struct ll_sb_info *sbi = ll_i2sbi(inode);
1241         ENTRY;
1242
1243         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1244                inode->i_generation, inode);
1245
1246         if (S_ISDIR(inode->i_mode)) {
1247                 /* these should have been cleared in ll_file_release */
1248                 LASSERT(lli->lli_sai == NULL);
1249                 LASSERT(lli->lli_opendir_key == NULL);
1250                 LASSERT(lli->lli_opendir_pid == 0);
1251         }
1252
1253         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1254         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1255                          null_if_equal, inode);
1256
1257         LASSERT(!lli->lli_open_fd_write_count);
1258         LASSERT(!lli->lli_open_fd_read_count);
1259         LASSERT(!lli->lli_open_fd_exec_count);
1260
1261         if (lli->lli_mds_write_och)
1262                 ll_md_real_close(inode, FMODE_WRITE);
1263         if (lli->lli_mds_exec_och)
1264                 ll_md_real_close(inode, FMODE_EXEC);
1265         if (lli->lli_mds_read_och)
1266                 ll_md_real_close(inode, FMODE_READ);
1267
1268         if (lli->lli_smd) {
1269                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1270                                   null_if_equal, inode);
1271
1272                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1273                 lli->lli_smd = NULL;
1274         }
1275
1276         if (lli->lli_symlink_name) {
1277                 OBD_FREE(lli->lli_symlink_name,
1278                          strlen(lli->lli_symlink_name) + 1);
1279                 lli->lli_symlink_name = NULL;
1280         }
1281
1282         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1283                 LASSERT(lli->lli_posix_acl == NULL);
1284                 if (lli->lli_remote_perms) {
1285                         free_rmtperm_hash(lli->lli_remote_perms);
1286                         lli->lli_remote_perms = NULL;
1287                 }
1288         }
1289 #ifdef CONFIG_FS_POSIX_ACL
1290         else if (lli->lli_posix_acl) {
1291                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1292                 LASSERT(lli->lli_remote_perms == NULL);
1293                 posix_acl_release(lli->lli_posix_acl);
1294                 lli->lli_posix_acl = NULL;
1295         }
1296 #endif
1297         lli->lli_inode_magic = LLI_INODE_DEAD;
1298
1299 #ifdef HAVE_EXPORT___IGET
1300         spin_lock(&sbi->ll_deathrow_lock);
1301         list_del_init(&lli->lli_dead_list);
1302         spin_unlock(&sbi->ll_deathrow_lock);
1303 #endif
1304         ll_clear_inode_capas(inode);
1305
1306         EXIT;
1307 }
1308
1309 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1310                   struct md_open_data **mod)
1311 {
1312         struct lustre_md md;
1313         struct ll_sb_info *sbi = ll_i2sbi(inode);
1314         struct ptlrpc_request *request = NULL;
1315         int rc;
1316         ENTRY;
1317
1318         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1319                                      LUSTRE_OPC_ANY, NULL);
1320         if (IS_ERR(op_data))
1321                 RETURN(PTR_ERR(op_data));
1322
1323         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1324                         &request, mod);
1325         if (rc) {
1326                 ptlrpc_req_finished(request);
1327                 if (rc == -ENOENT) {
1328                         inode->i_nlink = 0;
1329                         /* Unlinked special device node? Or just a race?
1330                          * Pretend we done everything. */
1331                         if (!S_ISREG(inode->i_mode) &&
1332                             !S_ISDIR(inode->i_mode))
1333                                 rc = inode_setattr(inode, &op_data->op_attr);
1334                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1335                         CERROR("md_setattr fails: rc = %d\n", rc);
1336                 }
1337                 RETURN(rc);
1338         }
1339
1340         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1341                               sbi->ll_md_exp, &md);
1342         if (rc) {
1343                 ptlrpc_req_finished(request);
1344                 RETURN(rc);
1345         }
1346
1347         /* We call inode_setattr to adjust timestamps.
1348          * If there is at least some data in file, we cleared ATTR_SIZE
1349          * above to avoid invoking vmtruncate, otherwise it is important
1350          * to call vmtruncate in inode_setattr to update inode->i_size
1351          * (bug 6196) */
1352         rc = inode_setattr(inode, &op_data->op_attr);
1353
1354         /* Extract epoch data if obtained. */
1355         op_data->op_handle = md.body->handle;
1356         op_data->op_ioepoch = md.body->ioepoch;
1357
1358         ll_update_inode(inode, &md);
1359         ptlrpc_req_finished(request);
1360
1361         RETURN(rc);
1362 }
1363
1364 /* Close IO epoch and send Size-on-MDS attribute update. */
1365 static int ll_setattr_done_writing(struct inode *inode,
1366                                    struct md_op_data *op_data,
1367                                    struct md_open_data *mod)
1368 {
1369         struct ll_inode_info *lli = ll_i2info(inode);
1370         int rc = 0;
1371         ENTRY;
1372
1373         LASSERT(op_data != NULL);
1374         if (!S_ISREG(inode->i_mode))
1375                 RETURN(0);
1376
1377         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1378                op_data->op_ioepoch, PFID(&lli->lli_fid));
1379
1380         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1381         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1382         if (rc == -EAGAIN) {
1383                 /* MDS has instructed us to obtain Size-on-MDS attribute
1384                  * from OSTs and send setattr to back to MDS. */
1385                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1386                                          op_data->op_ioepoch);
1387         } else if (rc) {
1388                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1389                        inode->i_ino, rc);
1390         }
1391         RETURN(rc);
1392 }
1393
1394 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1395 {
1396         struct ll_sb_info *sbi = ll_i2sbi(inode);
1397         struct ll_inode_info *lli = ll_i2info(inode);
1398         struct lov_stripe_md *lsm = lli->lli_smd;
1399         int rc;
1400         ldlm_policy_data_t policy = { .l_extent = {new_size,
1401                                                    OBD_OBJECT_EOF } };
1402         struct lustre_handle lockh = { 0 };
1403         int local_lock = 0; /* 0 - no local lock;
1404                              * 1 - lock taken by lock_extent;
1405                              * 2 - by obd_match*/
1406         int ast_flags;
1407         int err;
1408         ENTRY;
1409
1410         UNLOCK_INODE_MUTEX(inode);
1411         UP_WRITE_I_ALLOC_SEM(inode);
1412
1413         if (sbi->ll_lockless_truncate_enable &&
1414             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1415                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1416                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1417                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1418                 if (rc > 0) {
1419                         local_lock = 2;
1420                         rc = 0;
1421                 } else if (rc == 0) {
1422                         rc = ll_file_punch(inode, new_size, 1);
1423                 }
1424         } else {
1425                 /* XXX when we fix the AST intents to pass the discard-range
1426                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1427                  * XXX here. */
1428                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1429                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1430                                     &lockh, ast_flags);
1431                 if (likely(rc == 0))
1432                         local_lock = 1;
1433         }
1434
1435         LOCK_INODE_MUTEX(inode);
1436         DOWN_WRITE_I_ALLOC_SEM(inode);
1437
1438         if (likely(rc == 0)) {
1439                 /* Only ll_inode_size_lock is taken at this level.
1440                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1441                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1442                  * ll_truncate dropped ll_inode_size_lock() */
1443                 ll_inode_size_lock(inode, 0);
1444                 if (!local_lock) {
1445                         spin_lock(&lli->lli_lock);
1446                         lli->lli_flags |= LLIF_SRVLOCK;
1447                         spin_unlock(&lli->lli_lock);
1448                 }
1449                 rc = vmtruncate(inode, new_size);
1450                 if (!local_lock) {
1451                         spin_lock(&lli->lli_lock);
1452                         lli->lli_flags &= ~LLIF_SRVLOCK;
1453                         spin_unlock(&lli->lli_lock);
1454                 }
1455                 if (rc != 0) {
1456                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1457                         ll_inode_size_unlock(inode, 0);
1458                 }
1459         }
1460
1461         if (local_lock) {
1462                 if (local_lock == 2)
1463                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1464                 else
1465                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1466                 if (unlikely(err != 0)){
1467                         CERROR("extent unlock failed: err=%d,"
1468                                " unlock method =%d\n", err, local_lock);
1469                         if (rc == 0)
1470                                 rc = err;
1471                 }
1472         }
1473         RETURN(rc);
1474 }
1475
1476 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1477  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1478  * keep these values until such a time that objects are allocated for it.
1479  * We do the MDS operations first, as it is checking permissions for us.
1480  * We don't to the MDS RPC if there is nothing that we want to store there,
1481  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1482  * going to do an RPC anyways.
1483  *
1484  * If we are doing a truncate, we will send the mtime and ctime updates
1485  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1486  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1487  * at the same time.
1488  */
1489 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1490 {
1491         struct ll_inode_info *lli = ll_i2info(inode);
1492         struct lov_stripe_md *lsm = lli->lli_smd;
1493         struct ll_sb_info *sbi = ll_i2sbi(inode);
1494         struct md_op_data *op_data = NULL;
1495         struct md_open_data *mod = NULL;
1496         int ia_valid = attr->ia_valid;
1497         int rc = 0, rc1 = 0;
1498         ENTRY;
1499
1500         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1501                attr->ia_valid);
1502         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1503
1504         if (ia_valid & ATTR_SIZE) {
1505                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1506                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1507                                attr->ia_size, ll_file_maxbytes(inode));
1508                         RETURN(-EFBIG);
1509                 }
1510
1511                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1512         }
1513
1514         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1515         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1516                 if (current->fsuid != inode->i_uid &&
1517                     !cfs_capable(CFS_CAP_FOWNER))
1518                         RETURN(-EPERM);
1519         }
1520
1521         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1522         if (attr->ia_valid & ATTR_CTIME) {
1523                 attr->ia_ctime = CURRENT_TIME;
1524                 attr->ia_valid |= ATTR_CTIME_SET;
1525         }
1526         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1527                 attr->ia_atime = CURRENT_TIME;
1528                 attr->ia_valid |= ATTR_ATIME_SET;
1529         }
1530         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1531                 attr->ia_mtime = CURRENT_TIME;
1532                 attr->ia_valid |= ATTR_MTIME_SET;
1533         }
1534         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1535                 /* To avoid stale mtime on mds, obtain it from ost and send
1536                    to mds. */
1537                 rc = ll_glimpse_size(inode, 0);
1538                 if (rc)
1539                         RETURN(rc);
1540
1541                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1542                 attr->ia_mtime = inode->i_mtime;
1543         }
1544
1545         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1546                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1547                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1548                        cfs_time_current_sec());
1549
1550         /* NB: ATTR_SIZE will only be set after this point if the size
1551          * resides on the MDS, ie, this file has no objects. */
1552         if (lsm)
1553                 attr->ia_valid &= ~ATTR_SIZE;
1554
1555         /* We always do an MDS RPC, even if we're only changing the size;
1556          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1557
1558         OBD_ALLOC_PTR(op_data);
1559         if (op_data == NULL)
1560                 RETURN(-ENOMEM);
1561
1562         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1563
1564         /* Open epoch for truncate. */
1565         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1566             (ia_valid & ATTR_SIZE))
1567                 op_data->op_flags = MF_EPOCH_OPEN;
1568
1569         rc = ll_md_setattr(inode, op_data, &mod);
1570         if (rc)
1571                 GOTO(out, rc);
1572
1573         if (op_data->op_ioepoch)
1574                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1575                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1576
1577         if (!lsm || !S_ISREG(inode->i_mode)) {
1578                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1579                 GOTO(out, rc = 0);
1580         }
1581
1582         /* We really need to get our PW lock before we change inode->i_size.
1583          * If we don't we can race with other i_size updaters on our node, like
1584          * ll_file_read.  We can also race with i_size propogation to other
1585          * nodes through dirtying and writeback of final cached pages.  This
1586          * last one is especially bad for racing o_append users on other
1587          * nodes. */
1588         if (ia_valid & ATTR_SIZE) {
1589                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1590         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1591                 obd_flag flags;
1592                 struct obd_info oinfo = { { { 0 } } };
1593                 struct obdo *oa;
1594
1595                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1596                        inode->i_ino, LTIME_S(attr->ia_mtime));
1597
1598                 OBDO_ALLOC(oa);
1599                 if (oa) {
1600                         oa->o_id = lsm->lsm_object_id;
1601                         oa->o_gr = lsm->lsm_object_gr;
1602                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1603
1604                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1605                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1606                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1607                                 OBD_MD_FLGROUP;
1608
1609                         obdo_from_inode(oa, inode, flags);
1610
1611                         oinfo.oi_oa = oa;
1612                         oinfo.oi_md = lsm;
1613                         oinfo.oi_capa = ll_mdscapa_get(inode);
1614
1615                         /* XXX: this looks unnecessary now. */
1616                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1617                         capa_put(oinfo.oi_capa);
1618                         if (rc)
1619                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1620                         OBDO_FREE(oa);
1621                 } else {
1622                         rc = -ENOMEM;
1623                 }
1624         }
1625         EXIT;
1626 out:
1627         if (op_data) {
1628                 if (op_data->op_ioepoch)
1629                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1630                 ll_finish_md_op_data(op_data);
1631         }
1632         return rc ? rc : rc1;
1633 }
1634
1635 int ll_setattr(struct dentry *de, struct iattr *attr)
1636 {
1637         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1638             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1639                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1640
1641         if ((de->d_inode->i_mode & S_ISUID) &&
1642             !(attr->ia_mode & S_ISUID) &&
1643             !(attr->ia_valid & ATTR_KILL_SUID))
1644                 attr->ia_valid |= ATTR_KILL_SUID;
1645
1646         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1647             !(attr->ia_mode & S_ISGID) &&
1648             !(attr->ia_valid & ATTR_KILL_SGID))
1649                 attr->ia_valid |= ATTR_KILL_SGID;
1650
1651         return ll_setattr_raw(de->d_inode, attr);
1652 }
1653
1654 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1655                        __u64 max_age, __u32 flags)
1656 {
1657         struct ll_sb_info *sbi = ll_s2sbi(sb);
1658         struct obd_statfs obd_osfs;
1659         int rc;
1660         ENTRY;
1661
1662         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1663         if (rc) {
1664                 CERROR("md_statfs fails: rc = %d\n", rc);
1665                 RETURN(rc);
1666         }
1667
1668         osfs->os_type = sb->s_magic;
1669
1670         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1671                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1672
1673         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1674                               &obd_osfs, max_age, flags);
1675         if (rc) {
1676                 CERROR("obd_statfs fails: rc = %d\n", rc);
1677                 RETURN(rc);
1678         }
1679
1680         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1681                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1682                obd_osfs.os_files);
1683
1684         osfs->os_bsize = obd_osfs.os_bsize;
1685         osfs->os_blocks = obd_osfs.os_blocks;
1686         osfs->os_bfree = obd_osfs.os_bfree;
1687         osfs->os_bavail = obd_osfs.os_bavail;
1688
1689         /* If we don't have as many objects free on the OST as inodes
1690          * on the MDS, we reduce the total number of inodes to
1691          * compensate, so that the "inodes in use" number is correct.
1692          */
1693         if (obd_osfs.os_ffree < osfs->os_ffree) {
1694                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1695                         obd_osfs.os_ffree;
1696                 osfs->os_ffree = obd_osfs.os_ffree;
1697         }
1698
1699         RETURN(rc);
1700 }
1701 #ifndef HAVE_STATFS_DENTRY_PARAM
1702 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1703 {
1704 #else
1705 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1706 {
1707         struct super_block *sb = de->d_sb;
1708 #endif
1709         struct obd_statfs osfs;
1710         int rc;
1711
1712         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1713         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1714
1715         /* For now we will always get up-to-date statfs values, but in the
1716          * future we may allow some amount of caching on the client (e.g.
1717          * from QOS or lprocfs updates). */
1718         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1719         if (rc)
1720                 return rc;
1721
1722         statfs_unpack(sfs, &osfs);
1723
1724         /* We need to downshift for all 32-bit kernels, because we can't
1725          * tell if the kernel is being called via sys_statfs64() or not.
1726          * Stop before overflowing f_bsize - in which case it is better
1727          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1728         if (sizeof(long) < 8) {
1729                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1730                         sfs->f_bsize <<= 1;
1731
1732                         osfs.os_blocks >>= 1;
1733                         osfs.os_bfree >>= 1;
1734                         osfs.os_bavail >>= 1;
1735                 }
1736         }
1737
1738         sfs->f_blocks = osfs.os_blocks;
1739         sfs->f_bfree = osfs.os_bfree;
1740         sfs->f_bavail = osfs.os_bavail;
1741
1742         return 0;
1743 }
1744
1745 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1746 {
1747         struct ll_inode_info *lli;
1748         struct lov_stripe_md *lsm;
1749
1750         lli = ll_i2info(inode);
1751         LASSERT(lli->lli_size_sem_owner != current);
1752         down(&lli->lli_size_sem);
1753         LASSERT(lli->lli_size_sem_owner == NULL);
1754         lli->lli_size_sem_owner = current;
1755         lsm = lli->lli_smd;
1756         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1757                  lsm, lock_lsm);
1758         if (lock_lsm)
1759                 lov_stripe_lock(lsm);
1760 }
1761
1762 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1763 {
1764         struct ll_inode_info *lli;
1765         struct lov_stripe_md *lsm;
1766
1767         lli = ll_i2info(inode);
1768         lsm = lli->lli_smd;
1769         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1770                  lsm, unlock_lsm);
1771         if (unlock_lsm)
1772                 lov_stripe_unlock(lsm);
1773         LASSERT(lli->lli_size_sem_owner == current);
1774         lli->lli_size_sem_owner = NULL;
1775         up(&lli->lli_size_sem);
1776 }
1777
1778 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1779 {
1780         struct ll_inode_info *lli = ll_i2info(inode);
1781
1782         dump_lsm(D_INODE, lsm);
1783         dump_lsm(D_INODE, lli->lli_smd);
1784         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1785                  "lsm must be joined lsm %p\n", lsm);
1786         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1787         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1788                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1789         lli->lli_smd = lsm;
1790         lli->lli_maxbytes = lsm->lsm_maxbytes;
1791         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1792                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1793 }
1794
1795 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1796 {
1797         struct ll_inode_info *lli = ll_i2info(inode);
1798         struct mdt_body *body = md->body;
1799         struct lov_stripe_md *lsm = md->lsm;
1800         struct ll_sb_info *sbi = ll_i2sbi(inode);
1801
1802         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1803         if (lsm != NULL) {
1804                 if (lli->lli_smd == NULL) {
1805                         if (lsm->lsm_magic != LOV_MAGIC &&
1806                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1807                                 dump_lsm(D_ERROR, lsm);
1808                                 LBUG();
1809                         }
1810                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1811                                lsm, inode->i_ino, inode->i_generation, inode);
1812                         /* ll_inode_size_lock() requires it is only called
1813                          * with lli_smd != NULL or lock_lsm == 0 or we can
1814                          * race between lock/unlock.  bug 9547 */
1815                         lli->lli_smd = lsm;
1816                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1817                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1818                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1819                 } else {
1820                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1821                              lli->lli_smd->lsm_stripe_count ==
1822                                         lsm->lsm_stripe_count) {
1823                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1824                                         CERROR("lsm mismatch for inode %ld\n",
1825                                                 inode->i_ino);
1826                                         CERROR("lli_smd:\n");
1827                                         dump_lsm(D_ERROR, lli->lli_smd);
1828                                         CERROR("lsm:\n");
1829                                         dump_lsm(D_ERROR, lsm);
1830                                         LBUG();
1831                                 }
1832                         } else
1833                                 ll_replace_lsm(inode, lsm);
1834                 }
1835                 if (lli->lli_smd != lsm)
1836                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1837         }
1838
1839         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1840                 if (body->valid & OBD_MD_FLRMTPERM)
1841                         ll_update_remote_perm(inode, md->remote_perm);
1842         }
1843 #ifdef CONFIG_FS_POSIX_ACL
1844         else if (body->valid & OBD_MD_FLACL) {
1845                 spin_lock(&lli->lli_lock);
1846                 if (lli->lli_posix_acl)
1847                         posix_acl_release(lli->lli_posix_acl);
1848                 lli->lli_posix_acl = md->posix_acl;
1849                 spin_unlock(&lli->lli_lock);
1850         }
1851 #endif
1852         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1853
1854         if (body->valid & OBD_MD_FLATIME &&
1855             body->atime > LTIME_S(inode->i_atime))
1856                 LTIME_S(inode->i_atime) = body->atime;
1857
1858         /* mtime is always updated with ctime, but can be set in past.
1859            As write and utime(2) may happen within 1 second, and utime's
1860            mtime has a priority over write's one, so take mtime from mds
1861            for the same ctimes. */
1862         if (body->valid & OBD_MD_FLCTIME &&
1863             body->ctime >= LTIME_S(inode->i_ctime)) {
1864                 LTIME_S(inode->i_ctime) = body->ctime;
1865                 if (body->valid & OBD_MD_FLMTIME) {
1866                         CDEBUG(D_INODE, "setting ino %lu mtime "
1867                                "from %lu to "LPU64"\n", inode->i_ino,
1868                                LTIME_S(inode->i_mtime), body->mtime);
1869                         LTIME_S(inode->i_mtime) = body->mtime;
1870                 }
1871         }
1872         if (body->valid & OBD_MD_FLMODE)
1873                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1874         if (body->valid & OBD_MD_FLTYPE)
1875                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1876         if (S_ISREG(inode->i_mode)) {
1877                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1878         } else {
1879                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1880         }
1881 #ifdef HAVE_INODE_BLKSIZE
1882         inode->i_blksize = 1<<inode->i_blkbits;
1883 #endif
1884         if (body->valid & OBD_MD_FLUID)
1885                 inode->i_uid = body->uid;
1886         if (body->valid & OBD_MD_FLGID)
1887                 inode->i_gid = body->gid;
1888         if (body->valid & OBD_MD_FLFLAGS)
1889                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1890         if (body->valid & OBD_MD_FLNLINK)
1891                 inode->i_nlink = body->nlink;
1892         if (body->valid & OBD_MD_FLRDEV)
1893                 inode->i_rdev = old_decode_dev(body->rdev);
1894
1895         if (body->valid & OBD_MD_FLID) {
1896                 /* FID shouldn't be changed! */
1897                 if (fid_is_sane(&lli->lli_fid)) {
1898                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1899                                  "Trying to change FID "DFID
1900                                  " to the "DFID", inode %lu/%u(%p)\n",
1901                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1902                                  inode->i_ino, inode->i_generation, inode);
1903                 } else
1904                         lli->lli_fid = body->fid1;
1905         }
1906
1907         LASSERT(fid_seq(&lli->lli_fid) != 0);
1908
1909         if (body->valid & OBD_MD_FLSIZE) {
1910                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1911                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1912                         struct lustre_handle lockh;
1913                         ldlm_mode_t mode;
1914
1915                         /* As it is possible a blocking ast has been processed
1916                          * by this time, we need to check there is an UPDATE
1917                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1918                          * it. */
1919                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1920                                                &lockh);
1921                         if (mode) {
1922                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1923                                                       LLIF_EPOCH_PENDING |
1924                                                       LLIF_SOM_DIRTY)) {
1925                                         CERROR("ino %lu flags %lu still has "
1926                                                "size authority! do not trust "
1927                                                "the size got from MDS\n",
1928                                                inode->i_ino, lli->lli_flags);
1929                                 } else {
1930                                         /* Use old size assignment to avoid
1931                                          * deadlock bz14138 & bz14326 */
1932                                         inode->i_size = body->size;
1933                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1934                                 }
1935                                 ldlm_lock_decref(&lockh, mode);
1936                         }
1937                 } else {
1938                         /* Use old size assignment to avoid
1939                          * deadlock bz14138 & bz14326 */
1940                         inode->i_size = body->size;
1941                 }
1942
1943                 if (body->valid & OBD_MD_FLBLOCKS)
1944                         inode->i_blocks = body->blocks;
1945         }
1946
1947         if (body->valid & OBD_MD_FLMDSCAPA) {
1948                 LASSERT(md->mds_capa);
1949                 ll_add_capa(inode, md->mds_capa);
1950         }
1951         if (body->valid & OBD_MD_FLOSSCAPA) {
1952                 LASSERT(md->oss_capa);
1953                 ll_add_capa(inode, md->oss_capa);
1954         }
1955 }
1956
1957 static struct backing_dev_info ll_backing_dev_info = {
1958         .ra_pages       = 0,    /* No readahead */
1959 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1960         .capabilities   = 0,    /* Does contribute to dirty memory */
1961 #else
1962         .memory_backed  = 0,    /* Does contribute to dirty memory */
1963 #endif
1964 };
1965
1966 void ll_read_inode2(struct inode *inode, void *opaque)
1967 {
1968         struct lustre_md *md = opaque;
1969         struct ll_inode_info *lli = ll_i2info(inode);
1970         ENTRY;
1971
1972         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1973                inode->i_ino, inode->i_generation, inode);
1974
1975         ll_lli_init(lli);
1976
1977         LASSERT(!lli->lli_smd);
1978
1979         /* Core attributes from the MDS first.  This is a new inode, and
1980          * the VFS doesn't zero times in the core inode so we have to do
1981          * it ourselves.  They will be overwritten by either MDS or OST
1982          * attributes - we just need to make sure they aren't newer. */
1983         LTIME_S(inode->i_mtime) = 0;
1984         LTIME_S(inode->i_atime) = 0;
1985         LTIME_S(inode->i_ctime) = 0;
1986         inode->i_rdev = 0;
1987         ll_update_inode(inode, md);
1988
1989         /* OIDEBUG(inode); */
1990
1991         if (S_ISREG(inode->i_mode)) {
1992                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1993                 inode->i_op = &ll_file_inode_operations;
1994                 inode->i_fop = sbi->ll_fop;
1995                 inode->i_mapping->a_ops = &ll_aops;
1996                 EXIT;
1997         } else if (S_ISDIR(inode->i_mode)) {
1998                 inode->i_op = &ll_dir_inode_operations;
1999                 inode->i_fop = &ll_dir_operations;
2000                 inode->i_mapping->a_ops = &ll_dir_aops;
2001                 EXIT;
2002         } else if (S_ISLNK(inode->i_mode)) {
2003                 inode->i_op = &ll_fast_symlink_inode_operations;
2004                 EXIT;
2005         } else {
2006                 inode->i_op = &ll_special_inode_operations;
2007
2008                 init_special_inode(inode, inode->i_mode,
2009                                    kdev_t_to_nr(inode->i_rdev));
2010
2011                 /* initializing backing dev info. */
2012                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2013
2014                 EXIT;
2015         }
2016 }
2017
2018 void ll_delete_inode(struct inode *inode)
2019 {
2020         struct ll_sb_info *sbi = ll_i2sbi(inode);
2021         int rc;
2022         ENTRY;
2023
2024         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2025         if (rc) {
2026                 CERROR("fid_delete() failed, rc %d\n", rc);
2027         }
2028         truncate_inode_pages(&inode->i_data, 0);
2029         clear_inode(inode);
2030
2031         EXIT;
2032 }
2033
2034 int ll_iocontrol(struct inode *inode, struct file *file,
2035                  unsigned int cmd, unsigned long arg)
2036 {
2037         struct ll_sb_info *sbi = ll_i2sbi(inode);
2038         struct ptlrpc_request *req = NULL;
2039         int rc, flags = 0;
2040         ENTRY;
2041
2042         switch(cmd) {
2043         case EXT3_IOC_GETFLAGS: {
2044                 struct mdt_body *body;
2045                 struct obd_capa *oc;
2046
2047                 oc = ll_mdscapa_get(inode);
2048                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2049                                 OBD_MD_FLFLAGS, 0, &req);
2050                 capa_put(oc);
2051                 if (rc) {
2052                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2053                         RETURN(-abs(rc));
2054                 }
2055
2056                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2057
2058                 flags = body->flags;
2059
2060                 ptlrpc_req_finished(req);
2061
2062                 RETURN(put_user(flags, (int *)arg));
2063         }
2064         case EXT3_IOC_SETFLAGS: {
2065                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2066                 struct obd_info oinfo = { { { 0 } } };
2067                 struct md_op_data *op_data;
2068
2069                 if (get_user(flags, (int *)arg))
2070                         RETURN(-EFAULT);
2071
2072                 oinfo.oi_md = lsm;
2073                 OBDO_ALLOC(oinfo.oi_oa);
2074                 if (!oinfo.oi_oa)
2075                         RETURN(-ENOMEM);
2076
2077                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2078                                              LUSTRE_OPC_ANY, NULL);
2079                 if (IS_ERR(op_data))
2080                         RETURN(PTR_ERR(op_data));
2081
2082                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2083                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2084                 rc = md_setattr(sbi->ll_md_exp, op_data,
2085                                 NULL, 0, NULL, 0, &req, NULL);
2086                 ll_finish_md_op_data(op_data);
2087                 ptlrpc_req_finished(req);
2088                 if (rc || lsm == NULL) {
2089                         OBDO_FREE(oinfo.oi_oa);
2090                         RETURN(rc);
2091                 }
2092
2093                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2094                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2095                 oinfo.oi_oa->o_flags = flags;
2096                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2097                                        OBD_MD_FLGROUP;
2098                 oinfo.oi_capa = ll_mdscapa_get(inode);
2099
2100                 obdo_from_inode(oinfo.oi_oa, inode,
2101                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2102                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2103                 capa_put(oinfo.oi_capa);
2104                 OBDO_FREE(oinfo.oi_oa);
2105                 if (rc) {
2106                         if (rc != -EPERM && rc != -EACCES)
2107                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2108                         RETURN(rc);
2109                 }
2110
2111                 inode->i_flags = ll_ext_to_inode_flags(flags |
2112                                                        MDS_BFLAG_EXT_FLAGS);
2113                 RETURN(0);
2114         }
2115         default:
2116                 RETURN(-ENOSYS);
2117         }
2118
2119         RETURN(0);
2120 }
2121
2122 int ll_flush_ctx(struct inode *inode)
2123 {
2124         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2125
2126         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2127
2128         obd_set_info_async(sbi->ll_md_exp,
2129                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2130                            0, NULL, NULL);
2131         obd_set_info_async(sbi->ll_dt_exp,
2132                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2133                            0, NULL, NULL);
2134         return 0;
2135 }
2136
2137 /* umount -f client means force down, don't save state */
2138 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2139 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2140 {
2141         struct super_block *sb = vfsmnt->mnt_sb;
2142 #else
2143 void ll_umount_begin(struct super_block *sb)
2144 {
2145 #endif
2146         struct lustre_sb_info *lsi = s2lsi(sb);
2147         struct ll_sb_info *sbi = ll_s2sbi(sb);
2148         struct obd_device *obd;
2149         struct obd_ioctl_data ioc_data = { 0 };
2150         ENTRY;
2151
2152 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2153         if (!(flags & MNT_FORCE)) {
2154                 EXIT;
2155                 return;
2156         }
2157 #endif
2158
2159         /* Tell the MGC we got umount -f */
2160         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2161
2162         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2163                sb->s_count, atomic_read(&sb->s_active));
2164
2165         obd = class_exp2obd(sbi->ll_md_exp);
2166         if (obd == NULL) {
2167                 CERROR("Invalid MDC connection handle "LPX64"\n",
2168                        sbi->ll_md_exp->exp_handle.h_cookie);
2169                 EXIT;
2170                 return;
2171         }
2172         obd->obd_force = 1;
2173         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2174                       &ioc_data, NULL);
2175
2176         obd = class_exp2obd(sbi->ll_dt_exp);
2177         if (obd == NULL) {
2178                 CERROR("Invalid LOV connection handle "LPX64"\n",
2179                        sbi->ll_dt_exp->exp_handle.h_cookie);
2180                 EXIT;
2181                 return;
2182         }
2183
2184         obd->obd_force = 1;
2185         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2186                       &ioc_data, NULL);
2187
2188         /* Really, we'd like to wait until there are no requests outstanding,
2189          * and then continue.  For now, we just invalidate the requests,
2190          * schedule, and hope.
2191          */
2192         schedule();
2193
2194         EXIT;
2195 }
2196
2197 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2198 {
2199         struct ll_sb_info *sbi = ll_s2sbi(sb);
2200         int err;
2201         __u32 read_only;
2202
2203         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2204                 read_only = *flags & MS_RDONLY;
2205                 err = obd_set_info_async(sbi->ll_md_exp,
2206                                          sizeof(KEY_READ_ONLY),
2207                                          KEY_READ_ONLY, sizeof(read_only),
2208                                          &read_only, NULL);
2209                 if (err) {
2210                         CERROR("Failed to change the read-only flag during "
2211                                "remount: %d\n", err);
2212                         return err;
2213                 }
2214
2215                 if (read_only)
2216                         sb->s_flags |= MS_RDONLY;
2217                 else
2218                         sb->s_flags &= ~MS_RDONLY;
2219         }
2220         return 0;
2221 }
2222
2223 int ll_prep_inode(struct inode **inode,
2224                   struct ptlrpc_request *req,
2225                   struct super_block *sb)
2226 {
2227         struct ll_sb_info *sbi = NULL;
2228         struct lustre_md md;
2229         int rc = 0;
2230         ENTRY;
2231
2232         LASSERT(*inode || sb);
2233         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2234         prune_deathrow(sbi, 1);
2235         memset(&md, 0, sizeof(struct lustre_md));
2236
2237         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2238                               sbi->ll_md_exp, &md);
2239         if (rc)
2240                 RETURN(rc);
2241
2242         if (*inode) {
2243                 ll_update_inode(*inode, &md);
2244         } else {
2245                 LASSERT(sb != NULL);
2246
2247                 /*
2248                  * At this point server returns to client's same fid as client
2249                  * generated for creating. So using ->fid1 is okay here.
2250                  */
2251                 LASSERT(fid_is_sane(&md.body->fid1));
2252
2253                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2254                 if (*inode == NULL || is_bad_inode(*inode)) {
2255                         if (md.lsm)
2256                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2257 #ifdef CONFIG_FS_POSIX_ACL
2258                         if (md.posix_acl) {
2259                                 posix_acl_release(md.posix_acl);
2260                                 md.posix_acl = NULL;
2261                         }
2262 #endif
2263                         rc = -ENOMEM;
2264                         CERROR("new_inode -fatal: rc %d\n", rc);
2265                         GOTO(out, rc);
2266                 }
2267         }
2268
2269         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2270                          ll_i2info(*inode)->lli_smd);
2271 out:
2272         md_free_lustre_md(sbi->ll_md_exp, &md);
2273         RETURN(rc);
2274 }
2275
2276 char *llap_origins[] = {
2277         [LLAP_ORIGIN_UNKNOWN] = "--",
2278         [LLAP_ORIGIN_READPAGE] = "rp",
2279         [LLAP_ORIGIN_READAHEAD] = "ra",
2280         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2281         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2282         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2283 };
2284
2285 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2286                                              struct list_head *list)
2287 {
2288         struct ll_async_page *llap;
2289         struct list_head *pos;
2290
2291         list_for_each(pos, list) {
2292                 if (pos == head)
2293                         return NULL;
2294                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2295                 if (llap->llap_page == NULL)
2296                         continue;
2297                 return llap;
2298         }
2299         LBUG();
2300         return NULL;
2301 }
2302
2303 int ll_obd_statfs(struct inode *inode, void *arg)
2304 {
2305         struct ll_sb_info *sbi = NULL;
2306         struct obd_export *exp;
2307         char *buf = NULL;
2308         struct obd_ioctl_data *data = NULL;
2309         __u32 type;
2310         int len = 0, rc;
2311
2312         if (!inode || !(sbi = ll_i2sbi(inode)))
2313                 GOTO(out_statfs, rc = -EINVAL);
2314
2315         rc = obd_ioctl_getdata(&buf, &len, arg);
2316         if (rc)
2317                 GOTO(out_statfs, rc);
2318
2319         data = (void*)buf;
2320         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2321             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2322                 GOTO(out_statfs, rc = -EINVAL);
2323
2324         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2325         if (type == LL_STATFS_MDC)
2326                 exp = sbi->ll_md_exp;
2327         else if (type == LL_STATFS_LOV)
2328                 exp = sbi->ll_dt_exp;
2329         else
2330                 GOTO(out_statfs, rc = -ENODEV);
2331
2332         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2333         if (rc)
2334                 GOTO(out_statfs, rc);
2335 out_statfs:
2336         if (buf)
2337                 obd_ioctl_freedata(buf, len);
2338         return rc;
2339 }
2340
2341 int ll_process_config(struct lustre_cfg *lcfg)
2342 {
2343         char *ptr;
2344         void *sb;
2345         struct lprocfs_static_vars lvars;
2346         unsigned long x;
2347         int rc = 0;
2348
2349         lprocfs_llite_init_vars(&lvars);
2350
2351         /* The instance name contains the sb: lustre-client-aacfe000 */
2352         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2353         if (!ptr || !*(++ptr))
2354                 return -EINVAL;
2355         if (sscanf(ptr, "%lx", &x) != 1)
2356                 return -EINVAL;
2357         sb = (void *)x;
2358         /* This better be a real Lustre superblock! */
2359         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2360
2361         /* Note we have not called client_common_fill_super yet, so
2362            proc fns must be able to handle that! */
2363         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2364                                       lcfg, sb);
2365         return(rc);
2366 }
2367
2368 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2369 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2370                                        struct inode *i1, struct inode *i2,
2371                                        const char *name, int namelen,
2372                                        int mode, __u32 opc, void *data)
2373 {
2374         LASSERT(i1 != NULL);
2375
2376         if (namelen > ll_i2sbi(i1)->ll_namelen)
2377                 return ERR_PTR(-ENAMETOOLONG);
2378
2379         if (op_data == NULL)
2380                 OBD_ALLOC_PTR(op_data);
2381
2382         if (op_data == NULL)
2383                 return ERR_PTR(-ENOMEM);
2384
2385         ll_i2gids(op_data->op_suppgids, i1, i2);
2386         op_data->op_fid1 = *ll_inode2fid(i1);
2387         op_data->op_capa1 = ll_mdscapa_get(i1);
2388
2389         if (i2) {
2390                 op_data->op_fid2 = *ll_inode2fid(i2);
2391                 op_data->op_capa2 = ll_mdscapa_get(i2);
2392         } else {
2393                 fid_zero(&op_data->op_fid2);
2394                 op_data->op_capa2 = NULL;
2395         }
2396
2397         op_data->op_name = name;
2398         op_data->op_namelen = namelen;
2399         op_data->op_mode = mode;
2400         op_data->op_mod_time = cfs_time_current_sec();
2401         op_data->op_fsuid = current->fsuid;
2402         op_data->op_fsgid = current->fsgid;
2403         op_data->op_cap = cfs_curproc_cap_pack();
2404         op_data->op_bias = MDS_CHECK_SPLIT;
2405         op_data->op_opc = opc;
2406         op_data->op_mds = 0;
2407         op_data->op_data = data;
2408
2409         return op_data;
2410 }
2411
2412 void ll_finish_md_op_data(struct md_op_data *op_data)
2413 {
2414         capa_put(op_data->op_capa1);
2415         capa_put(op_data->op_capa2);
2416         OBD_FREE_PTR(op_data);
2417 }
2418
2419 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2420 {
2421         struct ll_sb_info *sbi;
2422
2423         LASSERT((seq != NULL) && (vfs != NULL));
2424         sbi = ll_s2sbi(vfs->mnt_sb);
2425
2426         if (sbi->ll_flags & LL_SBI_NOLCK)
2427                 seq_puts(seq, ",nolock");
2428
2429         if (sbi->ll_flags & LL_SBI_FLOCK)
2430                 seq_puts(seq, ",flock");
2431
2432         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2433                 seq_puts(seq, ",localflock");
2434
2435         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2436                 seq_puts(seq, ",user_xattr");
2437
2438         if (sbi->ll_flags & LL_SBI_ACL)
2439                 seq_puts(seq, ",acl");
2440
2441         RETURN(0);
2442 }