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_V3 };
 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 obd_capa *oc = NULL;
 285         struct obd_statfs osfs;
 286         struct ptlrpc_request *request = NULL;
 287         struct lustre_handle dt_conn = {0, };
 288         struct lustre_handle md_conn = {0, };
 289         struct obd_connect_data *data = NULL;
 290         struct obd_uuid *uuid;
 291         struct lustre_md lmd;
 292         obd_valid valid;
 293         int size, err, checksum;
 294         ENTRY;
 295
 296         obd = class_name2obd(md);
 297         if (!obd) {
 298                 CERROR("MD %s: not setup or attached\n", md);
 299                 RETURN(-EINVAL);
 300         }
 301
 302         OBD_ALLOC_PTR(data);
 303         if (data == NULL)
 304                 RETURN(-ENOMEM);
 305
 306         if (proc_lustre_fs_root) {
 307                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 308                                                   dt, md);
 309                 if (err < 0)
 310                         CERROR("could not register mount in /proc/fs/lustre\n");
 311         }
 312
 313         /* indicate the features supported by this client */
 314         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 315                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 316                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 317                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 318                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 319                                   OBD_CONNECT_LOV_V3;
 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         fid_zero(&sbi->ll_root_fid);
 566         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 567         if (err) {
 568                 CERROR("cannot mds_connect: rc = %d\n", err);
 569                 GOTO(out_lock_cn_cb, err);
 570         }
 571         if (!fid_is_sane(&sbi->ll_root_fid)) {
 572                 CERROR("Invalid root fid during mount\n");
 573                 GOTO(out_lock_cn_cb, err = -EINVAL);
 574         }
 575         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 576
 577         sb->s_op = &lustre_super_operations;
 578         sb->s_export_op = &lustre_export_operations;
 579
 580         /* make root inode
 581          * XXX: move this to after cbd setup? */
 582         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 583         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 584                 valid |= OBD_MD_FLRMTPERM;
 585         else if (sbi->ll_flags & LL_SBI_ACL)
 586                 valid |= OBD_MD_FLACL;
 587
 588         err = md_getattr(sbi->ll_md_exp, &sbi->ll_root_fid, oc, valid, 0,
 589                          &request);
 590         if (oc)
 591                 free_capa(oc);
 592         if (err) {
 593                 CERROR("md_getattr failed for root: rc = %d\n", err);
 594                 GOTO(out_lock_cn_cb, err);
 595         }
 596         memset(&lmd, 0, sizeof(lmd));
 597         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 598                                sbi->ll_md_exp, &lmd);
 599         if (err) {
 600                 CERROR("failed to understand root inode md: rc = %d\n", err);
 601                 ptlrpc_req_finished (request);
 602                 GOTO(out_lock_cn_cb, err);
 603         }
 604
 605         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 606         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 607         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 608         ptlrpc_req_finished(request);
 609
 610         if (root == NULL || is_bad_inode(root)) {
 611                 if (lmd.lsm)
 612                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 613 #ifdef CONFIG_FS_POSIX_ACL
 614                 if (lmd.posix_acl) {
 615                         posix_acl_release(lmd.posix_acl);
 616                         lmd.posix_acl = NULL;
 617                 }
 618 #endif
 619                 CERROR("lustre_lite: bad iget4 for root\n");
 620                 GOTO(out_root, err = -EBADF);
 621         }
 622
 623         err = ll_close_thread_start(&sbi->ll_lcq);
 624         if (err) {
 625                 CERROR("cannot start close thread: rc %d\n", err);
 626                 GOTO(out_root, err);
 627         }
 628
 629 #ifdef CONFIG_FS_POSIX_ACL
 630         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 631                 rct_init(&sbi->ll_rct);
 632                 et_init(&sbi->ll_et);
 633         }
 634 #endif
 635
 636         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 637         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 638                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 639                                  NULL);
 640
 641         sb->s_root = d_alloc_root(root);
 642         if (data != NULL)
 643                 OBD_FREE(data, sizeof(*data));
 644
 645         sb->s_root->d_op = &ll_d_root_ops;
 646
 647         sbi->ll_sdev_orig = sb->s_dev;
 648
 649         /* We set sb->s_dev equal on all lustre clients in order to support
 650          * NFS export clustering.  NFSD requires that the FSID be the same
 651          * on all clients. */
 652         /* s_dev is also used in lt_compare() to compare two fs, but that is
 653          * only a node-local comparison. */
 654         uuid = obd_get_uuid(sbi->ll_md_exp);
 655         if (uuid != NULL)
 656                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 657
 658         RETURN(err);
 659 out_root:
 660         if (root)
 661                 iput(root);
 662 out_lock_cn_cb:
 663         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
 664                                       ll_extent_lock_cancel_cb);
 665 out_page_rm_cb:
 666         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 667                                        ll_page_removal_cb);
 668         obd_fid_fini(sbi->ll_dt_exp);
 669 out_dt:
 670         obd_disconnect(sbi->ll_dt_exp);
 671         sbi->ll_dt_exp = NULL;
 672 out_md_fid:
 673         obd_fid_fini(sbi->ll_md_exp);
 674 out_md:
 675         obd_disconnect(sbi->ll_md_exp);
 676         sbi->ll_md_exp = NULL;
 677 out:
 678         if (data != NULL)
 679                 OBD_FREE_PTR(data);
 680         lprocfs_unregister_mountpoint(sbi);
 681         return err;
 682 }
 683
 684 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 685 {
 686         int size, rc;
 687
 688         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 689         size = sizeof(int);
 690         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 691                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 692         if (rc)
 693                 CERROR("Get max mdsize error rc %d \n", rc);
 694
 695         RETURN(rc);
 696 }
 697
 698 void ll_dump_inode(struct inode *inode)
 699 {
 700         struct list_head *tmp;
 701         int dentry_count = 0;
 702
 703         LASSERT(inode != NULL);
 704
 705         list_for_each(tmp, &inode->i_dentry)
 706                 dentry_count++;
 707
 708         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 709                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 710                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 711 }
 712
 713 void lustre_dump_dentry(struct dentry *dentry, int recur)
 714 {
 715         struct list_head *tmp;
 716         int subdirs = 0;
 717
 718         LASSERT(dentry != NULL);
 719
 720         list_for_each(tmp, &dentry->d_subdirs)
 721                 subdirs++;
 722
 723         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 724                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 725                dentry->d_name.len, dentry->d_name.name,
 726                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 727                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 728                dentry->d_flags, dentry->d_fsdata, subdirs);
 729         if (dentry->d_inode != NULL)
 730                 ll_dump_inode(dentry->d_inode);
 731
 732         if (recur == 0)
 733                 return;
 734
 735         list_for_each(tmp, &dentry->d_subdirs) {
 736                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 737                 lustre_dump_dentry(d, recur - 1);
 738         }
 739 }
 740
 741 #ifdef HAVE_EXPORT___IGET
 742 static void prune_dir_dentries(struct inode *inode)
 743 {
 744         struct dentry *dentry, *prev = NULL;
 745
 746         /* due to lustre specific logic, a directory
 747          * can have few dentries - a bug from VFS POV */
 748 restart:
 749         spin_lock(&dcache_lock);
 750         if (!list_empty(&inode->i_dentry)) {
 751                 dentry = list_entry(inode->i_dentry.prev,
 752                                     struct dentry, d_alias);
 753                 /* in order to prevent infinite loops we
 754                  * break if previous dentry is busy */
 755                 if (dentry != prev) {
 756                         prev = dentry;
 757                         dget_locked(dentry);
 758                         spin_unlock(&dcache_lock);
 759
 760                         /* try to kill all child dentries */
 761                         lock_dentry(dentry);
 762                         shrink_dcache_parent(dentry);
 763                         unlock_dentry(dentry);
 764                         dput(dentry);
 765
 766                         /* now try to get rid of current dentry */
 767                         d_prune_aliases(inode);
 768                         goto restart;
 769                 }
 770         }
 771         spin_unlock(&dcache_lock);
 772 }
 773
 774 static void prune_deathrow_one(struct ll_inode_info *lli)
 775 {
 776         struct inode *inode = ll_info2i(lli);
 777
 778         /* first, try to drop any dentries - they hold a ref on the inode */
 779         if (S_ISDIR(inode->i_mode))
 780                 prune_dir_dentries(inode);
 781         else
 782                 d_prune_aliases(inode);
 783
 784
 785         /* if somebody still uses it, leave it */
 786         LASSERT(atomic_read(&inode->i_count) > 0);
 787         if (atomic_read(&inode->i_count) > 1)
 788                 goto out;
 789
 790         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 791                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 792
 793         /* seems nobody uses it anymore */
 794         inode->i_nlink = 0;
 795
 796 out:
 797         iput(inode);
 798         return;
 799 }
 800
 801 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 802 {
 803         struct ll_inode_info *lli;
 804         int empty;
 805
 806         do {
 807                 if (need_resched() && try)
 808                         break;
 809
 810                 if (try) {
 811                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 812                                 break;
 813                 } else {
 814                         spin_lock(&sbi->ll_deathrow_lock);
 815                 }
 816
 817                 empty = 1;
 818                 lli = NULL;
 819                 if (!list_empty(&sbi->ll_deathrow)) {
 820                         lli = list_entry(sbi->ll_deathrow.next,
 821                                          struct ll_inode_info,
 822                                          lli_dead_list);
 823                         list_del_init(&lli->lli_dead_list);
 824                         if (!list_empty(&sbi->ll_deathrow))
 825                                 empty = 0;
 826                 }
 827                 spin_unlock(&sbi->ll_deathrow_lock);
 828
 829                 if (lli)
 830                         prune_deathrow_one(lli);
 831
 832         } while (empty == 0);
 833 }
 834 #else /* !HAVE_EXPORT___IGET */
 835 #define prune_deathrow(sbi, try) do {} while (0)
 836 #endif /* HAVE_EXPORT___IGET */
 837
 838 void client_common_put_super(struct super_block *sb)
 839 {
 840         struct ll_sb_info *sbi = ll_s2sbi(sb);
 841         ENTRY;
 842
 843 #ifdef CONFIG_FS_POSIX_ACL
 844         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 845                 et_fini(&sbi->ll_et);
 846                 rct_fini(&sbi->ll_rct);
 847         }
 848 #endif
 849
 850         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 851
 852         ll_close_thread_shutdown(sbi->ll_lcq);
 853
 854         /* destroy inodes in deathrow */
 855         prune_deathrow(sbi, 0);
 856
 857         list_del(&sbi->ll_conn_chain);
 858
 859         obd_fid_fini(sbi->ll_dt_exp);
 860         obd_disconnect(sbi->ll_dt_exp);
 861         sbi->ll_dt_exp = NULL;
 862
 863         lprocfs_unregister_mountpoint(sbi);
 864
 865         obd_fid_fini(sbi->ll_md_exp);
 866         obd_disconnect(sbi->ll_md_exp);
 867         sbi->ll_md_exp = NULL;
 868
 869         EXIT;
 870 }
 871
 872 void ll_kill_super(struct super_block *sb)
 873 {
 874         struct ll_sb_info *sbi;
 875
 876         ENTRY;
 877
 878         /* not init sb ?*/
 879         if (!(sb->s_flags & MS_ACTIVE))
 880                 return;
 881
 882         sbi = ll_s2sbi(sb);
 883         /* we need restore s_dev from changed for clustred NFS before put_super
 884          * because new kernels have cached s_dev and change sb->s_dev in
 885          * put_super not affected real removing devices */
 886         if (sbi)
 887                 sb->s_dev = sbi->ll_sdev_orig;
 888         EXIT;
 889 }
 890
 891 char *ll_read_opt(const char *opt, char *data)
 892 {
 893         char *value;
 894         char *retval;
 895         ENTRY;
 896
 897         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 898         if (strncmp(opt, data, strlen(opt)))
 899                 RETURN(NULL);
 900         if ((value = strchr(data, '=')) == NULL)
 901                 RETURN(NULL);
 902
 903         value++;
 904         OBD_ALLOC(retval, strlen(value) + 1);
 905         if (!retval) {
 906                 CERROR("out of memory!\n");
 907                 RETURN(NULL);
 908         }
 909
 910         memcpy(retval, value, strlen(value)+1);
 911         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 912         RETURN(retval);
 913 }
 914
 915 static inline int ll_set_opt(const char *opt, char *data, int fl)
 916 {
 917         if (strncmp(opt, data, strlen(opt)) != 0)
 918                 return(0);
 919         else
 920                 return(fl);
 921 }
 922
 923 /* non-client-specific mount options are parsed in lmd_parse */
 924 static int ll_options(char *options, int *flags)
 925 {
 926         int tmp;
 927         char *s1 = options, *s2;
 928         ENTRY;
 929
 930         if (!options)
 931                 RETURN(0);
 932
 933         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 934
 935         while (*s1) {
 936                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 937                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 938                 if (tmp) {
 939                         *flags |= tmp;
 940                         goto next;
 941                 }
 942                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 943                 if (tmp) {
 944                         *flags |= tmp;
 945                         goto next;
 946                 }
 947                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 948                 if (tmp) {
 949                         *flags |= tmp;
 950                         goto next;
 951                 }
 952                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 953                 if (tmp) {
 954                         *flags &= ~tmp;
 955                         goto next;
 956                 }
 957                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 958                 if (tmp) {
 959                         *flags |= tmp;
 960                         goto next;
 961                 }
 962                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 963                 if (tmp) {
 964                         *flags &= ~tmp;
 965                         goto next;
 966                 }
 967                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 968                 if (tmp) {
 969                         /* Ignore deprecated mount option.  The client will
 970                          * always try to mount with ACL support, whether this
 971                          * is used depends on whether server supports it. */
 972                         goto next;
 973                 }
 974                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 975                 if (tmp) {
 976                         goto next;
 977                 }
 978                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 979                 if (tmp) {
 980                         *flags |= tmp;
 981                         goto next;
 982                 }
 983
 984                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 985                 if (tmp) {
 986                         *flags |= tmp;
 987                         goto next;
 988                 }
 989                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 990                 if (tmp) {
 991                         *flags &= ~tmp;
 992                         goto next;
 993                 }
 994                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 995                 if (tmp) {
 996                         *flags |= tmp;
 997                         goto next;
 998                 }
 999                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
1000                 if (tmp) {
1001                         *flags &= ~tmp;
1002                         goto next;
1003                 }
1004
1005                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1006                                    s1);
1007                 RETURN(-EINVAL);
1008
1009 next:
1010                 /* Find next opt */
1011                 s2 = strchr(s1, ',');
1012                 if (s2 == NULL)
1013                         break;
1014                 s1 = s2 + 1;
1015         }
1016         RETURN(0);
1017 }
1018
1019 void ll_lli_init(struct ll_inode_info *lli)
1020 {
1021         lli->lli_inode_magic = LLI_INODE_MAGIC;
1022         sema_init(&lli->lli_size_sem, 1);
1023         sema_init(&lli->lli_write_sem, 1);
1024         lli->lli_flags = 0;
1025         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1026         spin_lock_init(&lli->lli_lock);
1027         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1028         INIT_LIST_HEAD(&lli->lli_close_list);
1029         lli->lli_inode_magic = LLI_INODE_MAGIC;
1030         sema_init(&lli->lli_och_sem, 1);
1031         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1032         lli->lli_mds_exec_och = NULL;
1033         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1034         lli->lli_open_fd_exec_count = 0;
1035         INIT_LIST_HEAD(&lli->lli_dead_list);
1036         lli->lli_remote_perms = NULL;
1037         lli->lli_rmtperm_utime = 0;
1038         sema_init(&lli->lli_rmtperm_sem, 1);
1039         INIT_LIST_HEAD(&lli->lli_oss_capas);
1040 }
1041
1042 int ll_fill_super(struct super_block *sb)
1043 {
1044         struct lustre_profile *lprof;
1045         struct lustre_sb_info *lsi = s2lsi(sb);
1046         struct ll_sb_info *sbi;
1047         char  *dt = NULL, *md = NULL;
1048         char  *profilenm = get_profile_name(sb);
1049         struct config_llog_instance cfg = {0, };
1050         char   ll_instance[sizeof(sb) * 2 + 1];
1051         int    err;
1052         ENTRY;
1053
1054         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1055
1056         cfs_module_get();
1057
1058         /* client additional sb info */
1059         lsi->lsi_llsbi = sbi = ll_init_sbi();
1060         if (!sbi) {
1061                 cfs_module_put();
1062                 RETURN(-ENOMEM);
1063         }
1064
1065         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1066         if (err)
1067                 GOTO(out_free, err);
1068
1069         /* Generate a string unique to this super, in case some joker tries
1070            to mount the same fs at two mount points.
1071            Use the address of the super itself.*/
1072         sprintf(ll_instance, "%p", sb);
1073         cfg.cfg_instance = ll_instance;
1074         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1075
1076         /* set up client obds */
1077         err = lustre_process_log(sb, profilenm, &cfg);
1078         if (err < 0) {
1079                 CERROR("Unable to process log: %d\n", err);
1080                 GOTO(out_free, err);
1081         }
1082
1083         lprof = class_get_profile(profilenm);
1084         if (lprof == NULL) {
1085                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1086                                    " read from the MGS.  Does that filesystem "
1087                                    "exist?\n", profilenm);
1088                 GOTO(out_free, err = -EINVAL);
1089         }
1090         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1091                lprof->lp_md, lprof->lp_dt);
1092
1093         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1094                   strlen(ll_instance) + 2);
1095         if (!dt)
1096                 GOTO(out_free, err = -ENOMEM);
1097         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1098
1099         OBD_ALLOC(md, strlen(lprof->lp_md) +
1100                   strlen(ll_instance) + 2);
1101         if (!md)
1102                 GOTO(out_free, err = -ENOMEM);
1103         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1104
1105         /* connections, registrations, sb setup */
1106         err = client_common_fill_super(sb, md, dt);
1107
1108 out_free:
1109         if (md)
1110                 OBD_FREE(md, strlen(md) + 1);
1111         if (dt)
1112                 OBD_FREE(dt, strlen(dt) + 1);
1113         if (err)
1114                 ll_put_super(sb);
1115         else
1116                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1117
1118         RETURN(err);
1119 } /* ll_fill_super */
1120
1121
1122 void ll_put_super(struct super_block *sb)
1123 {
1124         struct config_llog_instance cfg;
1125         char   ll_instance[sizeof(sb) * 2 + 1];
1126         struct obd_device *obd;
1127         struct lustre_sb_info *lsi = s2lsi(sb);
1128         struct ll_sb_info *sbi = ll_s2sbi(sb);
1129         char *profilenm = get_profile_name(sb);
1130         int force = 1, next;
1131         ENTRY;
1132
1133         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1134
1135         ll_print_capa_stat(sbi);
1136
1137         sprintf(ll_instance, "%p", sb);
1138         cfg.cfg_instance = ll_instance;
1139         lustre_end_log(sb, NULL, &cfg);
1140
1141         if (sbi->ll_md_exp) {
1142                 obd = class_exp2obd(sbi->ll_md_exp);
1143                 if (obd)
1144                         force = obd->obd_force;
1145         }
1146
1147         /* We need to set force before the lov_disconnect in
1148            lustre_common_put_super, since l_d cleans up osc's as well. */
1149         if (force) {
1150                 next = 0;
1151                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1152                                                      &next)) != NULL) {
1153                         obd->obd_force = force;
1154                 }
1155         }
1156
1157         if (sbi->ll_lcq) {
1158                 /* Only if client_common_fill_super succeeded */
1159                 client_common_put_super(sb);
1160         }
1161         next = 0;
1162         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1163                 class_manual_cleanup(obd);
1164         }
1165
1166         if (profilenm)
1167                 class_del_profile(profilenm);
1168
1169         ll_free_sbi(sb);
1170         lsi->lsi_llsbi = NULL;
1171
1172         lustre_common_put_super(sb);
1173
1174         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1175
1176         cfs_module_put();
1177
1178         EXIT;
1179 } /* client_put_super */
1180
1181 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1182
1183 #if defined(HAVE_CACHE_RETURN_INT)
1184 static int
1185 #else
1186 static void
1187 #endif
1188 ll_shrink_cache(int priority, unsigned int gfp_mask)
1189 {
1190         struct ll_sb_info *sbi;
1191         int count = 0;
1192
1193         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1194                 count += llap_shrink_cache(sbi, priority);
1195
1196 #if defined(HAVE_CACHE_RETURN_INT)
1197         return count;
1198 #endif
1199 }
1200
1201 struct cache_definition ll_cache_definition = {
1202         .name = "llap_cache",
1203         .shrink = ll_shrink_cache
1204 };
1205 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1206
1207 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1208 {
1209         struct inode *inode = NULL;
1210         /* NOTE: we depend on atomic igrab() -bzzz */
1211         lock_res_and_lock(lock);
1212         if (lock->l_ast_data) {
1213                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1214                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1215                         inode = igrab(lock->l_ast_data);
1216                 } else {
1217                         inode = lock->l_ast_data;
1218                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1219                                                 D_INFO : D_WARNING,
1220                                         lock, __FILE__, __func__, __LINE__,
1221                                         "l_ast_data %p is bogus: magic %08x",
1222                                         lock->l_ast_data, lli->lli_inode_magic);
1223                         inode = NULL;
1224                 }
1225         }
1226         unlock_res_and_lock(lock);
1227         return inode;
1228 }
1229
1230 static int null_if_equal(struct ldlm_lock *lock, void *data)
1231 {
1232         if (data == lock->l_ast_data) {
1233                 lock->l_ast_data = NULL;
1234
1235                 if (lock->l_req_mode != lock->l_granted_mode)
1236                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1237         }
1238
1239         return LDLM_ITER_CONTINUE;
1240 }
1241
1242 void ll_clear_inode(struct inode *inode)
1243 {
1244         struct ll_inode_info *lli = ll_i2info(inode);
1245         struct ll_sb_info *sbi = ll_i2sbi(inode);
1246         ENTRY;
1247
1248         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1249                inode->i_generation, inode);
1250
1251         if (S_ISDIR(inode->i_mode)) {
1252                 /* these should have been cleared in ll_file_release */
1253                 LASSERT(lli->lli_sai == NULL);
1254                 LASSERT(lli->lli_opendir_key == NULL);
1255                 LASSERT(lli->lli_opendir_pid == 0);
1256         }
1257
1258         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1259         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1260                          null_if_equal, inode);
1261
1262         LASSERT(!lli->lli_open_fd_write_count);
1263         LASSERT(!lli->lli_open_fd_read_count);
1264         LASSERT(!lli->lli_open_fd_exec_count);
1265
1266         if (lli->lli_mds_write_och)
1267                 ll_md_real_close(inode, FMODE_WRITE);
1268         if (lli->lli_mds_exec_och)
1269                 ll_md_real_close(inode, FMODE_EXEC);
1270         if (lli->lli_mds_read_och)
1271                 ll_md_real_close(inode, FMODE_READ);
1272
1273         if (lli->lli_smd) {
1274                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1275                                   null_if_equal, inode);
1276
1277                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1278                 lli->lli_smd = NULL;
1279         }
1280
1281         if (lli->lli_symlink_name) {
1282                 OBD_FREE(lli->lli_symlink_name,
1283                          strlen(lli->lli_symlink_name) + 1);
1284                 lli->lli_symlink_name = NULL;
1285         }
1286
1287         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1288                 LASSERT(lli->lli_posix_acl == NULL);
1289                 if (lli->lli_remote_perms) {
1290                         free_rmtperm_hash(lli->lli_remote_perms);
1291                         lli->lli_remote_perms = NULL;
1292                 }
1293         }
1294 #ifdef CONFIG_FS_POSIX_ACL
1295         else if (lli->lli_posix_acl) {
1296                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1297                 LASSERT(lli->lli_remote_perms == NULL);
1298                 posix_acl_release(lli->lli_posix_acl);
1299                 lli->lli_posix_acl = NULL;
1300         }
1301 #endif
1302         lli->lli_inode_magic = LLI_INODE_DEAD;
1303
1304 #ifdef HAVE_EXPORT___IGET
1305         spin_lock(&sbi->ll_deathrow_lock);
1306         list_del_init(&lli->lli_dead_list);
1307         spin_unlock(&sbi->ll_deathrow_lock);
1308 #endif
1309         ll_clear_inode_capas(inode);
1310
1311         EXIT;
1312 }
1313
1314 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1315                   struct md_open_data **mod)
1316 {
1317         struct lustre_md md;
1318         struct ll_sb_info *sbi = ll_i2sbi(inode);
1319         struct ptlrpc_request *request = NULL;
1320         int rc;
1321         ENTRY;
1322
1323         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1324                                      LUSTRE_OPC_ANY, NULL);
1325         if (IS_ERR(op_data))
1326                 RETURN(PTR_ERR(op_data));
1327
1328         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1329                         &request, mod);
1330         if (rc) {
1331                 ptlrpc_req_finished(request);
1332                 if (rc == -ENOENT) {
1333                         inode->i_nlink = 0;
1334                         /* Unlinked special device node? Or just a race?
1335                          * Pretend we done everything. */
1336                         if (!S_ISREG(inode->i_mode) &&
1337                             !S_ISDIR(inode->i_mode))
1338                                 rc = inode_setattr(inode, &op_data->op_attr);
1339                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1340                         CERROR("md_setattr fails: rc = %d\n", rc);
1341                 }
1342                 RETURN(rc);
1343         }
1344
1345         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1346                               sbi->ll_md_exp, &md);
1347         if (rc) {
1348                 ptlrpc_req_finished(request);
1349                 RETURN(rc);
1350         }
1351
1352         /* We call inode_setattr to adjust timestamps.
1353          * If there is at least some data in file, we cleared ATTR_SIZE
1354          * above to avoid invoking vmtruncate, otherwise it is important
1355          * to call vmtruncate in inode_setattr to update inode->i_size
1356          * (bug 6196) */
1357         rc = inode_setattr(inode, &op_data->op_attr);
1358
1359         /* Extract epoch data if obtained. */
1360         op_data->op_handle = md.body->handle;
1361         op_data->op_ioepoch = md.body->ioepoch;
1362
1363         ll_update_inode(inode, &md);
1364         ptlrpc_req_finished(request);
1365
1366         RETURN(rc);
1367 }
1368
1369 /* Close IO epoch and send Size-on-MDS attribute update. */
1370 static int ll_setattr_done_writing(struct inode *inode,
1371                                    struct md_op_data *op_data,
1372                                    struct md_open_data *mod)
1373 {
1374         struct ll_inode_info *lli = ll_i2info(inode);
1375         int rc = 0;
1376         ENTRY;
1377
1378         LASSERT(op_data != NULL);
1379         if (!S_ISREG(inode->i_mode))
1380                 RETURN(0);
1381
1382         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1383                op_data->op_ioepoch, PFID(&lli->lli_fid));
1384
1385         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1386         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1387         if (rc == -EAGAIN) {
1388                 /* MDS has instructed us to obtain Size-on-MDS attribute
1389                  * from OSTs and send setattr to back to MDS. */
1390                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1391                                          op_data->op_ioepoch);
1392         } else if (rc) {
1393                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1394                        inode->i_ino, rc);
1395         }
1396         RETURN(rc);
1397 }
1398
1399 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1400 {
1401         struct ll_sb_info *sbi = ll_i2sbi(inode);
1402         struct ll_inode_info *lli = ll_i2info(inode);
1403         struct lov_stripe_md *lsm = lli->lli_smd;
1404         int rc;
1405         ldlm_policy_data_t policy = { .l_extent = {new_size,
1406                                                    OBD_OBJECT_EOF } };
1407         struct lustre_handle lockh = { 0 };
1408         int local_lock = 0; /* 0 - no local lock;
1409                              * 1 - lock taken by lock_extent;
1410                              * 2 - by obd_match*/
1411         int ast_flags;
1412         int err;
1413         ENTRY;
1414
1415         UNLOCK_INODE_MUTEX(inode);
1416         UP_WRITE_I_ALLOC_SEM(inode);
1417
1418         if (sbi->ll_lockless_truncate_enable &&
1419             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1420                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1421                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1422                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1423                 if (rc > 0) {
1424                         local_lock = 2;
1425                         rc = 0;
1426                 } else if (rc == 0) {
1427                         rc = ll_file_punch(inode, new_size, 1);
1428                 }
1429         } else {
1430                 /* XXX when we fix the AST intents to pass the discard-range
1431                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1432                  * XXX here. */
1433                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1434                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1435                                     &lockh, ast_flags);
1436                 if (likely(rc == 0))
1437                         local_lock = 1;
1438         }
1439
1440         LOCK_INODE_MUTEX(inode);
1441         DOWN_WRITE_I_ALLOC_SEM(inode);
1442
1443         if (likely(rc == 0)) {
1444                 /* Only ll_inode_size_lock is taken at this level.
1445                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1446                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1447                  * ll_truncate dropped ll_inode_size_lock() */
1448                 ll_inode_size_lock(inode, 0);
1449                 if (!local_lock) {
1450                         spin_lock(&lli->lli_lock);
1451                         lli->lli_flags |= LLIF_SRVLOCK;
1452                         spin_unlock(&lli->lli_lock);
1453                 }
1454                 rc = vmtruncate(inode, new_size);
1455                 if (!local_lock) {
1456                         spin_lock(&lli->lli_lock);
1457                         lli->lli_flags &= ~LLIF_SRVLOCK;
1458                         spin_unlock(&lli->lli_lock);
1459                 }
1460                 if (rc != 0) {
1461                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1462                         ll_inode_size_unlock(inode, 0);
1463                 }
1464         }
1465
1466         if (local_lock) {
1467                 if (local_lock == 2)
1468                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1469                 else
1470                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1471                 if (unlikely(err != 0)){
1472                         CERROR("extent unlock failed: err=%d,"
1473                                " unlock method =%d\n", err, local_lock);
1474                         if (rc == 0)
1475                                 rc = err;
1476                 }
1477         }
1478         RETURN(rc);
1479 }
1480
1481 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1482  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1483  * keep these values until such a time that objects are allocated for it.
1484  * We do the MDS operations first, as it is checking permissions for us.
1485  * We don't to the MDS RPC if there is nothing that we want to store there,
1486  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1487  * going to do an RPC anyways.
1488  *
1489  * If we are doing a truncate, we will send the mtime and ctime updates
1490  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1491  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1492  * at the same time.
1493  */
1494 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1495 {
1496         struct ll_inode_info *lli = ll_i2info(inode);
1497         struct lov_stripe_md *lsm = lli->lli_smd;
1498         struct ll_sb_info *sbi = ll_i2sbi(inode);
1499         struct md_op_data *op_data = NULL;
1500         struct md_open_data *mod = NULL;
1501         int ia_valid = attr->ia_valid;
1502         int rc = 0, rc1 = 0;
1503         ENTRY;
1504
1505         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1506                attr->ia_valid);
1507         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1508
1509         if (ia_valid & ATTR_SIZE) {
1510                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1511                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1512                                attr->ia_size, ll_file_maxbytes(inode));
1513                         RETURN(-EFBIG);
1514                 }
1515
1516                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1517         }
1518
1519         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1520         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1521                 if (current->fsuid != inode->i_uid &&
1522                     !cfs_capable(CFS_CAP_FOWNER))
1523                         RETURN(-EPERM);
1524         }
1525
1526         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1527         if (attr->ia_valid & ATTR_CTIME) {
1528                 attr->ia_ctime = CURRENT_TIME;
1529                 attr->ia_valid |= ATTR_CTIME_SET;
1530         }
1531         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1532                 attr->ia_atime = CURRENT_TIME;
1533                 attr->ia_valid |= ATTR_ATIME_SET;
1534         }
1535         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1536                 attr->ia_mtime = CURRENT_TIME;
1537                 attr->ia_valid |= ATTR_MTIME_SET;
1538         }
1539         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1540                 /* To avoid stale mtime on mds, obtain it from ost and send
1541                    to mds. */
1542                 rc = ll_glimpse_size(inode, 0);
1543                 if (rc)
1544                         RETURN(rc);
1545
1546                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1547                 attr->ia_mtime = inode->i_mtime;
1548         }
1549
1550         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1551                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1552                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1553                        cfs_time_current_sec());
1554
1555         /* NB: ATTR_SIZE will only be set after this point if the size
1556          * resides on the MDS, ie, this file has no objects. */
1557         if (lsm)
1558                 attr->ia_valid &= ~ATTR_SIZE;
1559
1560         /* We always do an MDS RPC, even if we're only changing the size;
1561          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1562
1563         OBD_ALLOC_PTR(op_data);
1564         if (op_data == NULL)
1565                 RETURN(-ENOMEM);
1566
1567         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1568
1569         /* Open epoch for truncate. */
1570         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1571             (ia_valid & ATTR_SIZE))
1572                 op_data->op_flags = MF_EPOCH_OPEN;
1573
1574         rc = ll_md_setattr(inode, op_data, &mod);
1575         if (rc)
1576                 GOTO(out, rc);
1577
1578         if (op_data->op_ioepoch)
1579                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1580                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1581
1582         if (!lsm || !S_ISREG(inode->i_mode)) {
1583                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1584                 GOTO(out, rc = 0);
1585         }
1586
1587         /* We really need to get our PW lock before we change inode->i_size.
1588          * If we don't we can race with other i_size updaters on our node, like
1589          * ll_file_read.  We can also race with i_size propogation to other
1590          * nodes through dirtying and writeback of final cached pages.  This
1591          * last one is especially bad for racing o_append users on other
1592          * nodes. */
1593         if (ia_valid & ATTR_SIZE) {
1594                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1595         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1596                 obd_flag flags;
1597                 struct obd_info oinfo = { { { 0 } } };
1598                 struct obdo *oa;
1599
1600                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1601                        inode->i_ino, LTIME_S(attr->ia_mtime));
1602
1603                 OBDO_ALLOC(oa);
1604                 if (oa) {
1605                         oa->o_id = lsm->lsm_object_id;
1606                         oa->o_gr = lsm->lsm_object_gr;
1607                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1608
1609                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1610                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1611                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1612                                 OBD_MD_FLGROUP;
1613
1614                         obdo_from_inode(oa, inode, flags);
1615
1616                         oinfo.oi_oa = oa;
1617                         oinfo.oi_md = lsm;
1618                         oinfo.oi_capa = ll_mdscapa_get(inode);
1619
1620                         /* XXX: this looks unnecessary now. */
1621                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1622                         capa_put(oinfo.oi_capa);
1623                         if (rc)
1624                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1625                         OBDO_FREE(oa);
1626                 } else {
1627                         rc = -ENOMEM;
1628                 }
1629         }
1630         EXIT;
1631 out:
1632         if (op_data) {
1633                 if (op_data->op_ioepoch)
1634                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1635                 ll_finish_md_op_data(op_data);
1636         }
1637         return rc ? rc : rc1;
1638 }
1639
1640 int ll_setattr(struct dentry *de, struct iattr *attr)
1641 {
1642         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1643             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1644                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1645
1646         if ((de->d_inode->i_mode & S_ISUID) &&
1647             !(attr->ia_mode & S_ISUID) &&
1648             !(attr->ia_valid & ATTR_KILL_SUID))
1649                 attr->ia_valid |= ATTR_KILL_SUID;
1650
1651         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1652             !(attr->ia_mode & S_ISGID) &&
1653             !(attr->ia_valid & ATTR_KILL_SGID))
1654                 attr->ia_valid |= ATTR_KILL_SGID;
1655
1656         return ll_setattr_raw(de->d_inode, attr);
1657 }
1658
1659 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1660                        __u64 max_age, __u32 flags)
1661 {
1662         struct ll_sb_info *sbi = ll_s2sbi(sb);
1663         struct obd_statfs obd_osfs;
1664         int rc;
1665         ENTRY;
1666
1667         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1668         if (rc) {
1669                 CERROR("md_statfs fails: rc = %d\n", rc);
1670                 RETURN(rc);
1671         }
1672
1673         osfs->os_type = sb->s_magic;
1674
1675         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1676                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1677
1678         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1679                               &obd_osfs, max_age, flags);
1680         if (rc) {
1681                 CERROR("obd_statfs fails: rc = %d\n", rc);
1682                 RETURN(rc);
1683         }
1684
1685         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1686                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1687                obd_osfs.os_files);
1688
1689         osfs->os_bsize = obd_osfs.os_bsize;
1690         osfs->os_blocks = obd_osfs.os_blocks;
1691         osfs->os_bfree = obd_osfs.os_bfree;
1692         osfs->os_bavail = obd_osfs.os_bavail;
1693
1694         /* If we don't have as many objects free on the OST as inodes
1695          * on the MDS, we reduce the total number of inodes to
1696          * compensate, so that the "inodes in use" number is correct.
1697          */
1698         if (obd_osfs.os_ffree < osfs->os_ffree) {
1699                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1700                         obd_osfs.os_ffree;
1701                 osfs->os_ffree = obd_osfs.os_ffree;
1702         }
1703
1704         RETURN(rc);
1705 }
1706 #ifndef HAVE_STATFS_DENTRY_PARAM
1707 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1708 {
1709 #else
1710 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1711 {
1712         struct super_block *sb = de->d_sb;
1713 #endif
1714         struct obd_statfs osfs;
1715         int rc;
1716
1717         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1718         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1719
1720         /* For now we will always get up-to-date statfs values, but in the
1721          * future we may allow some amount of caching on the client (e.g.
1722          * from QOS or lprocfs updates). */
1723         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1724         if (rc)
1725                 return rc;
1726
1727         statfs_unpack(sfs, &osfs);
1728
1729         /* We need to downshift for all 32-bit kernels, because we can't
1730          * tell if the kernel is being called via sys_statfs64() or not.
1731          * Stop before overflowing f_bsize - in which case it is better
1732          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1733         if (sizeof(long) < 8) {
1734                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1735                         sfs->f_bsize <<= 1;
1736
1737                         osfs.os_blocks >>= 1;
1738                         osfs.os_bfree >>= 1;
1739                         osfs.os_bavail >>= 1;
1740                 }
1741         }
1742
1743         sfs->f_blocks = osfs.os_blocks;
1744         sfs->f_bfree = osfs.os_bfree;
1745         sfs->f_bavail = osfs.os_bavail;
1746
1747         return 0;
1748 }
1749
1750 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1751 {
1752         struct ll_inode_info *lli;
1753         struct lov_stripe_md *lsm;
1754
1755         lli = ll_i2info(inode);
1756         LASSERT(lli->lli_size_sem_owner != current);
1757         down(&lli->lli_size_sem);
1758         LASSERT(lli->lli_size_sem_owner == NULL);
1759         lli->lli_size_sem_owner = current;
1760         lsm = lli->lli_smd;
1761         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1762                  lsm, lock_lsm);
1763         if (lock_lsm)
1764                 lov_stripe_lock(lsm);
1765 }
1766
1767 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1768 {
1769         struct ll_inode_info *lli;
1770         struct lov_stripe_md *lsm;
1771
1772         lli = ll_i2info(inode);
1773         lsm = lli->lli_smd;
1774         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1775                  lsm, unlock_lsm);
1776         if (unlock_lsm)
1777                 lov_stripe_unlock(lsm);
1778         LASSERT(lli->lli_size_sem_owner == current);
1779         lli->lli_size_sem_owner = NULL;
1780         up(&lli->lli_size_sem);
1781 }
1782
1783 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1784 {
1785         struct ll_inode_info *lli = ll_i2info(inode);
1786
1787         dump_lsm(D_INODE, lsm);
1788         dump_lsm(D_INODE, lli->lli_smd);
1789         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1790                  "lsm must be joined lsm %p\n", lsm);
1791         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1792         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1793                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1794         lli->lli_smd = lsm;
1795         lli->lli_maxbytes = lsm->lsm_maxbytes;
1796         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1797                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1798 }
1799
1800 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1801 {
1802         struct ll_inode_info *lli = ll_i2info(inode);
1803         struct mdt_body *body = md->body;
1804         struct lov_stripe_md *lsm = md->lsm;
1805         struct ll_sb_info *sbi = ll_i2sbi(inode);
1806
1807         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1808         if (lsm != NULL) {
1809                 if (lli->lli_smd == NULL) {
1810                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1811                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1812                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1813                                 dump_lsm(D_ERROR, lsm);
1814                                 LBUG();
1815                         }
1816                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1817                                lsm, inode->i_ino, inode->i_generation, inode);
1818                         /* ll_inode_size_lock() requires it is only called
1819                          * with lli_smd != NULL or lock_lsm == 0 or we can
1820                          * race between lock/unlock.  bug 9547 */
1821                         lli->lli_smd = lsm;
1822                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1823                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1824                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1825                 } else {
1826                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1827                              lli->lli_smd->lsm_stripe_count ==
1828                                         lsm->lsm_stripe_count) {
1829                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1830                                         CERROR("lsm mismatch for inode %ld\n",
1831                                                 inode->i_ino);
1832                                         CERROR("lli_smd:\n");
1833                                         dump_lsm(D_ERROR, lli->lli_smd);
1834                                         CERROR("lsm:\n");
1835                                         dump_lsm(D_ERROR, lsm);
1836                                         LBUG();
1837                                 }
1838                         } else
1839                                 ll_replace_lsm(inode, lsm);
1840                 }
1841                 if (lli->lli_smd != lsm)
1842                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1843         }
1844
1845         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1846                 if (body->valid & OBD_MD_FLRMTPERM)
1847                         ll_update_remote_perm(inode, md->remote_perm);
1848         }
1849 #ifdef CONFIG_FS_POSIX_ACL
1850         else if (body->valid & OBD_MD_FLACL) {
1851                 spin_lock(&lli->lli_lock);
1852                 if (lli->lli_posix_acl)
1853                         posix_acl_release(lli->lli_posix_acl);
1854                 lli->lli_posix_acl = md->posix_acl;
1855                 spin_unlock(&lli->lli_lock);
1856         }
1857 #endif
1858         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1859
1860         if (body->valid & OBD_MD_FLATIME &&
1861             body->atime > LTIME_S(inode->i_atime))
1862                 LTIME_S(inode->i_atime) = body->atime;
1863
1864         /* mtime is always updated with ctime, but can be set in past.
1865            As write and utime(2) may happen within 1 second, and utime's
1866            mtime has a priority over write's one, so take mtime from mds
1867            for the same ctimes. */
1868         if (body->valid & OBD_MD_FLCTIME &&
1869             body->ctime >= LTIME_S(inode->i_ctime)) {
1870                 LTIME_S(inode->i_ctime) = body->ctime;
1871                 if (body->valid & OBD_MD_FLMTIME) {
1872                         CDEBUG(D_INODE, "setting ino %lu mtime "
1873                                "from %lu to "LPU64"\n", inode->i_ino,
1874                                LTIME_S(inode->i_mtime), body->mtime);
1875                         LTIME_S(inode->i_mtime) = body->mtime;
1876                 }
1877         }
1878         if (body->valid & OBD_MD_FLMODE)
1879                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1880         if (body->valid & OBD_MD_FLTYPE)
1881                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1882         LASSERT(inode->i_mode != 0);
1883         if (S_ISREG(inode->i_mode)) {
1884                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1885         } else {
1886                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1887         }
1888 #ifdef HAVE_INODE_BLKSIZE
1889         inode->i_blksize = 1<<inode->i_blkbits;
1890 #endif
1891         if (body->valid & OBD_MD_FLUID)
1892                 inode->i_uid = body->uid;
1893         if (body->valid & OBD_MD_FLGID)
1894                 inode->i_gid = body->gid;
1895         if (body->valid & OBD_MD_FLFLAGS)
1896                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1897         if (body->valid & OBD_MD_FLNLINK)
1898                 inode->i_nlink = body->nlink;
1899         if (body->valid & OBD_MD_FLRDEV)
1900                 inode->i_rdev = old_decode_dev(body->rdev);
1901
1902         if (body->valid & OBD_MD_FLID) {
1903                 /* FID shouldn't be changed! */
1904                 if (fid_is_sane(&lli->lli_fid)) {
1905                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1906                                  "Trying to change FID "DFID
1907                                  " to the "DFID", inode %lu/%u(%p)\n",
1908                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1909                                  inode->i_ino, inode->i_generation, inode);
1910                 } else
1911                         lli->lli_fid = body->fid1;
1912         }
1913
1914         LASSERT(fid_seq(&lli->lli_fid) != 0);
1915
1916         if (body->valid & OBD_MD_FLSIZE) {
1917                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1918                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1919                         struct lustre_handle lockh;
1920                         ldlm_mode_t mode;
1921
1922                         /* As it is possible a blocking ast has been processed
1923                          * by this time, we need to check there is an UPDATE
1924                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1925                          * it. */
1926                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1927                                                &lockh);
1928                         if (mode) {
1929                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1930                                                       LLIF_EPOCH_PENDING |
1931                                                       LLIF_SOM_DIRTY)) {
1932                                         CERROR("ino %lu flags %lu still has "
1933                                                "size authority! do not trust "
1934                                                "the size got from MDS\n",
1935                                                inode->i_ino, lli->lli_flags);
1936                                 } else {
1937                                         /* Use old size assignment to avoid
1938                                          * deadlock bz14138 & bz14326 */
1939                                         inode->i_size = body->size;
1940                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1941                                 }
1942                                 ldlm_lock_decref(&lockh, mode);
1943                         }
1944                 } else {
1945                         /* Use old size assignment to avoid
1946                          * deadlock bz14138 & bz14326 */
1947                         inode->i_size = body->size;
1948                 }
1949
1950                 if (body->valid & OBD_MD_FLBLOCKS)
1951                         inode->i_blocks = body->blocks;
1952         }
1953
1954         if (body->valid & OBD_MD_FLMDSCAPA) {
1955                 LASSERT(md->mds_capa);
1956                 ll_add_capa(inode, md->mds_capa);
1957         }
1958         if (body->valid & OBD_MD_FLOSSCAPA) {
1959                 LASSERT(md->oss_capa);
1960                 ll_add_capa(inode, md->oss_capa);
1961         }
1962 }
1963
1964 static struct backing_dev_info ll_backing_dev_info = {
1965         .ra_pages       = 0,    /* No readahead */
1966 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1967         .capabilities   = 0,    /* Does contribute to dirty memory */
1968 #else
1969         .memory_backed  = 0,    /* Does contribute to dirty memory */
1970 #endif
1971 };
1972
1973 void ll_read_inode2(struct inode *inode, void *opaque)
1974 {
1975         struct lustre_md *md = opaque;
1976         struct ll_inode_info *lli = ll_i2info(inode);
1977         ENTRY;
1978
1979         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1980                inode->i_ino, inode->i_generation, inode);
1981
1982         ll_lli_init(lli);
1983
1984         LASSERT(!lli->lli_smd);
1985
1986         /* Core attributes from the MDS first.  This is a new inode, and
1987          * the VFS doesn't zero times in the core inode so we have to do
1988          * it ourselves.  They will be overwritten by either MDS or OST
1989          * attributes - we just need to make sure they aren't newer. */
1990         LTIME_S(inode->i_mtime) = 0;
1991         LTIME_S(inode->i_atime) = 0;
1992         LTIME_S(inode->i_ctime) = 0;
1993         inode->i_rdev = 0;
1994         ll_update_inode(inode, md);
1995
1996         /* OIDEBUG(inode); */
1997
1998         if (S_ISREG(inode->i_mode)) {
1999                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2000                 inode->i_op = &ll_file_inode_operations;
2001                 inode->i_fop = sbi->ll_fop;
2002                 inode->i_mapping->a_ops = &ll_aops;
2003                 EXIT;
2004         } else if (S_ISDIR(inode->i_mode)) {
2005                 inode->i_op = &ll_dir_inode_operations;
2006                 inode->i_fop = &ll_dir_operations;
2007                 inode->i_mapping->a_ops = &ll_dir_aops;
2008                 EXIT;
2009         } else if (S_ISLNK(inode->i_mode)) {
2010                 inode->i_op = &ll_fast_symlink_inode_operations;
2011                 EXIT;
2012         } else {
2013                 inode->i_op = &ll_special_inode_operations;
2014
2015                 init_special_inode(inode, inode->i_mode,
2016                                    kdev_t_to_nr(inode->i_rdev));
2017
2018                 /* initializing backing dev info. */
2019                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2020
2021                 EXIT;
2022         }
2023 }
2024
2025 void ll_delete_inode(struct inode *inode)
2026 {
2027         struct ll_sb_info *sbi = ll_i2sbi(inode);
2028         int rc;
2029         ENTRY;
2030
2031         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2032         if (rc) {
2033                 CERROR("fid_delete() failed, rc %d\n", rc);
2034         }
2035         truncate_inode_pages(&inode->i_data, 0);
2036         clear_inode(inode);
2037
2038         EXIT;
2039 }
2040
2041 int ll_iocontrol(struct inode *inode, struct file *file,
2042                  unsigned int cmd, unsigned long arg)
2043 {
2044         struct ll_sb_info *sbi = ll_i2sbi(inode);
2045         struct ptlrpc_request *req = NULL;
2046         int rc, flags = 0;
2047         ENTRY;
2048
2049         switch(cmd) {
2050         case EXT3_IOC_GETFLAGS: {
2051                 struct mdt_body *body;
2052                 struct obd_capa *oc;
2053
2054                 oc = ll_mdscapa_get(inode);
2055                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2056                                 OBD_MD_FLFLAGS, 0, &req);
2057                 capa_put(oc);
2058                 if (rc) {
2059                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2060                         RETURN(-abs(rc));
2061                 }
2062
2063                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2064
2065                 flags = body->flags;
2066
2067                 ptlrpc_req_finished(req);
2068
2069                 RETURN(put_user(flags, (int *)arg));
2070         }
2071         case EXT3_IOC_SETFLAGS: {
2072                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2073                 struct obd_info oinfo = { { { 0 } } };
2074                 struct md_op_data *op_data;
2075
2076                 if (get_user(flags, (int *)arg))
2077                         RETURN(-EFAULT);
2078
2079                 oinfo.oi_md = lsm;
2080                 OBDO_ALLOC(oinfo.oi_oa);
2081                 if (!oinfo.oi_oa)
2082                         RETURN(-ENOMEM);
2083
2084                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2085                                              LUSTRE_OPC_ANY, NULL);
2086                 if (IS_ERR(op_data))
2087                         RETURN(PTR_ERR(op_data));
2088
2089                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2090                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2091                 rc = md_setattr(sbi->ll_md_exp, op_data,
2092                                 NULL, 0, NULL, 0, &req, NULL);
2093                 ll_finish_md_op_data(op_data);
2094                 ptlrpc_req_finished(req);
2095                 if (rc || lsm == NULL) {
2096                         OBDO_FREE(oinfo.oi_oa);
2097                         RETURN(rc);
2098                 }
2099
2100                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2101                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2102                 oinfo.oi_oa->o_flags = flags;
2103                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2104                                        OBD_MD_FLGROUP;
2105                 oinfo.oi_capa = ll_mdscapa_get(inode);
2106
2107                 obdo_from_inode(oinfo.oi_oa, inode,
2108                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2109                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2110                 capa_put(oinfo.oi_capa);
2111                 OBDO_FREE(oinfo.oi_oa);
2112                 if (rc) {
2113                         if (rc != -EPERM && rc != -EACCES)
2114                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2115                         RETURN(rc);
2116                 }
2117
2118                 inode->i_flags = ll_ext_to_inode_flags(flags |
2119                                                        MDS_BFLAG_EXT_FLAGS);
2120                 RETURN(0);
2121         }
2122         default:
2123                 RETURN(-ENOSYS);
2124         }
2125
2126         RETURN(0);
2127 }
2128
2129 int ll_flush_ctx(struct inode *inode)
2130 {
2131         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2132
2133         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2134
2135         obd_set_info_async(sbi->ll_md_exp,
2136                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2137                            0, NULL, NULL);
2138         obd_set_info_async(sbi->ll_dt_exp,
2139                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2140                            0, NULL, NULL);
2141         return 0;
2142 }
2143
2144 /* umount -f client means force down, don't save state */
2145 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2146 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2147 {
2148         struct super_block *sb = vfsmnt->mnt_sb;
2149 #else
2150 void ll_umount_begin(struct super_block *sb)
2151 {
2152 #endif
2153         struct lustre_sb_info *lsi = s2lsi(sb);
2154         struct ll_sb_info *sbi = ll_s2sbi(sb);
2155         struct obd_device *obd;
2156         struct obd_ioctl_data ioc_data = { 0 };
2157         ENTRY;
2158
2159 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2160         if (!(flags & MNT_FORCE)) {
2161                 EXIT;
2162                 return;
2163         }
2164 #endif
2165
2166         /* Tell the MGC we got umount -f */
2167         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2168
2169         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2170                sb->s_count, atomic_read(&sb->s_active));
2171
2172         obd = class_exp2obd(sbi->ll_md_exp);
2173         if (obd == NULL) {
2174                 CERROR("Invalid MDC connection handle "LPX64"\n",
2175                        sbi->ll_md_exp->exp_handle.h_cookie);
2176                 EXIT;
2177                 return;
2178         }
2179         obd->obd_force = 1;
2180         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2181                       &ioc_data, NULL);
2182
2183         obd = class_exp2obd(sbi->ll_dt_exp);
2184         if (obd == NULL) {
2185                 CERROR("Invalid LOV connection handle "LPX64"\n",
2186                        sbi->ll_dt_exp->exp_handle.h_cookie);
2187                 EXIT;
2188                 return;
2189         }
2190
2191         obd->obd_force = 1;
2192         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2193                       &ioc_data, NULL);
2194
2195         /* Really, we'd like to wait until there are no requests outstanding,
2196          * and then continue.  For now, we just invalidate the requests,
2197          * schedule, and hope.
2198          */
2199         schedule();
2200
2201         EXIT;
2202 }
2203
2204 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2205 {
2206         struct ll_sb_info *sbi = ll_s2sbi(sb);
2207         int err;
2208         __u32 read_only;
2209
2210         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2211                 read_only = *flags & MS_RDONLY;
2212                 err = obd_set_info_async(sbi->ll_md_exp,
2213                                          sizeof(KEY_READ_ONLY),
2214                                          KEY_READ_ONLY, sizeof(read_only),
2215                                          &read_only, NULL);
2216                 if (err) {
2217                         CERROR("Failed to change the read-only flag during "
2218                                "remount: %d\n", err);
2219                         return err;
2220                 }
2221
2222                 if (read_only)
2223                         sb->s_flags |= MS_RDONLY;
2224                 else
2225                         sb->s_flags &= ~MS_RDONLY;
2226         }
2227         return 0;
2228 }
2229
2230 int ll_prep_inode(struct inode **inode,
2231                   struct ptlrpc_request *req,
2232                   struct super_block *sb)
2233 {
2234         struct ll_sb_info *sbi = NULL;
2235         struct lustre_md md;
2236         int rc = 0;
2237         ENTRY;
2238
2239         LASSERT(*inode || sb);
2240         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2241         prune_deathrow(sbi, 1);
2242         memset(&md, 0, sizeof(struct lustre_md));
2243
2244         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2245                               sbi->ll_md_exp, &md);
2246         if (rc)
2247                 RETURN(rc);
2248
2249         if (*inode) {
2250                 ll_update_inode(*inode, &md);
2251         } else {
2252                 LASSERT(sb != NULL);
2253
2254                 /*
2255                  * At this point server returns to client's same fid as client
2256                  * generated for creating. So using ->fid1 is okay here.
2257                  */
2258                 LASSERT(fid_is_sane(&md.body->fid1));
2259
2260                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2261                 if (*inode == NULL || is_bad_inode(*inode)) {
2262                         if (md.lsm)
2263                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2264 #ifdef CONFIG_FS_POSIX_ACL
2265                         if (md.posix_acl) {
2266                                 posix_acl_release(md.posix_acl);
2267                                 md.posix_acl = NULL;
2268                         }
2269 #endif
2270                         rc = -ENOMEM;
2271                         CERROR("new_inode -fatal: rc %d\n", rc);
2272                         GOTO(out, rc);
2273                 }
2274         }
2275
2276         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2277                          ll_i2info(*inode)->lli_smd);
2278 out:
2279         md_free_lustre_md(sbi->ll_md_exp, &md);
2280         RETURN(rc);
2281 }
2282
2283 char *llap_origins[] = {
2284         [LLAP_ORIGIN_UNKNOWN] = "--",
2285         [LLAP_ORIGIN_READPAGE] = "rp",
2286         [LLAP_ORIGIN_READAHEAD] = "ra",
2287         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2288         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2289         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2290 };
2291
2292 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2293                                              struct list_head *list)
2294 {
2295         struct ll_async_page *llap;
2296         struct list_head *pos;
2297
2298         list_for_each(pos, list) {
2299                 if (pos == head)
2300                         return NULL;
2301                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2302                 if (llap->llap_page == NULL)
2303                         continue;
2304                 return llap;
2305         }
2306         LBUG();
2307         return NULL;
2308 }
2309
2310 int ll_obd_statfs(struct inode *inode, void *arg)
2311 {
2312         struct ll_sb_info *sbi = NULL;
2313         struct obd_export *exp;
2314         char *buf = NULL;
2315         struct obd_ioctl_data *data = NULL;
2316         __u32 type;
2317         int len = 0, rc;
2318
2319         if (!inode || !(sbi = ll_i2sbi(inode)))
2320                 GOTO(out_statfs, rc = -EINVAL);
2321
2322         rc = obd_ioctl_getdata(&buf, &len, arg);
2323         if (rc)
2324                 GOTO(out_statfs, rc);
2325
2326         data = (void*)buf;
2327         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2328             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2329                 GOTO(out_statfs, rc = -EINVAL);
2330
2331         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2332         if (type == LL_STATFS_MDC)
2333                 exp = sbi->ll_md_exp;
2334         else if (type == LL_STATFS_LOV)
2335                 exp = sbi->ll_dt_exp;
2336         else
2337                 GOTO(out_statfs, rc = -ENODEV);
2338
2339         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2340         if (rc)
2341                 GOTO(out_statfs, rc);
2342 out_statfs:
2343         if (buf)
2344                 obd_ioctl_freedata(buf, len);
2345         return rc;
2346 }
2347
2348 int ll_process_config(struct lustre_cfg *lcfg)
2349 {
2350         char *ptr;
2351         void *sb;
2352         struct lprocfs_static_vars lvars;
2353         unsigned long x;
2354         int rc = 0;
2355
2356         lprocfs_llite_init_vars(&lvars);
2357
2358         /* The instance name contains the sb: lustre-client-aacfe000 */
2359         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2360         if (!ptr || !*(++ptr))
2361                 return -EINVAL;
2362         if (sscanf(ptr, "%lx", &x) != 1)
2363                 return -EINVAL;
2364         sb = (void *)x;
2365         /* This better be a real Lustre superblock! */
2366         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2367
2368         /* Note we have not called client_common_fill_super yet, so
2369            proc fns must be able to handle that! */
2370         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2371                                       lcfg, sb);
2372         return(rc);
2373 }
2374
2375 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2376 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2377                                        struct inode *i1, struct inode *i2,
2378                                        const char *name, int namelen,
2379                                        int mode, __u32 opc, void *data)
2380 {
2381         LASSERT(i1 != NULL);
2382
2383         if (namelen > ll_i2sbi(i1)->ll_namelen)
2384                 return ERR_PTR(-ENAMETOOLONG);
2385
2386         if (op_data == NULL)
2387                 OBD_ALLOC_PTR(op_data);
2388
2389         if (op_data == NULL)
2390                 return ERR_PTR(-ENOMEM);
2391
2392         ll_i2gids(op_data->op_suppgids, i1, i2);
2393         op_data->op_fid1 = *ll_inode2fid(i1);
2394         op_data->op_capa1 = ll_mdscapa_get(i1);
2395
2396         if (i2) {
2397                 op_data->op_fid2 = *ll_inode2fid(i2);
2398                 op_data->op_capa2 = ll_mdscapa_get(i2);
2399         } else {
2400                 fid_zero(&op_data->op_fid2);
2401                 op_data->op_capa2 = NULL;
2402         }
2403
2404         op_data->op_name = name;
2405         op_data->op_namelen = namelen;
2406         op_data->op_mode = mode;
2407         op_data->op_mod_time = cfs_time_current_sec();
2408         op_data->op_fsuid = current->fsuid;
2409         op_data->op_fsgid = current->fsgid;
2410         op_data->op_cap = cfs_curproc_cap_pack();
2411         op_data->op_bias = MDS_CHECK_SPLIT;
2412         op_data->op_opc = opc;
2413         op_data->op_mds = 0;
2414         op_data->op_data = data;
2415
2416         return op_data;
2417 }
2418
2419 void ll_finish_md_op_data(struct md_op_data *op_data)
2420 {
2421         capa_put(op_data->op_capa1);
2422         capa_put(op_data->op_capa2);
2423         OBD_FREE_PTR(op_data);
2424 }
2425
2426 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2427 {
2428         struct ll_sb_info *sbi;
2429
2430         LASSERT((seq != NULL) && (vfs != NULL));
2431         sbi = ll_s2sbi(vfs->mnt_sb);
2432
2433         if (sbi->ll_flags & LL_SBI_NOLCK)
2434                 seq_puts(seq, ",nolock");
2435
2436         if (sbi->ll_flags & LL_SBI_FLOCK)
2437                 seq_puts(seq, ",flock");
2438
2439         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2440                 seq_puts(seq, ",localflock");
2441
2442         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2443                 seq_puts(seq, ",user_xattr");
2444
2445         if (sbi->ll_flags & LL_SBI_ACL)
2446                 seq_puts(seq, ",acl");
2447
2448         RETURN(0);
2449 }