lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <obd_cksum.h>
  57 #include <lustre_cache.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) ffz(~(n))
  70 #endif
  71
  72 static inline void ll_pglist_fini(struct ll_sb_info *sbi)
  73 {
  74         struct page *page;
  75         int i;
  76
  77         if (sbi->ll_pglist == NULL)
  78                 return;
  79
  80         for_each_possible_cpu(i) {
  81                 page = sbi->ll_pglist[i]->llpd_page;
  82                 if (page) {
  83                         sbi->ll_pglist[i] = NULL;
  84                         __free_page(page);
  85                 }
  86         }
  87
  88         OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
  89         sbi->ll_pglist = NULL;
  90 }
  91
  92 static inline int ll_pglist_init(struct ll_sb_info *sbi)
  93 {
  94         struct ll_pglist_data *pd;
  95         unsigned long budget;
  96         int i, color = 0;
  97         ENTRY;
  98
  99         OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
 100         if (sbi->ll_pglist == NULL)
 101                 RETURN(-ENOMEM);
 102
 103         budget = sbi->ll_async_page_max / num_online_cpus();
 104         for_each_possible_cpu(i) {
 105                 struct page *page = alloc_pages_node(cpu_to_node(i),
 106                                                     GFP_KERNEL, 0);
 107                 if (page == NULL) {
 108                         ll_pglist_fini(sbi);
 109                         RETURN(-ENOMEM);
 110                 }
 111
 112                 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
 113                         color = 0;
 114
 115                 pd = (struct ll_pglist_data *)(page_address(page) + color);
 116                 memset(pd, 0, sizeof(*pd));
 117                 spin_lock_init(&pd->llpd_lock);
 118                 INIT_LIST_HEAD(&pd->llpd_list);
 119                 if (cpu_online(i))
 120                         pd->llpd_budget = budget;
 121                 pd->llpd_cpu = i;
 122                 pd->llpd_page = page;
 123                 atomic_set(&pd->llpd_sample_count, 0);
 124                 sbi->ll_pglist[i] = pd;
 125                 color += L1_CACHE_ALIGN(sizeof(*pd));
 126         }
 127
 128         RETURN(0);
 129 }
 130
 131 static struct ll_sb_info *ll_init_sbi(void)
 132 {
 133         struct ll_sb_info *sbi = NULL;
 134         unsigned long pages;
 135         struct sysinfo si;
 136         class_uuid_t uuid;
 137         int i;
 138         ENTRY;
 139
 140         OBD_ALLOC(sbi, sizeof(*sbi));
 141         if (!sbi)
 142                 RETURN(NULL);
 143
 144         OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
 145         if (sbi->ll_async_page_sample == NULL)
 146                 GOTO(out, 0);
 147
 148         spin_lock_init(&sbi->ll_lock);
 149         spin_lock_init(&sbi->ll_lco.lco_lock);
 150         spin_lock_init(&sbi->ll_pp_extent_lock);
 151         spin_lock_init(&sbi->ll_process_lock);
 152         sbi->ll_rw_stats_on = 0;
 153
 154         si_meminfo(&si);
 155         pages = si.totalram - si.totalhigh;
 156         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 157 #ifdef HAVE_BGL_SUPPORT
 158                 sbi->ll_async_page_max = pages / 4;
 159 #else
 160                 sbi->ll_async_page_max = pages / 2;
 161 #endif
 162         } else {
 163                 sbi->ll_async_page_max = (pages / 4) * 3;
 164         }
 165         lcounter_init(&sbi->ll_async_page_count);
 166         spin_lock_init(&sbi->ll_async_page_reblnc_lock);
 167         sbi->ll_async_page_sample_max = 64 * num_online_cpus();
 168         sbi->ll_async_page_reblnc_count = 0;
 169         sbi->ll_async_page_clock_hand = 0;
 170         if (ll_pglist_init(sbi))
 171                 GOTO(out, 0);
 172
 173         sbi->ll_ra_info.ra_max_pages = min(pages / 32,
 174                                            SBI_DEFAULT_READAHEAD_MAX);
 175         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 176                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 177         sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
 178         sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
 179         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 180         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 181
 182         ll_generate_random_uuid(uuid);
 183         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 184         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 185
 186         spin_lock(&ll_sb_lock);
 187         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 188         spin_unlock(&ll_sb_lock);
 189
 190 #ifdef ENABLE_LLITE_CHECKSUM
 191         sbi->ll_flags |= LL_SBI_CHECKSUM;
 192 #endif
 193
 194 #ifdef HAVE_LRU_RESIZE_SUPPORT
 195         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 196 #endif
 197
 198 #ifdef HAVE_EXPORT___IGET
 199         INIT_LIST_HEAD(&sbi->ll_deathrow);
 200         spin_lock_init(&sbi->ll_deathrow_lock);
 201 #endif
 202         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 203                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 204                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 205         }
 206
 207         /* metadata statahead is enabled by default */
 208         sbi->ll_sa_max = LL_SA_RPC_DEF;
 209
 210         RETURN(sbi);
 211
 212 out:
 213         if (sbi->ll_async_page_sample)
 214                 OBD_FREE(sbi->ll_async_page_sample,
 215                          sizeof(long) * num_possible_cpus());
 216         ll_pglist_fini(sbi);
 217         OBD_FREE(sbi, sizeof(*sbi));
 218         RETURN(NULL);
 219 }
 220
 221 void ll_free_sbi(struct super_block *sb)
 222 {
 223         struct ll_sb_info *sbi = ll_s2sbi(sb);
 224         ENTRY;
 225
 226         if (sbi != NULL) {
 227                 ll_pglist_fini(sbi);
 228                 spin_lock(&ll_sb_lock);
 229                 list_del(&sbi->ll_list);
 230                 spin_unlock(&ll_sb_lock);
 231                 lcounter_destroy(&sbi->ll_async_page_count);
 232                 OBD_FREE(sbi->ll_async_page_sample,
 233                          sizeof(long) * num_possible_cpus());
 234                 OBD_FREE(sbi, sizeof(*sbi));
 235         }
 236         EXIT;
 237 }
 238
 239 static struct dentry_operations ll_d_root_ops = {
 240 #ifdef DCACHE_LUSTRE_INVALID
 241         .d_compare = ll_dcompare,
 242 #endif
 243 };
 244
 245 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 246  * us to make MDS RPCs with large enough reply buffers to hold the
 247  * maximum-sized (= maximum striped) EA and cookie without having to
 248  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 249 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 250 {
 251         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC };
 252         __u32 valsize = sizeof(struct lov_desc);
 253         int rc, easize, def_easize, cookiesize;
 254         struct lov_desc desc;
 255         __u32 stripes;
 256         ENTRY;
 257
 258         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
 259                           &valsize, &desc, NULL);
 260         if (rc)
 261                 RETURN(rc);
 262
 263         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 264         lsm.lsm_stripe_count = stripes;
 265         easize = obd_size_diskmd(dt_exp, &lsm);
 266
 267         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 268         def_easize = obd_size_diskmd(dt_exp, &lsm);
 269
 270         cookiesize = stripes * sizeof(struct llog_cookie);
 271
 272         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 273                easize, cookiesize);
 274
 275         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 276         RETURN(rc);
 277 }
 278
 279 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 280 {
 281         struct inode *root = 0;
 282         struct ll_sb_info *sbi = ll_s2sbi(sb);
 283         struct obd_device *obd;
 284         struct lu_fid rootfid;
 285         struct obd_capa *oc = NULL;
 286         struct obd_statfs osfs;
 287         struct ptlrpc_request *request = NULL;
 288         struct lustre_handle dt_conn = {0, };
 289         struct lustre_handle md_conn = {0, };
 290         struct obd_connect_data *data = NULL;
 291         struct lustre_md lmd;
 292         obd_valid valid;
 293         int size, err, checksum;
 294         ENTRY;
 295
 296         obd = class_name2obd(md);
 297         if (!obd) {
 298                 CERROR("MD %s: not setup or attached\n", md);
 299                 RETURN(-EINVAL);
 300         }
 301
 302         OBD_ALLOC_PTR(data);
 303         if (data == NULL)
 304                 RETURN(-ENOMEM);
 305
 306         if (proc_lustre_fs_root) {
 307                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 308                                                   dt, md);
 309                 if (err < 0)
 310                         CERROR("could not register mount in /proc/fs/lustre\n");
 311         }
 312
 313         /* indicate the features supported by this client */
 314         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 315                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 316                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 317                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 318                                   OBD_CONNECT_FID      | OBD_CONNECT_AT;
 319
 320 #ifdef HAVE_LRU_RESIZE_SUPPORT
 321         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 322                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 323 #endif
 324 #ifdef CONFIG_FS_POSIX_ACL
 325         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 326 #endif
 327         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 328         data->ocd_version = LUSTRE_VERSION_CODE;
 329
 330         if (sb->s_flags & MS_RDONLY)
 331                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 332         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 333                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 334
 335 #ifdef HAVE_MS_FLOCK_LOCK
 336         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 337         sb->s_flags |= MS_FLOCK_LOCK;
 338 #endif
 339
 340         if (sbi->ll_flags & LL_SBI_FLOCK)
 341                 sbi->ll_fop = &ll_file_operations_flock;
 342         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 343                 sbi->ll_fop = &ll_file_operations;
 344         else
 345                 sbi->ll_fop = &ll_file_operations_noflock;
 346
 347         /* real client */
 348         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 349         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 350                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 351                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 352         } else {
 353                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 354                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 355         }
 356
 357         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 358         if (err == -EBUSY) {
 359                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 360                                    "recovery, of which this client is not a "
 361                                    "part. Please wait for recovery to complete,"
 362                                    " abort, or time out.\n", md);
 363                 GOTO(out, err);
 364         } else if (err) {
 365                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 366                 GOTO(out, err);
 367         }
 368         sbi->ll_md_exp = class_conn2export(&md_conn);
 369
 370         err = obd_fid_init(sbi->ll_md_exp);
 371         if (err) {
 372                 CERROR("Can't init metadata layer FID infrastructure, "
 373                        "rc %d\n", err);
 374                 GOTO(out_md, err);
 375         }
 376
 377         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 378         if (err)
 379                 GOTO(out_md_fid, err);
 380
 381         size = sizeof(*data);
 382         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 383                            KEY_CONN_DATA,  &size, data, NULL);
 384         if (err) {
 385                 CERROR("Get connect data failed: %d \n", err);
 386                 GOTO(out_md, err);
 387         }
 388
 389         LASSERT(osfs.os_bsize);
 390         sb->s_blocksize = osfs.os_bsize;
 391         sb->s_blocksize_bits = log2(osfs.os_bsize);
 392         sb->s_magic = LL_SUPER_MAGIC;
 393
 394         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 395          *         retval = in_file->f_op->sendfile(...);
 396          *         if (*ppos > max)
 397          *                 retval = -EOVERFLOW;
 398          *
 399          * it will check if *ppos is greater than max. However, max equals to
 400          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 401          * has been defined as a signed long long ineger in linux kernel. */
 402 #if BITS_PER_LONG == 64
 403         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 404 #else
 405         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 406 #endif
 407         sbi->ll_namelen = osfs.os_namelen;
 408         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 409
 410         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 411             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 412                 LCONSOLE_INFO("Disabling user_xattr feature because "
 413                               "it is not supported on the server\n");
 414                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 415         }
 416
 417         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 418 #ifdef MS_POSIXACL
 419                 sb->s_flags |= MS_POSIXACL;
 420 #endif
 421                 sbi->ll_flags |= LL_SBI_ACL;
 422         } else {
 423                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 424 #ifdef MS_POSIXACL
 425                 sb->s_flags &= ~MS_POSIXACL;
 426 #endif
 427                 sbi->ll_flags &= ~LL_SBI_ACL;
 428         }
 429
 430         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 431                 sbi->ll_flags |= LL_SBI_JOIN;
 432
 433         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 434                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 435                         /* sometimes local client claims to be remote, but mdt
 436                          * will disagree when client gss not applied. */
 437                         LCONSOLE_INFO("client claims to be remote, but server "
 438                                       "rejected, forced to be local.\n");
 439                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 440                 }
 441         } else {
 442                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 443                         /* with gss applied, remote client can not claim to be
 444                          * local, so mdt maybe force client to be remote. */
 445                         LCONSOLE_INFO("client claims to be local, but server "
 446                                       "rejected, forced to be remote.\n");
 447                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 448                 }
 449         }
 450
 451         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 452                 LCONSOLE_INFO("client enabled MDS capability!\n");
 453                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 454         }
 455
 456         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 457                 LCONSOLE_INFO("client enabled OSS capability!\n");
 458                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 459         }
 460
 461         sbi->ll_sdev_orig = sb->s_dev;
 462         /* We set sb->s_dev equal on all lustre clients in order to support
 463          * NFS export clustering.  NFSD requires that the FSID be the same
 464          * on all clients. */
 465         /* s_dev is also used in lt_compare() to compare two fs, but that is
 466          * only a node-local comparison. */
 467
 468         /* XXX: this will not work with LMV */
 469         sb->s_dev = get_uuid2int(sbi2mdc(sbi)->cl_target_uuid.uuid,
 470                                  strlen(sbi2mdc(sbi)->cl_target_uuid.uuid));
 471
 472         obd = class_name2obd(dt);
 473         if (!obd) {
 474                 CERROR("DT %s: not setup or attached\n", dt);
 475                 GOTO(out_md_fid, err = -ENODEV);
 476         }
 477
 478         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 479                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 480                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 481                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 482                                   OBD_CONNECT_AT;
 483         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 484                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 485
 486         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 487                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 488                  * disabled by default, because it can still be enabled on the
 489                  * fly via /proc. As a consequence, we still need to come to an
 490                  * agreement on the supported algorithms at connect time */
 491                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 492
 493                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 494                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 495                 else
 496                         /* send the list of supported checksum types */
 497                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 498         }
 499
 500 #ifdef HAVE_LRU_RESIZE_SUPPORT
 501         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 502 #endif
 503         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 504                "ocd_grant: %d\n", data->ocd_connect_flags,
 505                data->ocd_version, data->ocd_grant);
 506
 507         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 508         obd->obd_upcall.onu_upcall = ll_ocd_update;
 509         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 510
 511         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 512         if (err == -EBUSY) {
 513                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 514                                    "recovery, of which this client is not a "
 515                                    "part.  Please wait for recovery to "
 516                                    "complete, abort, or time out.\n", dt);
 517                 GOTO(out_md_fid, err);
 518         } else if (err) {
 519                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 520                 GOTO(out_md_fid, err);
 521         }
 522
 523         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 524
 525         err = obd_fid_init(sbi->ll_dt_exp);
 526         if (err) {
 527                 CERROR("Can't init data layer FID infrastructure, "
 528                        "rc %d\n", err);
 529                 GOTO(out_dt, err);
 530         }
 531
 532         spin_lock(&sbi->ll_lco.lco_lock);
 533         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 534         spin_unlock(&sbi->ll_lco.lco_lock);
 535
 536         err = obd_register_page_removal_cb(sbi->ll_dt_exp,
 537                                            ll_page_removal_cb,
 538                                            ll_pin_extent_cb);
 539         if (err) {
 540                 CERROR("cannot register page removal callback: rc = %d\n",err);
 541                 GOTO(out_dt, err);
 542         }
 543         err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
 544                                           ll_extent_lock_cancel_cb);
 545         if (err) {
 546                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
 547                 GOTO(out_page_rm_cb, err);
 548         }
 549
 550         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 551         if (err) {
 552                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 553                 GOTO(out_lock_cn_cb, err);
 554         }
 555
 556         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 557                                   0, NULL, NULL, NULL, 0, NULL);
 558         if (err < 0) {
 559                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 560                                    "filesystem. There must be at least one "
 561                                    "active OST for a client to start.\n");
 562                 GOTO(out_lock_cn_cb, err);
 563         }
 564
 565         if (!ll_async_page_slab) {
 566                 ll_async_page_slab_size =
 567                         size_round(sizeof(struct ll_async_page)) + err;
 568                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 569                                                           ll_async_page_slab_size,
 570                                                           0, 0);
 571                 if (!ll_async_page_slab)
 572                         GOTO(out_lock_cn_cb, err = -ENOMEM);
 573         }
 574
 575         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 576         if (err) {
 577                 CERROR("cannot mds_connect: rc = %d\n", err);
 578                 GOTO(out_lock_cn_cb, err);
 579         }
 580         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 581         sbi->ll_root_fid = rootfid;
 582
 583         sb->s_op = &lustre_super_operations;
 584         sb->s_export_op = &lustre_export_operations;
 585
 586         /* make root inode
 587          * XXX: move this to after cbd setup? */
 588         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 589         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 590                 valid |= OBD_MD_FLRMTPERM;
 591         else if (sbi->ll_flags & LL_SBI_ACL)
 592                 valid |= OBD_MD_FLACL;
 593
 594         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 595         if (oc)
 596                 free_capa(oc);
 597         if (err) {
 598                 CERROR("md_getattr failed for root: rc = %d\n", err);
 599                 GOTO(out_lock_cn_cb, err);
 600         }
 601         memset(&lmd, 0, sizeof(lmd));
 602         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 603                                sbi->ll_md_exp, &lmd);
 604         if (err) {
 605                 CERROR("failed to understand root inode md: rc = %d\n", err);
 606                 ptlrpc_req_finished (request);
 607                 GOTO(out_lock_cn_cb, err);
 608         }
 609
 610         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 611         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 612         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 613         ptlrpc_req_finished(request);
 614
 615         if (root == NULL || is_bad_inode(root)) {
 616                 if (lmd.lsm)
 617                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 618 #ifdef CONFIG_FS_POSIX_ACL
 619                 if (lmd.posix_acl) {
 620                         posix_acl_release(lmd.posix_acl);
 621                         lmd.posix_acl = NULL;
 622                 }
 623 #endif
 624                 CERROR("lustre_lite: bad iget4 for root\n");
 625                 GOTO(out_root, err = -EBADF);
 626         }
 627
 628         err = ll_close_thread_start(&sbi->ll_lcq);
 629         if (err) {
 630                 CERROR("cannot start close thread: rc %d\n", err);
 631                 GOTO(out_root, err);
 632         }
 633
 634 #ifdef CONFIG_FS_POSIX_ACL
 635         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 636                 rct_init(&sbi->ll_rct);
 637                 et_init(&sbi->ll_et);
 638         }
 639 #endif
 640
 641         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 642         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 643                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 644                                  NULL);
 645
 646         sb->s_root = d_alloc_root(root);
 647         if (data != NULL)
 648                 OBD_FREE(data, sizeof(*data));
 649         sb->s_root->d_op = &ll_d_root_ops;
 650         RETURN(err);
 651 out_root:
 652         if (root)
 653                 iput(root);
 654 out_lock_cn_cb:
 655         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
 656                                       ll_extent_lock_cancel_cb);
 657 out_page_rm_cb:
 658         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 659                                        ll_page_removal_cb);
 660         obd_fid_fini(sbi->ll_dt_exp);
 661 out_dt:
 662         obd_disconnect(sbi->ll_dt_exp);
 663         sbi->ll_dt_exp = NULL;
 664 out_md_fid:
 665         obd_fid_fini(sbi->ll_md_exp);
 666 out_md:
 667         obd_disconnect(sbi->ll_md_exp);
 668         sbi->ll_md_exp = NULL;
 669 out:
 670         if (data != NULL)
 671                 OBD_FREE_PTR(data);
 672         lprocfs_unregister_mountpoint(sbi);
 673         return err;
 674 }
 675
 676 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 677 {
 678         int size, rc;
 679
 680         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 681         size = sizeof(int);
 682         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 683                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 684         if (rc)
 685                 CERROR("Get max mdsize error rc %d \n", rc);
 686
 687         RETURN(rc);
 688 }
 689
 690 void ll_dump_inode(struct inode *inode)
 691 {
 692         struct list_head *tmp;
 693         int dentry_count = 0;
 694
 695         LASSERT(inode != NULL);
 696
 697         list_for_each(tmp, &inode->i_dentry)
 698                 dentry_count++;
 699
 700         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 701                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 702                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 703 }
 704
 705 void lustre_dump_dentry(struct dentry *dentry, int recur)
 706 {
 707         struct list_head *tmp;
 708         int subdirs = 0;
 709
 710         LASSERT(dentry != NULL);
 711
 712         list_for_each(tmp, &dentry->d_subdirs)
 713                 subdirs++;
 714
 715         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 716                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 717                dentry->d_name.len, dentry->d_name.name,
 718                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 719                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 720                dentry->d_flags, dentry->d_fsdata, subdirs);
 721         if (dentry->d_inode != NULL)
 722                 ll_dump_inode(dentry->d_inode);
 723
 724         if (recur == 0)
 725                 return;
 726
 727         list_for_each(tmp, &dentry->d_subdirs) {
 728                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 729                 lustre_dump_dentry(d, recur - 1);
 730         }
 731 }
 732
 733 #ifdef HAVE_EXPORT___IGET
 734 static void prune_dir_dentries(struct inode *inode)
 735 {
 736         struct dentry *dentry, *prev = NULL;
 737
 738         /* due to lustre specific logic, a directory
 739          * can have few dentries - a bug from VFS POV */
 740 restart:
 741         spin_lock(&dcache_lock);
 742         if (!list_empty(&inode->i_dentry)) {
 743                 dentry = list_entry(inode->i_dentry.prev,
 744                                     struct dentry, d_alias);
 745                 /* in order to prevent infinite loops we
 746                  * break if previous dentry is busy */
 747                 if (dentry != prev) {
 748                         prev = dentry;
 749                         dget_locked(dentry);
 750                         spin_unlock(&dcache_lock);
 751
 752                         /* try to kill all child dentries */
 753                         lock_dentry(dentry);
 754                         shrink_dcache_parent(dentry);
 755                         unlock_dentry(dentry);
 756                         dput(dentry);
 757
 758                         /* now try to get rid of current dentry */
 759                         d_prune_aliases(inode);
 760                         goto restart;
 761                 }
 762         }
 763         spin_unlock(&dcache_lock);
 764 }
 765
 766 static void prune_deathrow_one(struct ll_inode_info *lli)
 767 {
 768         struct inode *inode = ll_info2i(lli);
 769
 770         /* first, try to drop any dentries - they hold a ref on the inode */
 771         if (S_ISDIR(inode->i_mode))
 772                 prune_dir_dentries(inode);
 773         else
 774                 d_prune_aliases(inode);
 775
 776
 777         /* if somebody still uses it, leave it */
 778         LASSERT(atomic_read(&inode->i_count) > 0);
 779         if (atomic_read(&inode->i_count) > 1)
 780                 goto out;
 781
 782         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 783                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 784
 785         /* seems nobody uses it anymore */
 786         inode->i_nlink = 0;
 787
 788 out:
 789         iput(inode);
 790         return;
 791 }
 792
 793 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 794 {
 795         struct ll_inode_info *lli;
 796         int empty;
 797
 798         do {
 799                 if (need_resched() && try)
 800                         break;
 801
 802                 if (try) {
 803                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 804                                 break;
 805                 } else {
 806                         spin_lock(&sbi->ll_deathrow_lock);
 807                 }
 808
 809                 empty = 1;
 810                 lli = NULL;
 811                 if (!list_empty(&sbi->ll_deathrow)) {
 812                         lli = list_entry(sbi->ll_deathrow.next,
 813                                          struct ll_inode_info,
 814                                          lli_dead_list);
 815                         list_del_init(&lli->lli_dead_list);
 816                         if (!list_empty(&sbi->ll_deathrow))
 817                                 empty = 0;
 818                 }
 819                 spin_unlock(&sbi->ll_deathrow_lock);
 820
 821                 if (lli)
 822                         prune_deathrow_one(lli);
 823
 824         } while (empty == 0);
 825 }
 826 #else /* !HAVE_EXPORT___IGET */
 827 #define prune_deathrow(sbi, try) do {} while (0)
 828 #endif /* HAVE_EXPORT___IGET */
 829
 830 void client_common_put_super(struct super_block *sb)
 831 {
 832         struct ll_sb_info *sbi = ll_s2sbi(sb);
 833         ENTRY;
 834
 835 #ifdef CONFIG_FS_POSIX_ACL
 836         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 837                 et_fini(&sbi->ll_et);
 838                 rct_fini(&sbi->ll_rct);
 839         }
 840 #endif
 841
 842         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 843
 844         ll_close_thread_shutdown(sbi->ll_lcq);
 845
 846         /* destroy inodes in deathrow */
 847         prune_deathrow(sbi, 0);
 848
 849         list_del(&sbi->ll_conn_chain);
 850
 851         obd_fid_fini(sbi->ll_dt_exp);
 852         obd_disconnect(sbi->ll_dt_exp);
 853         sbi->ll_dt_exp = NULL;
 854
 855         lprocfs_unregister_mountpoint(sbi);
 856
 857         obd_fid_fini(sbi->ll_md_exp);
 858         obd_disconnect(sbi->ll_md_exp);
 859         sbi->ll_md_exp = NULL;
 860
 861         EXIT;
 862 }
 863
 864 void ll_kill_super(struct super_block *sb)
 865 {
 866         struct ll_sb_info *sbi;
 867
 868         ENTRY;
 869
 870         /* not init sb ?*/
 871         if (!(sb->s_flags & MS_ACTIVE))
 872                 return;
 873
 874         sbi = ll_s2sbi(sb);
 875         /* we need restore s_dev from changed for clustred NFS before put_super
 876          * because new kernels have cached s_dev and change sb->s_dev in
 877          * put_super not affected real removing devices */
 878         if (sbi)
 879                 sb->s_dev = sbi->ll_sdev_orig;
 880         EXIT;
 881 }
 882
 883 char *ll_read_opt(const char *opt, char *data)
 884 {
 885         char *value;
 886         char *retval;
 887         ENTRY;
 888
 889         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 890         if (strncmp(opt, data, strlen(opt)))
 891                 RETURN(NULL);
 892         if ((value = strchr(data, '=')) == NULL)
 893                 RETURN(NULL);
 894
 895         value++;
 896         OBD_ALLOC(retval, strlen(value) + 1);
 897         if (!retval) {
 898                 CERROR("out of memory!\n");
 899                 RETURN(NULL);
 900         }
 901
 902         memcpy(retval, value, strlen(value)+1);
 903         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 904         RETURN(retval);
 905 }
 906
 907 static inline int ll_set_opt(const char *opt, char *data, int fl)
 908 {
 909         if (strncmp(opt, data, strlen(opt)) != 0)
 910                 return(0);
 911         else
 912                 return(fl);
 913 }
 914
 915 /* non-client-specific mount options are parsed in lmd_parse */
 916 static int ll_options(char *options, int *flags)
 917 {
 918         int tmp;
 919         char *s1 = options, *s2;
 920         ENTRY;
 921
 922         if (!options)
 923                 RETURN(0);
 924
 925         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 926
 927         while (*s1) {
 928                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 929                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 930                 if (tmp) {
 931                         *flags |= tmp;
 932                         goto next;
 933                 }
 934                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 935                 if (tmp) {
 936                         *flags |= tmp;
 937                         goto next;
 938                 }
 939                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 940                 if (tmp) {
 941                         *flags |= tmp;
 942                         goto next;
 943                 }
 944                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 945                 if (tmp) {
 946                         *flags &= ~tmp;
 947                         goto next;
 948                 }
 949                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 950                 if (tmp) {
 951                         *flags |= tmp;
 952                         goto next;
 953                 }
 954                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 955                 if (tmp) {
 956                         *flags &= ~tmp;
 957                         goto next;
 958                 }
 959                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 960                 if (tmp) {
 961                         /* Ignore deprecated mount option.  The client will
 962                          * always try to mount with ACL support, whether this
 963                          * is used depends on whether server supports it. */
 964                         goto next;
 965                 }
 966                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 967                 if (tmp) {
 968                         goto next;
 969                 }
 970                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 971                 if (tmp) {
 972                         *flags |= tmp;
 973                         goto next;
 974                 }
 975
 976                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 977                 if (tmp) {
 978                         *flags |= tmp;
 979                         goto next;
 980                 }
 981                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 982                 if (tmp) {
 983                         *flags &= ~tmp;
 984                         goto next;
 985                 }
 986                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 987                 if (tmp) {
 988                         *flags |= tmp;
 989                         goto next;
 990                 }
 991                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 992                 if (tmp) {
 993                         *flags &= ~tmp;
 994                         goto next;
 995                 }
 996
 997                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 998                                    s1);
 999                 RETURN(-EINVAL);
1000
1001 next:
1002                 /* Find next opt */
1003                 s2 = strchr(s1, ',');
1004                 if (s2 == NULL)
1005                         break;
1006                 s1 = s2 + 1;
1007         }
1008         RETURN(0);
1009 }
1010
1011 void ll_lli_init(struct ll_inode_info *lli)
1012 {
1013         lli->lli_inode_magic = LLI_INODE_MAGIC;
1014         sema_init(&lli->lli_size_sem, 1);
1015         sema_init(&lli->lli_write_sem, 1);
1016         lli->lli_flags = 0;
1017         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1018         spin_lock_init(&lli->lli_lock);
1019         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1020         INIT_LIST_HEAD(&lli->lli_close_list);
1021         lli->lli_inode_magic = LLI_INODE_MAGIC;
1022         sema_init(&lli->lli_och_sem, 1);
1023         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1024         lli->lli_mds_exec_och = NULL;
1025         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1026         lli->lli_open_fd_exec_count = 0;
1027         INIT_LIST_HEAD(&lli->lli_dead_list);
1028         lli->lli_remote_perms = NULL;
1029         lli->lli_rmtperm_utime = 0;
1030         sema_init(&lli->lli_rmtperm_sem, 1);
1031         INIT_LIST_HEAD(&lli->lli_oss_capas);
1032 }
1033
1034 int ll_fill_super(struct super_block *sb)
1035 {
1036         struct lustre_profile *lprof;
1037         struct lustre_sb_info *lsi = s2lsi(sb);
1038         struct ll_sb_info *sbi;
1039         char  *dt = NULL, *md = NULL;
1040         char  *profilenm = get_profile_name(sb);
1041         struct config_llog_instance cfg = {0, };
1042         char   ll_instance[sizeof(sb) * 2 + 1];
1043         int    err;
1044         ENTRY;
1045
1046         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1047
1048         cfs_module_get();
1049
1050         /* client additional sb info */
1051         lsi->lsi_llsbi = sbi = ll_init_sbi();
1052         if (!sbi) {
1053                 cfs_module_put();
1054                 RETURN(-ENOMEM);
1055         }
1056
1057         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1058         if (err)
1059                 GOTO(out_free, err);
1060
1061         /* Generate a string unique to this super, in case some joker tries
1062            to mount the same fs at two mount points.
1063            Use the address of the super itself.*/
1064         sprintf(ll_instance, "%p", sb);
1065         cfg.cfg_instance = ll_instance;
1066         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1067
1068         /* set up client obds */
1069         err = lustre_process_log(sb, profilenm, &cfg);
1070         if (err < 0) {
1071                 CERROR("Unable to process log: %d\n", err);
1072                 GOTO(out_free, err);
1073         }
1074
1075         lprof = class_get_profile(profilenm);
1076         if (lprof == NULL) {
1077                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1078                                    " read from the MGS.  Does that filesystem "
1079                                    "exist?\n", profilenm);
1080                 GOTO(out_free, err = -EINVAL);
1081         }
1082         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1083                lprof->lp_md, lprof->lp_dt);
1084
1085         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1086                   strlen(ll_instance) + 2);
1087         if (!dt)
1088                 GOTO(out_free, err = -ENOMEM);
1089         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1090
1091         OBD_ALLOC(md, strlen(lprof->lp_md) +
1092                   strlen(ll_instance) + 2);
1093         if (!md)
1094                 GOTO(out_free, err = -ENOMEM);
1095         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1096
1097         /* connections, registrations, sb setup */
1098         err = client_common_fill_super(sb, md, dt);
1099
1100 out_free:
1101         if (md)
1102                 OBD_FREE(md, strlen(md) + 1);
1103         if (dt)
1104                 OBD_FREE(dt, strlen(dt) + 1);
1105         if (err)
1106                 ll_put_super(sb);
1107         else
1108                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1109
1110         RETURN(err);
1111 } /* ll_fill_super */
1112
1113
1114 void ll_put_super(struct super_block *sb)
1115 {
1116         struct config_llog_instance cfg;
1117         char   ll_instance[sizeof(sb) * 2 + 1];
1118         struct obd_device *obd;
1119         struct lustre_sb_info *lsi = s2lsi(sb);
1120         struct ll_sb_info *sbi = ll_s2sbi(sb);
1121         char *profilenm = get_profile_name(sb);
1122         int force = 1, next;
1123         ENTRY;
1124
1125         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1126
1127         ll_print_capa_stat(sbi);
1128
1129         sprintf(ll_instance, "%p", sb);
1130         cfg.cfg_instance = ll_instance;
1131         lustre_end_log(sb, NULL, &cfg);
1132
1133         if (sbi->ll_md_exp) {
1134                 obd = class_exp2obd(sbi->ll_md_exp);
1135                 if (obd)
1136                         force = obd->obd_force;
1137         }
1138
1139         /* We need to set force before the lov_disconnect in
1140            lustre_common_put_super, since l_d cleans up osc's as well. */
1141         if (force) {
1142                 next = 0;
1143                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1144                                                      &next)) != NULL) {
1145                         obd->obd_force = force;
1146                 }
1147         }
1148
1149         if (sbi->ll_lcq) {
1150                 /* Only if client_common_fill_super succeeded */
1151                 client_common_put_super(sb);
1152         }
1153         next = 0;
1154         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1155                 class_manual_cleanup(obd);
1156         }
1157
1158         if (profilenm)
1159                 class_del_profile(profilenm);
1160
1161         ll_free_sbi(sb);
1162         lsi->lsi_llsbi = NULL;
1163
1164         lustre_common_put_super(sb);
1165
1166         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1167
1168         cfs_module_put();
1169
1170         EXIT;
1171 } /* client_put_super */
1172
1173 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1174
1175 #if defined(HAVE_CACHE_RETURN_INT)
1176 static int
1177 #else
1178 static void
1179 #endif
1180 ll_shrink_cache(int priority, unsigned int gfp_mask)
1181 {
1182         struct ll_sb_info *sbi;
1183         int count = 0;
1184
1185         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1186                 count += llap_shrink_cache(sbi, priority);
1187
1188 #if defined(HAVE_CACHE_RETURN_INT)
1189         return count;
1190 #endif
1191 }
1192
1193 struct cache_definition ll_cache_definition = {
1194         .name = "llap_cache",
1195         .shrink = ll_shrink_cache
1196 };
1197 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1198
1199 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1200 {
1201         struct inode *inode = NULL;
1202         /* NOTE: we depend on atomic igrab() -bzzz */
1203         lock_res_and_lock(lock);
1204         if (lock->l_ast_data) {
1205                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1206                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1207                         inode = igrab(lock->l_ast_data);
1208                 } else {
1209                         inode = lock->l_ast_data;
1210                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1211                                                 D_INFO : D_WARNING,
1212                                         lock, __FILE__, __func__, __LINE__,
1213                                         "l_ast_data %p is bogus: magic %08x",
1214                                         lock->l_ast_data, lli->lli_inode_magic);
1215                         inode = NULL;
1216                 }
1217         }
1218         unlock_res_and_lock(lock);
1219         return inode;
1220 }
1221
1222 static int null_if_equal(struct ldlm_lock *lock, void *data)
1223 {
1224         if (data == lock->l_ast_data) {
1225                 lock->l_ast_data = NULL;
1226
1227                 if (lock->l_req_mode != lock->l_granted_mode)
1228                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1229         }
1230
1231         return LDLM_ITER_CONTINUE;
1232 }
1233
1234 void ll_clear_inode(struct inode *inode)
1235 {
1236         struct ll_inode_info *lli = ll_i2info(inode);
1237         struct ll_sb_info *sbi = ll_i2sbi(inode);
1238         ENTRY;
1239
1240         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1241                inode->i_generation, inode);
1242
1243         if (S_ISDIR(inode->i_mode)) {
1244                 /* these should have been cleared in ll_file_release */
1245                 LASSERT(lli->lli_sai == NULL);
1246                 LASSERT(lli->lli_opendir_key == NULL);
1247                 LASSERT(lli->lli_opendir_pid == 0);
1248         }
1249
1250         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1251         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1252                          null_if_equal, inode);
1253
1254         LASSERT(!lli->lli_open_fd_write_count);
1255         LASSERT(!lli->lli_open_fd_read_count);
1256         LASSERT(!lli->lli_open_fd_exec_count);
1257
1258         if (lli->lli_mds_write_och)
1259                 ll_md_real_close(inode, FMODE_WRITE);
1260         if (lli->lli_mds_exec_och)
1261                 ll_md_real_close(inode, FMODE_EXEC);
1262         if (lli->lli_mds_read_och)
1263                 ll_md_real_close(inode, FMODE_READ);
1264
1265         if (lli->lli_smd) {
1266                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1267                                   null_if_equal, inode);
1268
1269                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1270                 lli->lli_smd = NULL;
1271         }
1272
1273         if (lli->lli_symlink_name) {
1274                 OBD_FREE(lli->lli_symlink_name,
1275                          strlen(lli->lli_symlink_name) + 1);
1276                 lli->lli_symlink_name = NULL;
1277         }
1278
1279         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1280                 LASSERT(lli->lli_posix_acl == NULL);
1281                 if (lli->lli_remote_perms) {
1282                         free_rmtperm_hash(lli->lli_remote_perms);
1283                         lli->lli_remote_perms = NULL;
1284                 }
1285         }
1286 #ifdef CONFIG_FS_POSIX_ACL
1287         else if (lli->lli_posix_acl) {
1288                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1289                 LASSERT(lli->lli_remote_perms == NULL);
1290                 posix_acl_release(lli->lli_posix_acl);
1291                 lli->lli_posix_acl = NULL;
1292         }
1293 #endif
1294         lli->lli_inode_magic = LLI_INODE_DEAD;
1295
1296 #ifdef HAVE_EXPORT___IGET
1297         spin_lock(&sbi->ll_deathrow_lock);
1298         list_del_init(&lli->lli_dead_list);
1299         spin_unlock(&sbi->ll_deathrow_lock);
1300 #endif
1301         ll_clear_inode_capas(inode);
1302
1303         EXIT;
1304 }
1305
1306 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1307                   struct md_open_data **mod)
1308 {
1309         struct lustre_md md;
1310         struct ll_sb_info *sbi = ll_i2sbi(inode);
1311         struct ptlrpc_request *request = NULL;
1312         int rc;
1313         ENTRY;
1314
1315         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1316                                      LUSTRE_OPC_ANY, NULL);
1317         if (IS_ERR(op_data))
1318                 RETURN(PTR_ERR(op_data));
1319
1320         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1321                         &request, mod);
1322         if (rc) {
1323                 ptlrpc_req_finished(request);
1324                 if (rc == -ENOENT) {
1325                         inode->i_nlink = 0;
1326                         /* Unlinked special device node? Or just a race?
1327                          * Pretend we done everything. */
1328                         if (!S_ISREG(inode->i_mode) &&
1329                             !S_ISDIR(inode->i_mode))
1330                                 rc = inode_setattr(inode, &op_data->op_attr);
1331                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1332                         CERROR("md_setattr fails: rc = %d\n", rc);
1333                 }
1334                 RETURN(rc);
1335         }
1336
1337         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1338                               sbi->ll_md_exp, &md);
1339         if (rc) {
1340                 ptlrpc_req_finished(request);
1341                 RETURN(rc);
1342         }
1343
1344         /* We call inode_setattr to adjust timestamps.
1345          * If there is at least some data in file, we cleared ATTR_SIZE
1346          * above to avoid invoking vmtruncate, otherwise it is important
1347          * to call vmtruncate in inode_setattr to update inode->i_size
1348          * (bug 6196) */
1349         rc = inode_setattr(inode, &op_data->op_attr);
1350
1351         /* Extract epoch data if obtained. */
1352         op_data->op_handle = md.body->handle;
1353         op_data->op_ioepoch = md.body->ioepoch;
1354
1355         ll_update_inode(inode, &md);
1356         ptlrpc_req_finished(request);
1357
1358         RETURN(rc);
1359 }
1360
1361 /* Close IO epoch and send Size-on-MDS attribute update. */
1362 static int ll_setattr_done_writing(struct inode *inode,
1363                                    struct md_op_data *op_data,
1364                                    struct md_open_data *mod)
1365 {
1366         struct ll_inode_info *lli = ll_i2info(inode);
1367         int rc = 0;
1368         ENTRY;
1369
1370         LASSERT(op_data != NULL);
1371         if (!S_ISREG(inode->i_mode))
1372                 RETURN(0);
1373
1374         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1375                op_data->op_ioepoch, PFID(&lli->lli_fid));
1376
1377         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1378         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1379         if (rc == -EAGAIN) {
1380                 /* MDS has instructed us to obtain Size-on-MDS attribute
1381                  * from OSTs and send setattr to back to MDS. */
1382                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1383                                          op_data->op_ioepoch);
1384         } else if (rc) {
1385                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1386                        inode->i_ino, rc);
1387         }
1388         RETURN(rc);
1389 }
1390
1391 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1392 {
1393         struct ll_sb_info *sbi = ll_i2sbi(inode);
1394         struct ll_inode_info *lli = ll_i2info(inode);
1395         struct lov_stripe_md *lsm = lli->lli_smd;
1396         int rc;
1397         ldlm_policy_data_t policy = { .l_extent = {new_size,
1398                                                    OBD_OBJECT_EOF } };
1399         struct lustre_handle lockh = { 0 };
1400         int local_lock = 0; /* 0 - no local lock;
1401                              * 1 - lock taken by lock_extent;
1402                              * 2 - by obd_match*/
1403         int ast_flags;
1404         int err;
1405         ENTRY;
1406
1407         UNLOCK_INODE_MUTEX(inode);
1408         UP_WRITE_I_ALLOC_SEM(inode);
1409
1410         if (sbi->ll_lockless_truncate_enable &&
1411             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1412                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1413                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1414                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1415                 if (rc > 0) {
1416                         local_lock = 2;
1417                         rc = 0;
1418                 } else if (rc == 0) {
1419                         rc = ll_file_punch(inode, new_size, 1);
1420                 }
1421         } else {
1422                 /* XXX when we fix the AST intents to pass the discard-range
1423                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1424                  * XXX here. */
1425                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1426                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1427                                     &lockh, ast_flags);
1428                 if (likely(rc == 0))
1429                         local_lock = 1;
1430         }
1431
1432         LOCK_INODE_MUTEX(inode);
1433         DOWN_WRITE_I_ALLOC_SEM(inode);
1434
1435         if (likely(rc == 0)) {
1436                 /* Only ll_inode_size_lock is taken at this level.
1437                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1438                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1439                  * ll_truncate dropped ll_inode_size_lock() */
1440                 ll_inode_size_lock(inode, 0);
1441                 if (!local_lock) {
1442                         spin_lock(&lli->lli_lock);
1443                         lli->lli_flags |= LLIF_SRVLOCK;
1444                         spin_unlock(&lli->lli_lock);
1445                 }
1446                 rc = vmtruncate(inode, new_size);
1447                 if (!local_lock) {
1448                         spin_lock(&lli->lli_lock);
1449                         lli->lli_flags &= ~LLIF_SRVLOCK;
1450                         spin_unlock(&lli->lli_lock);
1451                 }
1452                 if (rc != 0) {
1453                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1454                         ll_inode_size_unlock(inode, 0);
1455                 }
1456         }
1457
1458         if (local_lock) {
1459                 if (local_lock == 2)
1460                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1461                 else
1462                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1463                 if (unlikely(err != 0)){
1464                         CERROR("extent unlock failed: err=%d,"
1465                                " unlock method =%d\n", err, local_lock);
1466                         if (rc == 0)
1467                                 rc = err;
1468                 }
1469         }
1470         RETURN(rc);
1471 }
1472
1473 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1474  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1475  * keep these values until such a time that objects are allocated for it.
1476  * We do the MDS operations first, as it is checking permissions for us.
1477  * We don't to the MDS RPC if there is nothing that we want to store there,
1478  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1479  * going to do an RPC anyways.
1480  *
1481  * If we are doing a truncate, we will send the mtime and ctime updates
1482  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1483  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1484  * at the same time.
1485  */
1486 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1487 {
1488         struct ll_inode_info *lli = ll_i2info(inode);
1489         struct lov_stripe_md *lsm = lli->lli_smd;
1490         struct ll_sb_info *sbi = ll_i2sbi(inode);
1491         struct md_op_data *op_data = NULL;
1492         struct md_open_data *mod = NULL;
1493         int ia_valid = attr->ia_valid;
1494         int rc = 0, rc1 = 0;
1495         ENTRY;
1496
1497         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1498                attr->ia_valid);
1499         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1500
1501         if (ia_valid & ATTR_SIZE) {
1502                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1503                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1504                                attr->ia_size, ll_file_maxbytes(inode));
1505                         RETURN(-EFBIG);
1506                 }
1507
1508                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1509         }
1510
1511         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1512         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1513                 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1514                         RETURN(-EPERM);
1515         }
1516
1517         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1518         if (attr->ia_valid & ATTR_CTIME) {
1519                 attr->ia_ctime = CURRENT_TIME;
1520                 attr->ia_valid |= ATTR_CTIME_SET;
1521         }
1522         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1523                 attr->ia_atime = CURRENT_TIME;
1524                 attr->ia_valid |= ATTR_ATIME_SET;
1525         }
1526         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1527                 attr->ia_mtime = CURRENT_TIME;
1528                 attr->ia_valid |= ATTR_MTIME_SET;
1529         }
1530         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1531                 /* To avoid stale mtime on mds, obtain it from ost and send
1532                    to mds. */
1533                 rc = ll_glimpse_size(inode, 0);
1534                 if (rc)
1535                         RETURN(rc);
1536
1537                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1538                 attr->ia_mtime = inode->i_mtime;
1539         }
1540
1541         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1542                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1543                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1544                        cfs_time_current_sec());
1545
1546         /* NB: ATTR_SIZE will only be set after this point if the size
1547          * resides on the MDS, ie, this file has no objects. */
1548         if (lsm)
1549                 attr->ia_valid &= ~ATTR_SIZE;
1550
1551         /* We always do an MDS RPC, even if we're only changing the size;
1552          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1553
1554         OBD_ALLOC_PTR(op_data);
1555         if (op_data == NULL)
1556                 RETURN(-ENOMEM);
1557
1558         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1559
1560         /* Open epoch for truncate. */
1561         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1562             (ia_valid & ATTR_SIZE))
1563                 op_data->op_flags = MF_EPOCH_OPEN;
1564
1565         rc = ll_md_setattr(inode, op_data, &mod);
1566         if (rc)
1567                 GOTO(out, rc);
1568
1569         if (op_data->op_ioepoch)
1570                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1571                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1572
1573         if (!lsm || !S_ISREG(inode->i_mode)) {
1574                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1575                 GOTO(out, rc = 0);
1576         }
1577
1578         /* We really need to get our PW lock before we change inode->i_size.
1579          * If we don't we can race with other i_size updaters on our node, like
1580          * ll_file_read.  We can also race with i_size propogation to other
1581          * nodes through dirtying and writeback of final cached pages.  This
1582          * last one is especially bad for racing o_append users on other
1583          * nodes. */
1584         if (ia_valid & ATTR_SIZE) {
1585                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1586         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1587                 obd_flag flags;
1588                 struct obd_info oinfo = { { { 0 } } };
1589                 struct obdo *oa;
1590
1591                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1592                        inode->i_ino, LTIME_S(attr->ia_mtime));
1593
1594                 OBDO_ALLOC(oa);
1595                 if (oa) {
1596                         oa->o_id = lsm->lsm_object_id;
1597                         oa->o_gr = lsm->lsm_object_gr;
1598                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1599
1600                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1601                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1602                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1603                                 OBD_MD_FLGROUP;
1604
1605                         obdo_from_inode(oa, inode, flags);
1606
1607                         oinfo.oi_oa = oa;
1608                         oinfo.oi_md = lsm;
1609                         oinfo.oi_capa = ll_mdscapa_get(inode);
1610
1611                         /* XXX: this looks unnecessary now. */
1612                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1613                         capa_put(oinfo.oi_capa);
1614                         if (rc)
1615                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1616                         OBDO_FREE(oa);
1617                 } else {
1618                         rc = -ENOMEM;
1619                 }
1620         }
1621         EXIT;
1622 out:
1623         if (op_data) {
1624                 if (op_data->op_ioepoch)
1625                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1626                 ll_finish_md_op_data(op_data);
1627         }
1628         return rc ? rc : rc1;
1629 }
1630
1631 int ll_setattr(struct dentry *de, struct iattr *attr)
1632 {
1633         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1634             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1635                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1636
1637         return ll_setattr_raw(de->d_inode, attr);
1638 }
1639
1640 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1641                        __u64 max_age, __u32 flags)
1642 {
1643         struct ll_sb_info *sbi = ll_s2sbi(sb);
1644         struct obd_statfs obd_osfs;
1645         int rc;
1646         ENTRY;
1647
1648         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1649         if (rc) {
1650                 CERROR("md_statfs fails: rc = %d\n", rc);
1651                 RETURN(rc);
1652         }
1653
1654         osfs->os_type = sb->s_magic;
1655
1656         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1657                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1658
1659         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1660                               &obd_osfs, max_age, flags);
1661         if (rc) {
1662                 CERROR("obd_statfs fails: rc = %d\n", rc);
1663                 RETURN(rc);
1664         }
1665
1666         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1667                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1668                obd_osfs.os_files);
1669
1670         osfs->os_bsize = obd_osfs.os_bsize;
1671         osfs->os_blocks = obd_osfs.os_blocks;
1672         osfs->os_bfree = obd_osfs.os_bfree;
1673         osfs->os_bavail = obd_osfs.os_bavail;
1674
1675         /* If we don't have as many objects free on the OST as inodes
1676          * on the MDS, we reduce the total number of inodes to
1677          * compensate, so that the "inodes in use" number is correct.
1678          */
1679         if (obd_osfs.os_ffree < osfs->os_ffree) {
1680                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1681                         obd_osfs.os_ffree;
1682                 osfs->os_ffree = obd_osfs.os_ffree;
1683         }
1684
1685         RETURN(rc);
1686 }
1687 #ifndef HAVE_STATFS_DENTRY_PARAM
1688 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1689 {
1690 #else
1691 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1692 {
1693         struct super_block *sb = de->d_sb;
1694 #endif
1695         struct obd_statfs osfs;
1696         int rc;
1697
1698         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1699         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1700
1701         /* For now we will always get up-to-date statfs values, but in the
1702          * future we may allow some amount of caching on the client (e.g.
1703          * from QOS or lprocfs updates). */
1704         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1705         if (rc)
1706                 return rc;
1707
1708         statfs_unpack(sfs, &osfs);
1709
1710         /* We need to downshift for all 32-bit kernels, because we can't
1711          * tell if the kernel is being called via sys_statfs64() or not.
1712          * Stop before overflowing f_bsize - in which case it is better
1713          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1714         if (sizeof(long) < 8) {
1715                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1716                         sfs->f_bsize <<= 1;
1717
1718                         osfs.os_blocks >>= 1;
1719                         osfs.os_bfree >>= 1;
1720                         osfs.os_bavail >>= 1;
1721                 }
1722         }
1723
1724         sfs->f_blocks = osfs.os_blocks;
1725         sfs->f_bfree = osfs.os_bfree;
1726         sfs->f_bavail = osfs.os_bavail;
1727
1728         return 0;
1729 }
1730
1731 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1732 {
1733         struct ll_inode_info *lli;
1734         struct lov_stripe_md *lsm;
1735
1736         lli = ll_i2info(inode);
1737         LASSERT(lli->lli_size_sem_owner != current);
1738         down(&lli->lli_size_sem);
1739         LASSERT(lli->lli_size_sem_owner == NULL);
1740         lli->lli_size_sem_owner = current;
1741         lsm = lli->lli_smd;
1742         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1743                  lsm, lock_lsm);
1744         if (lock_lsm)
1745                 lov_stripe_lock(lsm);
1746 }
1747
1748 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1749 {
1750         struct ll_inode_info *lli;
1751         struct lov_stripe_md *lsm;
1752
1753         lli = ll_i2info(inode);
1754         lsm = lli->lli_smd;
1755         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1756                  lsm, unlock_lsm);
1757         if (unlock_lsm)
1758                 lov_stripe_unlock(lsm);
1759         LASSERT(lli->lli_size_sem_owner == current);
1760         lli->lli_size_sem_owner = NULL;
1761         up(&lli->lli_size_sem);
1762 }
1763
1764 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1765 {
1766         struct ll_inode_info *lli = ll_i2info(inode);
1767
1768         dump_lsm(D_INODE, lsm);
1769         dump_lsm(D_INODE, lli->lli_smd);
1770         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1771                  "lsm must be joined lsm %p\n", lsm);
1772         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1773         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1774                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1775         lli->lli_smd = lsm;
1776         lli->lli_maxbytes = lsm->lsm_maxbytes;
1777         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1778                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1779 }
1780
1781 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1782 {
1783         struct ll_inode_info *lli = ll_i2info(inode);
1784         struct mdt_body *body = md->body;
1785         struct lov_stripe_md *lsm = md->lsm;
1786         struct ll_sb_info *sbi = ll_i2sbi(inode);
1787
1788         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1789         if (lsm != NULL) {
1790                 if (lli->lli_smd == NULL) {
1791                         if (lsm->lsm_magic != LOV_MAGIC &&
1792                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1793                                 dump_lsm(D_ERROR, lsm);
1794                                 LBUG();
1795                         }
1796                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1797                                lsm, inode->i_ino, inode->i_generation, inode);
1798                         /* ll_inode_size_lock() requires it is only called
1799                          * with lli_smd != NULL or lock_lsm == 0 or we can
1800                          * race between lock/unlock.  bug 9547 */
1801                         lli->lli_smd = lsm;
1802                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1803                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1804                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1805                 } else {
1806                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1807                              lli->lli_smd->lsm_stripe_count ==
1808                                         lsm->lsm_stripe_count) {
1809                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1810                                         CERROR("lsm mismatch for inode %ld\n",
1811                                                 inode->i_ino);
1812                                         CERROR("lli_smd:\n");
1813                                         dump_lsm(D_ERROR, lli->lli_smd);
1814                                         CERROR("lsm:\n");
1815                                         dump_lsm(D_ERROR, lsm);
1816                                         LBUG();
1817                                 }
1818                         } else
1819                                 ll_replace_lsm(inode, lsm);
1820                 }
1821                 if (lli->lli_smd != lsm)
1822                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1823         }
1824
1825         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1826                 if (body->valid & OBD_MD_FLRMTPERM)
1827                         ll_update_remote_perm(inode, md->remote_perm);
1828         }
1829 #ifdef CONFIG_FS_POSIX_ACL
1830         else if (body->valid & OBD_MD_FLACL) {
1831                 spin_lock(&lli->lli_lock);
1832                 if (lli->lli_posix_acl)
1833                         posix_acl_release(lli->lli_posix_acl);
1834                 lli->lli_posix_acl = md->posix_acl;
1835                 spin_unlock(&lli->lli_lock);
1836         }
1837 #endif
1838         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1839
1840         if (body->valid & OBD_MD_FLATIME &&
1841             body->atime > LTIME_S(inode->i_atime))
1842                 LTIME_S(inode->i_atime) = body->atime;
1843
1844         /* mtime is always updated with ctime, but can be set in past.
1845            As write and utime(2) may happen within 1 second, and utime's
1846            mtime has a priority over write's one, so take mtime from mds
1847            for the same ctimes. */
1848         if (body->valid & OBD_MD_FLCTIME &&
1849             body->ctime >= LTIME_S(inode->i_ctime)) {
1850                 LTIME_S(inode->i_ctime) = body->ctime;
1851                 if (body->valid & OBD_MD_FLMTIME) {
1852                         CDEBUG(D_INODE, "setting ino %lu mtime "
1853                                "from %lu to "LPU64"\n", inode->i_ino,
1854                                LTIME_S(inode->i_mtime), body->mtime);
1855                         LTIME_S(inode->i_mtime) = body->mtime;
1856                 }
1857         }
1858         if (body->valid & OBD_MD_FLMODE)
1859                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1860         if (body->valid & OBD_MD_FLTYPE)
1861                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1862         if (S_ISREG(inode->i_mode)) {
1863                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1864         } else {
1865                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1866         }
1867 #ifdef HAVE_INODE_BLKSIZE
1868         inode->i_blksize = 1<<inode->i_blkbits;
1869 #endif
1870         if (body->valid & OBD_MD_FLUID)
1871                 inode->i_uid = body->uid;
1872         if (body->valid & OBD_MD_FLGID)
1873                 inode->i_gid = body->gid;
1874         if (body->valid & OBD_MD_FLFLAGS)
1875                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1876         if (body->valid & OBD_MD_FLNLINK)
1877                 inode->i_nlink = body->nlink;
1878         if (body->valid & OBD_MD_FLRDEV)
1879                 inode->i_rdev = old_decode_dev(body->rdev);
1880
1881         if (body->valid & OBD_MD_FLID) {
1882                 /* FID shouldn't be changed! */
1883                 if (fid_is_sane(&lli->lli_fid)) {
1884                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1885                                  "Trying to change FID "DFID
1886                                  " to the "DFID", inode %lu/%u(%p)\n",
1887                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1888                                  inode->i_ino, inode->i_generation, inode);
1889                 } else
1890                         lli->lli_fid = body->fid1;
1891         }
1892
1893         LASSERT(fid_seq(&lli->lli_fid) != 0);
1894
1895         if (body->valid & OBD_MD_FLSIZE) {
1896                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1897                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1898                         struct lustre_handle lockh;
1899                         ldlm_mode_t mode;
1900
1901                         /* As it is possible a blocking ast has been processed
1902                          * by this time, we need to check there is an UPDATE
1903                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1904                          * it. */
1905                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1906                                                &lockh);
1907                         if (mode) {
1908                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1909                                                       LLIF_EPOCH_PENDING |
1910                                                       LLIF_SOM_DIRTY)) {
1911                                         CERROR("ino %lu flags %lu still has "
1912                                                "size authority! do not trust "
1913                                                "the size got from MDS\n",
1914                                                inode->i_ino, lli->lli_flags);
1915                                 } else {
1916                                         /* Use old size assignment to avoid
1917                                          * deadlock bz14138 & bz14326 */
1918                                         inode->i_size = body->size;
1919                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1920                                 }
1921                                 ldlm_lock_decref(&lockh, mode);
1922                         }
1923                 } else {
1924                         /* Use old size assignment to avoid
1925                          * deadlock bz14138 & bz14326 */
1926                         inode->i_size = body->size;
1927                 }
1928
1929                 if (body->valid & OBD_MD_FLBLOCKS)
1930                         inode->i_blocks = body->blocks;
1931         }
1932
1933         if (body->valid & OBD_MD_FLMDSCAPA) {
1934                 LASSERT(md->mds_capa);
1935                 ll_add_capa(inode, md->mds_capa);
1936         }
1937         if (body->valid & OBD_MD_FLOSSCAPA) {
1938                 LASSERT(md->oss_capa);
1939                 ll_add_capa(inode, md->oss_capa);
1940         }
1941 }
1942
1943 static struct backing_dev_info ll_backing_dev_info = {
1944         .ra_pages       = 0,    /* No readahead */
1945 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1946         .capabilities   = 0,    /* Does contribute to dirty memory */
1947 #else
1948         .memory_backed  = 0,    /* Does contribute to dirty memory */
1949 #endif
1950 };
1951
1952 void ll_read_inode2(struct inode *inode, void *opaque)
1953 {
1954         struct lustre_md *md = opaque;
1955         struct ll_inode_info *lli = ll_i2info(inode);
1956         ENTRY;
1957
1958         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1959                inode->i_ino, inode->i_generation, inode);
1960
1961         ll_lli_init(lli);
1962
1963         LASSERT(!lli->lli_smd);
1964
1965         /* Core attributes from the MDS first.  This is a new inode, and
1966          * the VFS doesn't zero times in the core inode so we have to do
1967          * it ourselves.  They will be overwritten by either MDS or OST
1968          * attributes - we just need to make sure they aren't newer. */
1969         LTIME_S(inode->i_mtime) = 0;
1970         LTIME_S(inode->i_atime) = 0;
1971         LTIME_S(inode->i_ctime) = 0;
1972         inode->i_rdev = 0;
1973         ll_update_inode(inode, md);
1974
1975         /* OIDEBUG(inode); */
1976
1977         if (S_ISREG(inode->i_mode)) {
1978                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1979                 inode->i_op = &ll_file_inode_operations;
1980                 inode->i_fop = sbi->ll_fop;
1981                 inode->i_mapping->a_ops = &ll_aops;
1982                 EXIT;
1983         } else if (S_ISDIR(inode->i_mode)) {
1984                 inode->i_op = &ll_dir_inode_operations;
1985                 inode->i_fop = &ll_dir_operations;
1986                 inode->i_mapping->a_ops = &ll_dir_aops;
1987                 EXIT;
1988         } else if (S_ISLNK(inode->i_mode)) {
1989                 inode->i_op = &ll_fast_symlink_inode_operations;
1990                 EXIT;
1991         } else {
1992                 inode->i_op = &ll_special_inode_operations;
1993
1994                 init_special_inode(inode, inode->i_mode,
1995                                    kdev_t_to_nr(inode->i_rdev));
1996
1997                 /* initializing backing dev info. */
1998                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
1999
2000                 EXIT;
2001         }
2002 }
2003
2004 void ll_delete_inode(struct inode *inode)
2005 {
2006         struct ll_sb_info *sbi = ll_i2sbi(inode);
2007         int rc;
2008         ENTRY;
2009
2010         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2011         if (rc) {
2012                 CERROR("fid_delete() failed, rc %d\n", rc);
2013         }
2014         truncate_inode_pages(&inode->i_data, 0);
2015         clear_inode(inode);
2016
2017         EXIT;
2018 }
2019
2020 int ll_iocontrol(struct inode *inode, struct file *file,
2021                  unsigned int cmd, unsigned long arg)
2022 {
2023         struct ll_sb_info *sbi = ll_i2sbi(inode);
2024         struct ptlrpc_request *req = NULL;
2025         int rc, flags = 0;
2026         ENTRY;
2027
2028         switch(cmd) {
2029         case EXT3_IOC_GETFLAGS: {
2030                 struct mdt_body *body;
2031                 struct obd_capa *oc;
2032
2033                 oc = ll_mdscapa_get(inode);
2034                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2035                                 OBD_MD_FLFLAGS, 0, &req);
2036                 capa_put(oc);
2037                 if (rc) {
2038                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2039                         RETURN(-abs(rc));
2040                 }
2041
2042                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2043
2044                 flags = body->flags;
2045
2046                 ptlrpc_req_finished(req);
2047
2048                 RETURN(put_user(flags, (int *)arg));
2049         }
2050         case EXT3_IOC_SETFLAGS: {
2051                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2052                 struct obd_info oinfo = { { { 0 } } };
2053                 struct md_op_data *op_data;
2054
2055                 if (get_user(flags, (int *)arg))
2056                         RETURN(-EFAULT);
2057
2058                 oinfo.oi_md = lsm;
2059                 OBDO_ALLOC(oinfo.oi_oa);
2060                 if (!oinfo.oi_oa)
2061                         RETURN(-ENOMEM);
2062
2063                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2064                                              LUSTRE_OPC_ANY, NULL);
2065                 if (IS_ERR(op_data))
2066                         RETURN(PTR_ERR(op_data));
2067
2068                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2069                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2070                 rc = md_setattr(sbi->ll_md_exp, op_data,
2071                                 NULL, 0, NULL, 0, &req, NULL);
2072                 ll_finish_md_op_data(op_data);
2073                 ptlrpc_req_finished(req);
2074                 if (rc || lsm == NULL) {
2075                         OBDO_FREE(oinfo.oi_oa);
2076                         RETURN(rc);
2077                 }
2078
2079                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2080                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2081                 oinfo.oi_oa->o_flags = flags;
2082                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2083                                        OBD_MD_FLGROUP;
2084                 oinfo.oi_capa = ll_mdscapa_get(inode);
2085
2086                 obdo_from_inode(oinfo.oi_oa, inode,
2087                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2088                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2089                 capa_put(oinfo.oi_capa);
2090                 OBDO_FREE(oinfo.oi_oa);
2091                 if (rc) {
2092                         if (rc != -EPERM && rc != -EACCES)
2093                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2094                         RETURN(rc);
2095                 }
2096
2097                 inode->i_flags = ll_ext_to_inode_flags(flags |
2098                                                        MDS_BFLAG_EXT_FLAGS);
2099                 RETURN(0);
2100         }
2101         default:
2102                 RETURN(-ENOSYS);
2103         }
2104
2105         RETURN(0);
2106 }
2107
2108 int ll_flush_ctx(struct inode *inode)
2109 {
2110         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2111
2112         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2113
2114         obd_set_info_async(sbi->ll_md_exp,
2115                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2116                            0, NULL, NULL);
2117         obd_set_info_async(sbi->ll_dt_exp,
2118                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2119                            0, NULL, NULL);
2120         return 0;
2121 }
2122
2123 /* umount -f client means force down, don't save state */
2124 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2125 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2126 {
2127         struct super_block *sb = vfsmnt->mnt_sb;
2128 #else
2129 void ll_umount_begin(struct super_block *sb)
2130 {
2131 #endif
2132         struct lustre_sb_info *lsi = s2lsi(sb);
2133         struct ll_sb_info *sbi = ll_s2sbi(sb);
2134         struct obd_device *obd;
2135         struct obd_ioctl_data ioc_data = { 0 };
2136         ENTRY;
2137
2138 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2139         if (!(flags & MNT_FORCE)) {
2140                 EXIT;
2141                 return;
2142         }
2143 #endif
2144
2145         /* Tell the MGC we got umount -f */
2146         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2147
2148         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2149                sb->s_count, atomic_read(&sb->s_active));
2150
2151         obd = class_exp2obd(sbi->ll_md_exp);
2152         if (obd == NULL) {
2153                 CERROR("Invalid MDC connection handle "LPX64"\n",
2154                        sbi->ll_md_exp->exp_handle.h_cookie);
2155                 EXIT;
2156                 return;
2157         }
2158         obd->obd_force = 1;
2159         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2160                       &ioc_data, NULL);
2161
2162         obd = class_exp2obd(sbi->ll_dt_exp);
2163         if (obd == NULL) {
2164                 CERROR("Invalid LOV connection handle "LPX64"\n",
2165                        sbi->ll_dt_exp->exp_handle.h_cookie);
2166                 EXIT;
2167                 return;
2168         }
2169
2170         obd->obd_force = 1;
2171         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2172                       &ioc_data, NULL);
2173
2174         /* Really, we'd like to wait until there are no requests outstanding,
2175          * and then continue.  For now, we just invalidate the requests,
2176          * schedule, and hope.
2177          */
2178         schedule();
2179
2180         EXIT;
2181 }
2182
2183 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2184 {
2185         struct ll_sb_info *sbi = ll_s2sbi(sb);
2186         int err;
2187         __u32 read_only;
2188
2189         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2190                 read_only = *flags & MS_RDONLY;
2191                 err = obd_set_info_async(sbi->ll_md_exp,
2192                                          sizeof(KEY_READ_ONLY),
2193                                          KEY_READ_ONLY, sizeof(read_only),
2194                                          &read_only, NULL);
2195                 if (err) {
2196                         CERROR("Failed to change the read-only flag during "
2197                                "remount: %d\n", err);
2198                         return err;
2199                 }
2200
2201                 if (read_only)
2202                         sb->s_flags |= MS_RDONLY;
2203                 else
2204                         sb->s_flags &= ~MS_RDONLY;
2205         }
2206         return 0;
2207 }
2208
2209 int ll_prep_inode(struct inode **inode,
2210                   struct ptlrpc_request *req,
2211                   struct super_block *sb)
2212 {
2213         struct ll_sb_info *sbi = NULL;
2214         struct lustre_md md;
2215         int rc = 0;
2216         ENTRY;
2217
2218         LASSERT(*inode || sb);
2219         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2220         prune_deathrow(sbi, 1);
2221         memset(&md, 0, sizeof(struct lustre_md));
2222
2223         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2224                               sbi->ll_md_exp, &md);
2225         if (rc)
2226                 RETURN(rc);
2227
2228         if (*inode) {
2229                 ll_update_inode(*inode, &md);
2230         } else {
2231                 LASSERT(sb != NULL);
2232
2233                 /*
2234                  * At this point server returns to client's same fid as client
2235                  * generated for creating. So using ->fid1 is okay here.
2236                  */
2237                 LASSERT(fid_is_sane(&md.body->fid1));
2238
2239                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2240                 if (*inode == NULL || is_bad_inode(*inode)) {
2241                         if (md.lsm)
2242                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2243 #ifdef CONFIG_FS_POSIX_ACL
2244                         if (md.posix_acl) {
2245                                 posix_acl_release(md.posix_acl);
2246                                 md.posix_acl = NULL;
2247                         }
2248 #endif
2249                         rc = -ENOMEM;
2250                         CERROR("new_inode -fatal: rc %d\n", rc);
2251                         GOTO(out, rc);
2252                 }
2253         }
2254
2255         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2256                          ll_i2info(*inode)->lli_smd);
2257 out:
2258         md_free_lustre_md(sbi->ll_md_exp, &md);
2259         RETURN(rc);
2260 }
2261
2262 char *llap_origins[] = {
2263         [LLAP_ORIGIN_UNKNOWN] = "--",
2264         [LLAP_ORIGIN_READPAGE] = "rp",
2265         [LLAP_ORIGIN_READAHEAD] = "ra",
2266         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2267         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2268         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2269 };
2270
2271 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2272                                              struct list_head *list)
2273 {
2274         struct ll_async_page *llap;
2275         struct list_head *pos;
2276
2277         list_for_each(pos, list) {
2278                 if (pos == head)
2279                         return NULL;
2280                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2281                 if (llap->llap_page == NULL)
2282                         continue;
2283                 return llap;
2284         }
2285         LBUG();
2286         return NULL;
2287 }
2288
2289 int ll_obd_statfs(struct inode *inode, void *arg)
2290 {
2291         struct ll_sb_info *sbi = NULL;
2292         struct obd_export *exp;
2293         char *buf = NULL;
2294         struct obd_ioctl_data *data = NULL;
2295         __u32 type;
2296         int len = 0, rc;
2297
2298         if (!inode || !(sbi = ll_i2sbi(inode)))
2299                 GOTO(out_statfs, rc = -EINVAL);
2300
2301         rc = obd_ioctl_getdata(&buf, &len, arg);
2302         if (rc)
2303                 GOTO(out_statfs, rc);
2304
2305         data = (void*)buf;
2306         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2307             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2308                 GOTO(out_statfs, rc = -EINVAL);
2309
2310         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2311         if (type == LL_STATFS_MDC)
2312                 exp = sbi->ll_md_exp;
2313         else if (type == LL_STATFS_LOV)
2314                 exp = sbi->ll_dt_exp;
2315         else
2316                 GOTO(out_statfs, rc = -ENODEV);
2317
2318         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2319         if (rc)
2320                 GOTO(out_statfs, rc);
2321 out_statfs:
2322         if (buf)
2323                 obd_ioctl_freedata(buf, len);
2324         return rc;
2325 }
2326
2327 int ll_process_config(struct lustre_cfg *lcfg)
2328 {
2329         char *ptr;
2330         void *sb;
2331         struct lprocfs_static_vars lvars;
2332         unsigned long x;
2333         int rc = 0;
2334
2335         lprocfs_llite_init_vars(&lvars);
2336
2337         /* The instance name contains the sb: lustre-client-aacfe000 */
2338         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2339         if (!ptr || !*(++ptr))
2340                 return -EINVAL;
2341         if (sscanf(ptr, "%lx", &x) != 1)
2342                 return -EINVAL;
2343         sb = (void *)x;
2344         /* This better be a real Lustre superblock! */
2345         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2346
2347         /* Note we have not called client_common_fill_super yet, so
2348            proc fns must be able to handle that! */
2349         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2350                                       lcfg, sb);
2351         return(rc);
2352 }
2353
2354 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2355 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2356                                        struct inode *i1, struct inode *i2,
2357                                        const char *name, int namelen,
2358                                        int mode, __u32 opc, void *data)
2359 {
2360         LASSERT(i1 != NULL);
2361
2362         if (namelen > ll_i2sbi(i1)->ll_namelen)
2363                 return ERR_PTR(-ENAMETOOLONG);
2364
2365         if (op_data == NULL)
2366                 OBD_ALLOC_PTR(op_data);
2367
2368         if (op_data == NULL)
2369                 return ERR_PTR(-ENOMEM);
2370
2371         ll_i2gids(op_data->op_suppgids, i1, i2);
2372         op_data->op_fid1 = *ll_inode2fid(i1);
2373         op_data->op_capa1 = ll_mdscapa_get(i1);
2374
2375         if (i2) {
2376                 op_data->op_fid2 = *ll_inode2fid(i2);
2377                 op_data->op_capa2 = ll_mdscapa_get(i2);
2378         } else {
2379                 fid_zero(&op_data->op_fid2);
2380                 op_data->op_capa2 = NULL;
2381         }
2382
2383         op_data->op_name = name;
2384         op_data->op_namelen = namelen;
2385         op_data->op_mode = mode;
2386         op_data->op_mod_time = cfs_time_current_sec();
2387         op_data->op_fsuid = current->fsuid;
2388         op_data->op_fsgid = current->fsgid;
2389         op_data->op_cap = current->cap_effective;
2390         op_data->op_bias = MDS_CHECK_SPLIT;
2391         op_data->op_opc = opc;
2392         op_data->op_mds = 0;
2393         op_data->op_data = data;
2394
2395         return op_data;
2396 }
2397
2398 void ll_finish_md_op_data(struct md_op_data *op_data)
2399 {
2400         capa_put(op_data->op_capa1);
2401         capa_put(op_data->op_capa2);
2402         OBD_FREE_PTR(op_data);
2403 }
2404
2405 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2406 {
2407         struct ll_sb_info *sbi;
2408
2409         LASSERT((seq != NULL) && (vfs != NULL));
2410         sbi = ll_s2sbi(vfs->mnt_sb);
2411
2412         if (sbi->ll_flags & LL_SBI_NOLCK)
2413                 seq_puts(seq, ",nolock");
2414
2415         if (sbi->ll_flags & LL_SBI_FLOCK)
2416                 seq_puts(seq, ",flock");
2417
2418         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2419                 seq_puts(seq, ",localflock");
2420
2421         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2422                 seq_puts(seq, ",user_xattr");
2423
2424         if (sbi->ll_flags & LL_SBI_ACL)
2425                 seq_puts(seq, ",acl");
2426
2427         RETURN(0);
2428 }