lustre/llite/llite_lib.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/random.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <obd_cksum.h>
  57 #include <lustre_cache.h>
  58 #include "llite_internal.h"
  59
  60 cfs_mem_cache_t *ll_file_data_slab;
  61
  62 LIST_HEAD(ll_super_blocks);
  63 spinlock_t ll_sb_lock = SPIN_LOCK_UNLOCKED;
  64
  65 extern struct address_space_operations ll_aops;
  66 extern struct address_space_operations ll_dir_aops;
  67
  68 #ifndef log2
  69 #define log2(n) ffz(~(n))
  70 #endif
  71
  72 static inline void ll_pglist_fini(struct ll_sb_info *sbi)
  73 {
  74         struct page *page;
  75         int i;
  76
  77         if (sbi->ll_pglist == NULL)
  78                 return;
  79
  80         for_each_possible_cpu(i) {
  81                 page = sbi->ll_pglist[i]->llpd_page;
  82                 if (page) {
  83                         sbi->ll_pglist[i] = NULL;
  84                         __free_page(page);
  85                 }
  86         }
  87
  88         OBD_FREE(sbi->ll_pglist, sizeof(void *)*num_possible_cpus());
  89         sbi->ll_pglist = NULL;
  90 }
  91
  92 static inline int ll_pglist_init(struct ll_sb_info *sbi)
  93 {
  94         struct ll_pglist_data *pd;
  95         unsigned long budget;
  96         int i, color = 0;
  97         ENTRY;
  98
  99         OBD_ALLOC(sbi->ll_pglist, sizeof(void *) * num_possible_cpus());
 100         if (sbi->ll_pglist == NULL)
 101                 RETURN(-ENOMEM);
 102
 103         budget = sbi->ll_async_page_max / num_online_cpus();
 104         for_each_possible_cpu(i) {
 105                 struct page *page = alloc_pages_node(cpu_to_node(i),
 106                                                     GFP_KERNEL, 0);
 107                 if (page == NULL) {
 108                         ll_pglist_fini(sbi);
 109                         RETURN(-ENOMEM);
 110                 }
 111
 112                 if (color + L1_CACHE_ALIGN(sizeof(*pd)) > PAGE_SIZE)
 113                         color = 0;
 114
 115                 pd = (struct ll_pglist_data *)(page_address(page) + color);
 116                 memset(pd, 0, sizeof(*pd));
 117                 spin_lock_init(&pd->llpd_lock);
 118                 INIT_LIST_HEAD(&pd->llpd_list);
 119                 if (cpu_online(i))
 120                         pd->llpd_budget = budget;
 121                 pd->llpd_cpu = i;
 122                 pd->llpd_page = page;
 123                 atomic_set(&pd->llpd_sample_count, 0);
 124                 sbi->ll_pglist[i] = pd;
 125                 color += L1_CACHE_ALIGN(sizeof(*pd));
 126         }
 127
 128         RETURN(0);
 129 }
 130
 131 static struct ll_sb_info *ll_init_sbi(void)
 132 {
 133         struct ll_sb_info *sbi = NULL;
 134         unsigned long pages;
 135         struct sysinfo si;
 136         class_uuid_t uuid;
 137         int i;
 138         ENTRY;
 139
 140         OBD_ALLOC(sbi, sizeof(*sbi));
 141         if (!sbi)
 142                 RETURN(NULL);
 143
 144         OBD_ALLOC(sbi->ll_async_page_sample, sizeof(long)*num_possible_cpus());
 145         if (sbi->ll_async_page_sample == NULL)
 146                 GOTO(out, 0);
 147
 148         spin_lock_init(&sbi->ll_lock);
 149         spin_lock_init(&sbi->ll_lco.lco_lock);
 150         spin_lock_init(&sbi->ll_pp_extent_lock);
 151         spin_lock_init(&sbi->ll_process_lock);
 152         sbi->ll_rw_stats_on = 0;
 153
 154         si_meminfo(&si);
 155         pages = si.totalram - si.totalhigh;
 156         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
 157 #ifdef HAVE_BGL_SUPPORT
 158                 sbi->ll_async_page_max = pages / 4;
 159 #else
 160                 sbi->ll_async_page_max = pages / 2;
 161 #endif
 162         } else {
 163                 sbi->ll_async_page_max = (pages / 4) * 3;
 164         }
 165         lcounter_init(&sbi->ll_async_page_count);
 166         spin_lock_init(&sbi->ll_async_page_reblnc_lock);
 167         sbi->ll_async_page_sample_max = 64 * num_online_cpus();
 168         sbi->ll_async_page_reblnc_count = 0;
 169         sbi->ll_async_page_clock_hand = 0;
 170         if (ll_pglist_init(sbi))
 171                 GOTO(out, 0);
 172
 173         sbi->ll_ra_info.ra_max_pages = min(pages / 32,
 174                                            SBI_DEFAULT_READAHEAD_MAX);
 175         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 176                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 177         sbi->ll_contention_time = SBI_DEFAULT_CONTENTION_SECONDS;
 178         sbi->ll_lockless_truncate_enable = SBI_DEFAULT_LOCKLESS_TRUNCATE_ENABLE;
 179         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 180         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 181
 182         ll_generate_random_uuid(uuid);
 183         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 184         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 185
 186         spin_lock(&ll_sb_lock);
 187         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 188         spin_unlock(&ll_sb_lock);
 189
 190 #ifdef ENABLE_LLITE_CHECKSUM
 191         sbi->ll_flags |= LL_SBI_CHECKSUM;
 192 #endif
 193
 194 #ifdef HAVE_LRU_RESIZE_SUPPORT
 195         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 196 #endif
 197
 198 #ifdef HAVE_EXPORT___IGET
 199         INIT_LIST_HEAD(&sbi->ll_deathrow);
 200         spin_lock_init(&sbi->ll_deathrow_lock);
 201 #endif
 202         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 203                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_r_hist.oh_lock);
 204                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].pp_w_hist.oh_lock);
 205         }
 206
 207         /* metadata statahead is enabled by default */
 208         sbi->ll_sa_max = LL_SA_RPC_DEF;
 209
 210         RETURN(sbi);
 211
 212 out:
 213         if (sbi->ll_async_page_sample)
 214                 OBD_FREE(sbi->ll_async_page_sample,
 215                          sizeof(long) * num_possible_cpus());
 216         ll_pglist_fini(sbi);
 217         OBD_FREE(sbi, sizeof(*sbi));
 218         RETURN(NULL);
 219 }
 220
 221 void ll_free_sbi(struct super_block *sb)
 222 {
 223         struct ll_sb_info *sbi = ll_s2sbi(sb);
 224         ENTRY;
 225
 226         if (sbi != NULL) {
 227                 ll_pglist_fini(sbi);
 228                 spin_lock(&ll_sb_lock);
 229                 list_del(&sbi->ll_list);
 230                 spin_unlock(&ll_sb_lock);
 231                 lcounter_destroy(&sbi->ll_async_page_count);
 232                 OBD_FREE(sbi->ll_async_page_sample,
 233                          sizeof(long) * num_possible_cpus());
 234                 OBD_FREE(sbi, sizeof(*sbi));
 235         }
 236         EXIT;
 237 }
 238
 239 static struct dentry_operations ll_d_root_ops = {
 240 #ifdef DCACHE_LUSTRE_INVALID
 241         .d_compare = ll_dcompare,
 242 #endif
 243 };
 244
 245 /* Initialize the default and maximum LOV EA and cookie sizes.  This allows
 246  * us to make MDS RPCs with large enough reply buffers to hold the
 247  * maximum-sized (= maximum striped) EA and cookie without having to
 248  * calculate this (via a call into the LOV + OSCs) each time we make an RPC. */
 249 static int ll_init_ea_size(struct obd_export *md_exp, struct obd_export *dt_exp)
 250 {
 251         struct lov_stripe_md lsm = { .lsm_magic = LOV_MAGIC_V3 };
 252         __u32 valsize = sizeof(struct lov_desc);
 253         int rc, easize, def_easize, cookiesize;
 254         struct lov_desc desc;
 255         __u32 stripes;
 256         ENTRY;
 257
 258         rc = obd_get_info(dt_exp, sizeof(KEY_LOVDESC), KEY_LOVDESC,
 259                           &valsize, &desc, NULL);
 260         if (rc)
 261                 RETURN(rc);
 262
 263         stripes = min(desc.ld_tgt_count, (__u32)LOV_MAX_STRIPE_COUNT);
 264         lsm.lsm_stripe_count = stripes;
 265         easize = obd_size_diskmd(dt_exp, &lsm);
 266
 267         lsm.lsm_stripe_count = desc.ld_default_stripe_count;
 268         def_easize = obd_size_diskmd(dt_exp, &lsm);
 269
 270         cookiesize = stripes * sizeof(struct llog_cookie);
 271
 272         CDEBUG(D_HA, "updating max_mdsize/max_cookiesize: %d/%d\n",
 273                easize, cookiesize);
 274
 275         rc = md_init_ea_size(md_exp, easize, def_easize, cookiesize);
 276         RETURN(rc);
 277 }
 278
 279 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 280 {
 281         struct inode *root = 0;
 282         struct ll_sb_info *sbi = ll_s2sbi(sb);
 283         struct obd_device *obd;
 284         struct lu_fid rootfid;
 285         struct obd_capa *oc = NULL;
 286         struct obd_statfs osfs;
 287         struct ptlrpc_request *request = NULL;
 288         struct lustre_handle dt_conn = {0, };
 289         struct lustre_handle md_conn = {0, };
 290         struct obd_connect_data *data = NULL;
 291         struct obd_uuid *uuid;
 292         struct lustre_md lmd;
 293         obd_valid valid;
 294         int size, err, checksum;
 295         ENTRY;
 296
 297         obd = class_name2obd(md);
 298         if (!obd) {
 299                 CERROR("MD %s: not setup or attached\n", md);
 300                 RETURN(-EINVAL);
 301         }
 302
 303         OBD_ALLOC_PTR(data);
 304         if (data == NULL)
 305                 RETURN(-ENOMEM);
 306
 307         if (proc_lustre_fs_root) {
 308                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 309                                                   dt, md);
 310                 if (err < 0)
 311                         CERROR("could not register mount in /proc/fs/lustre\n");
 312         }
 313
 314         /* indicate the features supported by this client */
 315         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 316                                   OBD_CONNECT_JOIN     | OBD_CONNECT_ATTRFID  |
 317                                   OBD_CONNECT_VERSION  | OBD_CONNECT_MDS_CAPA |
 318                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_CANCELSET|
 319                                   OBD_CONNECT_FID      | OBD_CONNECT_AT |
 320                                   OBD_CONNECT_LOV_V3;
 321
 322 #ifdef HAVE_LRU_RESIZE_SUPPORT
 323         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 324                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 325 #endif
 326 #ifdef CONFIG_FS_POSIX_ACL
 327         data->ocd_connect_flags |= OBD_CONNECT_ACL;
 328 #endif
 329         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 330         data->ocd_version = LUSTRE_VERSION_CODE;
 331
 332         if (sb->s_flags & MS_RDONLY)
 333                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 334         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 335                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 336
 337 #ifdef HAVE_MS_FLOCK_LOCK
 338         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 339         sb->s_flags |= MS_FLOCK_LOCK;
 340 #endif
 341
 342         if (sbi->ll_flags & LL_SBI_FLOCK)
 343                 sbi->ll_fop = &ll_file_operations_flock;
 344         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 345                 sbi->ll_fop = &ll_file_operations;
 346         else
 347                 sbi->ll_fop = &ll_file_operations_noflock;
 348
 349         /* real client */
 350         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 351         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 352                 data->ocd_connect_flags &= ~OBD_CONNECT_LCL_CLIENT;
 353                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT;
 354         } else {
 355                 data->ocd_connect_flags &= ~OBD_CONNECT_RMT_CLIENT;
 356                 data->ocd_connect_flags |= OBD_CONNECT_LCL_CLIENT;
 357         }
 358
 359         err = obd_connect(NULL, &md_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 360         if (err == -EBUSY) {
 361                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 362                                    "recovery, of which this client is not a "
 363                                    "part. Please wait for recovery to complete,"
 364                                    " abort, or time out.\n", md);
 365                 GOTO(out, err);
 366         } else if (err) {
 367                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 368                 GOTO(out, err);
 369         }
 370         sbi->ll_md_exp = class_conn2export(&md_conn);
 371
 372         err = obd_fid_init(sbi->ll_md_exp);
 373         if (err) {
 374                 CERROR("Can't init metadata layer FID infrastructure, "
 375                        "rc %d\n", err);
 376                 GOTO(out_md, err);
 377         }
 378
 379         err = obd_statfs(obd, &osfs, cfs_time_current_64() - HZ, 0);
 380         if (err)
 381                 GOTO(out_md_fid, err);
 382
 383         size = sizeof(*data);
 384         err = obd_get_info(sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 385                            KEY_CONN_DATA,  &size, data, NULL);
 386         if (err) {
 387                 CERROR("Get connect data failed: %d \n", err);
 388                 GOTO(out_md, err);
 389         }
 390
 391         LASSERT(osfs.os_bsize);
 392         sb->s_blocksize = osfs.os_bsize;
 393         sb->s_blocksize_bits = log2(osfs.os_bsize);
 394         sb->s_magic = LL_SUPER_MAGIC;
 395
 396         /* for bug 11559. in $LINUX/fs/read_write.c, function do_sendfile():
 397          *         retval = in_file->f_op->sendfile(...);
 398          *         if (*ppos > max)
 399          *                 retval = -EOVERFLOW;
 400          *
 401          * it will check if *ppos is greater than max. However, max equals to
 402          * s_maxbytes, which is a negative integer in a x86_64 box since loff_t
 403          * has been defined as a signed long long ineger in linux kernel. */
 404 #if BITS_PER_LONG == 64
 405         sb->s_maxbytes = PAGE_CACHE_MAXBYTES >> 1;
 406 #else
 407         sb->s_maxbytes = PAGE_CACHE_MAXBYTES;
 408 #endif
 409         sbi->ll_namelen = osfs.os_namelen;
 410         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 411
 412         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 413             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 414                 LCONSOLE_INFO("Disabling user_xattr feature because "
 415                               "it is not supported on the server\n");
 416                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 417         }
 418
 419         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 420 #ifdef MS_POSIXACL
 421                 sb->s_flags |= MS_POSIXACL;
 422 #endif
 423                 sbi->ll_flags |= LL_SBI_ACL;
 424         } else {
 425                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 426 #ifdef MS_POSIXACL
 427                 sb->s_flags &= ~MS_POSIXACL;
 428 #endif
 429                 sbi->ll_flags &= ~LL_SBI_ACL;
 430         }
 431
 432         if (data->ocd_connect_flags & OBD_CONNECT_JOIN)
 433                 sbi->ll_flags |= LL_SBI_JOIN;
 434
 435         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 436                 if (!(data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT)) {
 437                         /* sometimes local client claims to be remote, but mdt
 438                          * will disagree when client gss not applied. */
 439                         LCONSOLE_INFO("client claims to be remote, but server "
 440                                       "rejected, forced to be local.\n");
 441                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 442                 }
 443         } else {
 444                 if (!(data->ocd_connect_flags & OBD_CONNECT_LCL_CLIENT)) {
 445                         /* with gss applied, remote client can not claim to be
 446                          * local, so mdt maybe force client to be remote. */
 447                         LCONSOLE_INFO("client claims to be local, but server "
 448                                       "rejected, forced to be remote.\n");
 449                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 450                 }
 451         }
 452
 453         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 454                 LCONSOLE_INFO("client enabled MDS capability!\n");
 455                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 456         }
 457
 458         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 459                 LCONSOLE_INFO("client enabled OSS capability!\n");
 460                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 461         }
 462
 463         obd = class_name2obd(dt);
 464         if (!obd) {
 465                 CERROR("DT %s: not setup or attached\n", dt);
 466                 GOTO(out_md_fid, err = -ENODEV);
 467         }
 468
 469         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 470                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 471                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 472                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 473                                   OBD_CONNECT_AT;
 474         if (sbi->ll_flags & LL_SBI_OSS_CAPA)
 475                 data->ocd_connect_flags |= OBD_CONNECT_OSS_CAPA;
 476
 477         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 478                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 479                  * disabled by default, because it can still be enabled on the
 480                  * fly via /proc. As a consequence, we still need to come to an
 481                  * agreement on the supported algorithms at connect time */
 482                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 483
 484                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 485                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 486                 else
 487                         /* send the list of supported checksum types */
 488                         data->ocd_cksum_types = OBD_CKSUM_ALL;
 489         }
 490
 491 #ifdef HAVE_LRU_RESIZE_SUPPORT
 492         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 493 #endif
 494         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 495                "ocd_grant: %d\n", data->ocd_connect_flags,
 496                data->ocd_version, data->ocd_grant);
 497
 498         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 499         obd->obd_upcall.onu_upcall = ll_ocd_update;
 500         data->ocd_brw_size = PTLRPC_MAX_BRW_PAGES << CFS_PAGE_SHIFT;
 501
 502         err = obd_connect(NULL, &dt_conn, obd, &sbi->ll_sb_uuid, data, NULL);
 503         if (err == -EBUSY) {
 504                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 505                                    "recovery, of which this client is not a "
 506                                    "part.  Please wait for recovery to "
 507                                    "complete, abort, or time out.\n", dt);
 508                 GOTO(out_md_fid, err);
 509         } else if (err) {
 510                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 511                 GOTO(out_md_fid, err);
 512         }
 513
 514         sbi->ll_dt_exp = class_conn2export(&dt_conn);
 515
 516         err = obd_fid_init(sbi->ll_dt_exp);
 517         if (err) {
 518                 CERROR("Can't init data layer FID infrastructure, "
 519                        "rc %d\n", err);
 520                 GOTO(out_dt, err);
 521         }
 522
 523         spin_lock(&sbi->ll_lco.lco_lock);
 524         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 525         spin_unlock(&sbi->ll_lco.lco_lock);
 526
 527         err = obd_register_page_removal_cb(sbi->ll_dt_exp,
 528                                            ll_page_removal_cb,
 529                                            ll_pin_extent_cb);
 530         if (err) {
 531                 CERROR("cannot register page removal callback: rc = %d\n",err);
 532                 GOTO(out_dt, err);
 533         }
 534         err = obd_register_lock_cancel_cb(sbi->ll_dt_exp,
 535                                           ll_extent_lock_cancel_cb);
 536         if (err) {
 537                 CERROR("cannot register lock cancel callback: rc = %d\n", err);
 538                 GOTO(out_page_rm_cb, err);
 539         }
 540
 541         err = ll_init_ea_size(sbi->ll_md_exp, sbi->ll_dt_exp);;
 542         if (err) {
 543                 CERROR("cannot set max EA and cookie sizes: rc = %d\n", err);
 544                 GOTO(out_lock_cn_cb, err);
 545         }
 546
 547         err = obd_prep_async_page(sbi->ll_dt_exp, NULL, NULL, NULL,
 548                                   0, NULL, NULL, NULL, 0, NULL);
 549         if (err < 0) {
 550                 LCONSOLE_ERROR_MSG(0x151, "There are no OST's in this "
 551                                    "filesystem. There must be at least one "
 552                                    "active OST for a client to start.\n");
 553                 GOTO(out_lock_cn_cb, err);
 554         }
 555
 556         if (!ll_async_page_slab) {
 557                 ll_async_page_slab_size =
 558                         size_round(sizeof(struct ll_async_page)) + err;
 559                 ll_async_page_slab = cfs_mem_cache_create("ll_async_page",
 560                                                           ll_async_page_slab_size,
 561                                                           0, 0);
 562                 if (!ll_async_page_slab)
 563                         GOTO(out_lock_cn_cb, err = -ENOMEM);
 564         }
 565
 566         err = md_getstatus(sbi->ll_md_exp, &rootfid, &oc);
 567         if (err) {
 568                 CERROR("cannot mds_connect: rc = %d\n", err);
 569                 GOTO(out_lock_cn_cb, err);
 570         }
 571         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&rootfid));
 572         sbi->ll_root_fid = rootfid;
 573
 574         sb->s_op = &lustre_super_operations;
 575         sb->s_export_op = &lustre_export_operations;
 576
 577         /* make root inode
 578          * XXX: move this to after cbd setup? */
 579         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 580         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 581                 valid |= OBD_MD_FLRMTPERM;
 582         else if (sbi->ll_flags & LL_SBI_ACL)
 583                 valid |= OBD_MD_FLACL;
 584
 585         err = md_getattr(sbi->ll_md_exp, &rootfid, oc, valid, 0, &request);
 586         if (oc)
 587                 free_capa(oc);
 588         if (err) {
 589                 CERROR("md_getattr failed for root: rc = %d\n", err);
 590                 GOTO(out_lock_cn_cb, err);
 591         }
 592         memset(&lmd, 0, sizeof(lmd));
 593         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 594                                sbi->ll_md_exp, &lmd);
 595         if (err) {
 596                 CERROR("failed to understand root inode md: rc = %d\n", err);
 597                 ptlrpc_req_finished (request);
 598                 GOTO(out_lock_cn_cb, err);
 599         }
 600
 601         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 602         root = ll_iget(sb, ll_fid_build_ino(sbi, &sbi->ll_root_fid), &lmd);
 603         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 604         ptlrpc_req_finished(request);
 605
 606         if (root == NULL || is_bad_inode(root)) {
 607                 if (lmd.lsm)
 608                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 609 #ifdef CONFIG_FS_POSIX_ACL
 610                 if (lmd.posix_acl) {
 611                         posix_acl_release(lmd.posix_acl);
 612                         lmd.posix_acl = NULL;
 613                 }
 614 #endif
 615                 CERROR("lustre_lite: bad iget4 for root\n");
 616                 GOTO(out_root, err = -EBADF);
 617         }
 618
 619         err = ll_close_thread_start(&sbi->ll_lcq);
 620         if (err) {
 621                 CERROR("cannot start close thread: rc %d\n", err);
 622                 GOTO(out_root, err);
 623         }
 624
 625 #ifdef CONFIG_FS_POSIX_ACL
 626         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 627                 rct_init(&sbi->ll_rct);
 628                 et_init(&sbi->ll_et);
 629         }
 630 #endif
 631
 632         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 633         err = obd_set_info_async(sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 634                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 635                                  NULL);
 636
 637         sb->s_root = d_alloc_root(root);
 638         if (data != NULL)
 639                 OBD_FREE(data, sizeof(*data));
 640
 641         sb->s_root->d_op = &ll_d_root_ops;
 642
 643         sbi->ll_sdev_orig = sb->s_dev;
 644
 645         /* We set sb->s_dev equal on all lustre clients in order to support
 646          * NFS export clustering.  NFSD requires that the FSID be the same
 647          * on all clients. */
 648         /* s_dev is also used in lt_compare() to compare two fs, but that is
 649          * only a node-local comparison. */
 650         uuid = obd_get_uuid(sbi->ll_md_exp);
 651         if (uuid != NULL)
 652                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 653
 654         RETURN(err);
 655 out_root:
 656         if (root)
 657                 iput(root);
 658 out_lock_cn_cb:
 659         obd_unregister_lock_cancel_cb(sbi->ll_dt_exp,
 660                                       ll_extent_lock_cancel_cb);
 661 out_page_rm_cb:
 662         obd_unregister_page_removal_cb(sbi->ll_dt_exp,
 663                                        ll_page_removal_cb);
 664         obd_fid_fini(sbi->ll_dt_exp);
 665 out_dt:
 666         obd_disconnect(sbi->ll_dt_exp);
 667         sbi->ll_dt_exp = NULL;
 668 out_md_fid:
 669         obd_fid_fini(sbi->ll_md_exp);
 670 out_md:
 671         obd_disconnect(sbi->ll_md_exp);
 672         sbi->ll_md_exp = NULL;
 673 out:
 674         if (data != NULL)
 675                 OBD_FREE_PTR(data);
 676         lprocfs_unregister_mountpoint(sbi);
 677         return err;
 678 }
 679
 680 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 681 {
 682         int size, rc;
 683
 684         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 685         size = sizeof(int);
 686         rc = obd_get_info(sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 687                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 688         if (rc)
 689                 CERROR("Get max mdsize error rc %d \n", rc);
 690
 691         RETURN(rc);
 692 }
 693
 694 void ll_dump_inode(struct inode *inode)
 695 {
 696         struct list_head *tmp;
 697         int dentry_count = 0;
 698
 699         LASSERT(inode != NULL);
 700
 701         list_for_each(tmp, &inode->i_dentry)
 702                 dentry_count++;
 703
 704         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 705                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 706                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 707 }
 708
 709 void lustre_dump_dentry(struct dentry *dentry, int recur)
 710 {
 711         struct list_head *tmp;
 712         int subdirs = 0;
 713
 714         LASSERT(dentry != NULL);
 715
 716         list_for_each(tmp, &dentry->d_subdirs)
 717                 subdirs++;
 718
 719         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 720                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 721                dentry->d_name.len, dentry->d_name.name,
 722                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 723                dentry->d_parent, dentry->d_inode, atomic_read(&dentry->d_count),
 724                dentry->d_flags, dentry->d_fsdata, subdirs);
 725         if (dentry->d_inode != NULL)
 726                 ll_dump_inode(dentry->d_inode);
 727
 728         if (recur == 0)
 729                 return;
 730
 731         list_for_each(tmp, &dentry->d_subdirs) {
 732                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 733                 lustre_dump_dentry(d, recur - 1);
 734         }
 735 }
 736
 737 #ifdef HAVE_EXPORT___IGET
 738 static void prune_dir_dentries(struct inode *inode)
 739 {
 740         struct dentry *dentry, *prev = NULL;
 741
 742         /* due to lustre specific logic, a directory
 743          * can have few dentries - a bug from VFS POV */
 744 restart:
 745         spin_lock(&dcache_lock);
 746         if (!list_empty(&inode->i_dentry)) {
 747                 dentry = list_entry(inode->i_dentry.prev,
 748                                     struct dentry, d_alias);
 749                 /* in order to prevent infinite loops we
 750                  * break if previous dentry is busy */
 751                 if (dentry != prev) {
 752                         prev = dentry;
 753                         dget_locked(dentry);
 754                         spin_unlock(&dcache_lock);
 755
 756                         /* try to kill all child dentries */
 757                         lock_dentry(dentry);
 758                         shrink_dcache_parent(dentry);
 759                         unlock_dentry(dentry);
 760                         dput(dentry);
 761
 762                         /* now try to get rid of current dentry */
 763                         d_prune_aliases(inode);
 764                         goto restart;
 765                 }
 766         }
 767         spin_unlock(&dcache_lock);
 768 }
 769
 770 static void prune_deathrow_one(struct ll_inode_info *lli)
 771 {
 772         struct inode *inode = ll_info2i(lli);
 773
 774         /* first, try to drop any dentries - they hold a ref on the inode */
 775         if (S_ISDIR(inode->i_mode))
 776                 prune_dir_dentries(inode);
 777         else
 778                 d_prune_aliases(inode);
 779
 780
 781         /* if somebody still uses it, leave it */
 782         LASSERT(atomic_read(&inode->i_count) > 0);
 783         if (atomic_read(&inode->i_count) > 1)
 784                 goto out;
 785
 786         CDEBUG(D_INODE, "inode %lu/%u(%d) looks a good candidate for prune\n",
 787                inode->i_ino,inode->i_generation, atomic_read(&inode->i_count));
 788
 789         /* seems nobody uses it anymore */
 790         inode->i_nlink = 0;
 791
 792 out:
 793         iput(inode);
 794         return;
 795 }
 796
 797 static void prune_deathrow(struct ll_sb_info *sbi, int try)
 798 {
 799         struct ll_inode_info *lli;
 800         int empty;
 801
 802         do {
 803                 if (need_resched() && try)
 804                         break;
 805
 806                 if (try) {
 807                         if (!spin_trylock(&sbi->ll_deathrow_lock))
 808                                 break;
 809                 } else {
 810                         spin_lock(&sbi->ll_deathrow_lock);
 811                 }
 812
 813                 empty = 1;
 814                 lli = NULL;
 815                 if (!list_empty(&sbi->ll_deathrow)) {
 816                         lli = list_entry(sbi->ll_deathrow.next,
 817                                          struct ll_inode_info,
 818                                          lli_dead_list);
 819                         list_del_init(&lli->lli_dead_list);
 820                         if (!list_empty(&sbi->ll_deathrow))
 821                                 empty = 0;
 822                 }
 823                 spin_unlock(&sbi->ll_deathrow_lock);
 824
 825                 if (lli)
 826                         prune_deathrow_one(lli);
 827
 828         } while (empty == 0);
 829 }
 830 #else /* !HAVE_EXPORT___IGET */
 831 #define prune_deathrow(sbi, try) do {} while (0)
 832 #endif /* HAVE_EXPORT___IGET */
 833
 834 void client_common_put_super(struct super_block *sb)
 835 {
 836         struct ll_sb_info *sbi = ll_s2sbi(sb);
 837         ENTRY;
 838
 839 #ifdef CONFIG_FS_POSIX_ACL
 840         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 841                 et_fini(&sbi->ll_et);
 842                 rct_fini(&sbi->ll_rct);
 843         }
 844 #endif
 845
 846         obd_cancel_unused(sbi->ll_dt_exp, NULL, 0, NULL);
 847
 848         ll_close_thread_shutdown(sbi->ll_lcq);
 849
 850         /* destroy inodes in deathrow */
 851         prune_deathrow(sbi, 0);
 852
 853         list_del(&sbi->ll_conn_chain);
 854
 855         obd_fid_fini(sbi->ll_dt_exp);
 856         obd_disconnect(sbi->ll_dt_exp);
 857         sbi->ll_dt_exp = NULL;
 858
 859         lprocfs_unregister_mountpoint(sbi);
 860
 861         obd_fid_fini(sbi->ll_md_exp);
 862         obd_disconnect(sbi->ll_md_exp);
 863         sbi->ll_md_exp = NULL;
 864
 865         EXIT;
 866 }
 867
 868 void ll_kill_super(struct super_block *sb)
 869 {
 870         struct ll_sb_info *sbi;
 871
 872         ENTRY;
 873
 874         /* not init sb ?*/
 875         if (!(sb->s_flags & MS_ACTIVE))
 876                 return;
 877
 878         sbi = ll_s2sbi(sb);
 879         /* we need restore s_dev from changed for clustred NFS before put_super
 880          * because new kernels have cached s_dev and change sb->s_dev in
 881          * put_super not affected real removing devices */
 882         if (sbi)
 883                 sb->s_dev = sbi->ll_sdev_orig;
 884         EXIT;
 885 }
 886
 887 char *ll_read_opt(const char *opt, char *data)
 888 {
 889         char *value;
 890         char *retval;
 891         ENTRY;
 892
 893         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 894         if (strncmp(opt, data, strlen(opt)))
 895                 RETURN(NULL);
 896         if ((value = strchr(data, '=')) == NULL)
 897                 RETURN(NULL);
 898
 899         value++;
 900         OBD_ALLOC(retval, strlen(value) + 1);
 901         if (!retval) {
 902                 CERROR("out of memory!\n");
 903                 RETURN(NULL);
 904         }
 905
 906         memcpy(retval, value, strlen(value)+1);
 907         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 908         RETURN(retval);
 909 }
 910
 911 static inline int ll_set_opt(const char *opt, char *data, int fl)
 912 {
 913         if (strncmp(opt, data, strlen(opt)) != 0)
 914                 return(0);
 915         else
 916                 return(fl);
 917 }
 918
 919 /* non-client-specific mount options are parsed in lmd_parse */
 920 static int ll_options(char *options, int *flags)
 921 {
 922         int tmp;
 923         char *s1 = options, *s2;
 924         ENTRY;
 925
 926         if (!options)
 927                 RETURN(0);
 928
 929         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 930
 931         while (*s1) {
 932                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 933                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 934                 if (tmp) {
 935                         *flags |= tmp;
 936                         goto next;
 937                 }
 938                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 939                 if (tmp) {
 940                         *flags |= tmp;
 941                         goto next;
 942                 }
 943                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 944                 if (tmp) {
 945                         *flags |= tmp;
 946                         goto next;
 947                 }
 948                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 949                 if (tmp) {
 950                         *flags &= ~tmp;
 951                         goto next;
 952                 }
 953                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 954                 if (tmp) {
 955                         *flags |= tmp;
 956                         goto next;
 957                 }
 958                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 959                 if (tmp) {
 960                         *flags &= ~tmp;
 961                         goto next;
 962                 }
 963                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 964                 if (tmp) {
 965                         /* Ignore deprecated mount option.  The client will
 966                          * always try to mount with ACL support, whether this
 967                          * is used depends on whether server supports it. */
 968                         goto next;
 969                 }
 970                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 971                 if (tmp) {
 972                         goto next;
 973                 }
 974                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 975                 if (tmp) {
 976                         *flags |= tmp;
 977                         goto next;
 978                 }
 979
 980                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 981                 if (tmp) {
 982                         *flags |= tmp;
 983                         goto next;
 984                 }
 985                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 986                 if (tmp) {
 987                         *flags &= ~tmp;
 988                         goto next;
 989                 }
 990                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 991                 if (tmp) {
 992                         *flags |= tmp;
 993                         goto next;
 994                 }
 995                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 996                 if (tmp) {
 997                         *flags &= ~tmp;
 998                         goto next;
 999                 }
1000
1001                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1002                                    s1);
1003                 RETURN(-EINVAL);
1004
1005 next:
1006                 /* Find next opt */
1007                 s2 = strchr(s1, ',');
1008                 if (s2 == NULL)
1009                         break;
1010                 s1 = s2 + 1;
1011         }
1012         RETURN(0);
1013 }
1014
1015 void ll_lli_init(struct ll_inode_info *lli)
1016 {
1017         lli->lli_inode_magic = LLI_INODE_MAGIC;
1018         sema_init(&lli->lli_size_sem, 1);
1019         sema_init(&lli->lli_write_sem, 1);
1020         lli->lli_flags = 0;
1021         lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1022         spin_lock_init(&lli->lli_lock);
1023         INIT_LIST_HEAD(&lli->lli_pending_write_llaps);
1024         INIT_LIST_HEAD(&lli->lli_close_list);
1025         lli->lli_inode_magic = LLI_INODE_MAGIC;
1026         sema_init(&lli->lli_och_sem, 1);
1027         lli->lli_mds_read_och = lli->lli_mds_write_och = NULL;
1028         lli->lli_mds_exec_och = NULL;
1029         lli->lli_open_fd_read_count = lli->lli_open_fd_write_count = 0;
1030         lli->lli_open_fd_exec_count = 0;
1031         INIT_LIST_HEAD(&lli->lli_dead_list);
1032         lli->lli_remote_perms = NULL;
1033         lli->lli_rmtperm_utime = 0;
1034         sema_init(&lli->lli_rmtperm_sem, 1);
1035         INIT_LIST_HEAD(&lli->lli_oss_capas);
1036 }
1037
1038 int ll_fill_super(struct super_block *sb)
1039 {
1040         struct lustre_profile *lprof;
1041         struct lustre_sb_info *lsi = s2lsi(sb);
1042         struct ll_sb_info *sbi;
1043         char  *dt = NULL, *md = NULL;
1044         char  *profilenm = get_profile_name(sb);
1045         struct config_llog_instance cfg = {0, };
1046         char   ll_instance[sizeof(sb) * 2 + 1];
1047         int    err;
1048         ENTRY;
1049
1050         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1051
1052         cfs_module_get();
1053
1054         /* client additional sb info */
1055         lsi->lsi_llsbi = sbi = ll_init_sbi();
1056         if (!sbi) {
1057                 cfs_module_put();
1058                 RETURN(-ENOMEM);
1059         }
1060
1061         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1062         if (err)
1063                 GOTO(out_free, err);
1064
1065         /* Generate a string unique to this super, in case some joker tries
1066            to mount the same fs at two mount points.
1067            Use the address of the super itself.*/
1068         sprintf(ll_instance, "%p", sb);
1069         cfg.cfg_instance = ll_instance;
1070         cfg.cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1071
1072         /* set up client obds */
1073         err = lustre_process_log(sb, profilenm, &cfg);
1074         if (err < 0) {
1075                 CERROR("Unable to process log: %d\n", err);
1076                 GOTO(out_free, err);
1077         }
1078
1079         lprof = class_get_profile(profilenm);
1080         if (lprof == NULL) {
1081                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1082                                    " read from the MGS.  Does that filesystem "
1083                                    "exist?\n", profilenm);
1084                 GOTO(out_free, err = -EINVAL);
1085         }
1086         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1087                lprof->lp_md, lprof->lp_dt);
1088
1089         OBD_ALLOC(dt, strlen(lprof->lp_dt) +
1090                   strlen(ll_instance) + 2);
1091         if (!dt)
1092                 GOTO(out_free, err = -ENOMEM);
1093         sprintf(dt, "%s-%s", lprof->lp_dt, ll_instance);
1094
1095         OBD_ALLOC(md, strlen(lprof->lp_md) +
1096                   strlen(ll_instance) + 2);
1097         if (!md)
1098                 GOTO(out_free, err = -ENOMEM);
1099         sprintf(md, "%s-%s", lprof->lp_md, ll_instance);
1100
1101         /* connections, registrations, sb setup */
1102         err = client_common_fill_super(sb, md, dt);
1103
1104 out_free:
1105         if (md)
1106                 OBD_FREE(md, strlen(md) + 1);
1107         if (dt)
1108                 OBD_FREE(dt, strlen(dt) + 1);
1109         if (err)
1110                 ll_put_super(sb);
1111         else
1112                 LCONSOLE_WARN("Client %s has started\n", profilenm);
1113
1114         RETURN(err);
1115 } /* ll_fill_super */
1116
1117
1118 void ll_put_super(struct super_block *sb)
1119 {
1120         struct config_llog_instance cfg;
1121         char   ll_instance[sizeof(sb) * 2 + 1];
1122         struct obd_device *obd;
1123         struct lustre_sb_info *lsi = s2lsi(sb);
1124         struct ll_sb_info *sbi = ll_s2sbi(sb);
1125         char *profilenm = get_profile_name(sb);
1126         int force = 1, next;
1127         ENTRY;
1128
1129         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1130
1131         ll_print_capa_stat(sbi);
1132
1133         sprintf(ll_instance, "%p", sb);
1134         cfg.cfg_instance = ll_instance;
1135         lustre_end_log(sb, NULL, &cfg);
1136
1137         if (sbi->ll_md_exp) {
1138                 obd = class_exp2obd(sbi->ll_md_exp);
1139                 if (obd)
1140                         force = obd->obd_force;
1141         }
1142
1143         /* We need to set force before the lov_disconnect in
1144            lustre_common_put_super, since l_d cleans up osc's as well. */
1145         if (force) {
1146                 next = 0;
1147                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1148                                                      &next)) != NULL) {
1149                         obd->obd_force = force;
1150                 }
1151         }
1152
1153         if (sbi->ll_lcq) {
1154                 /* Only if client_common_fill_super succeeded */
1155                 client_common_put_super(sb);
1156         }
1157         next = 0;
1158         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1159                 class_manual_cleanup(obd);
1160         }
1161
1162         if (profilenm)
1163                 class_del_profile(profilenm);
1164
1165         ll_free_sbi(sb);
1166         lsi->lsi_llsbi = NULL;
1167
1168         lustre_common_put_super(sb);
1169
1170         LCONSOLE_WARN("client %s umount complete\n", ll_instance);
1171
1172         cfs_module_put();
1173
1174         EXIT;
1175 } /* client_put_super */
1176
1177 #if defined(HAVE_REGISTER_CACHE) || defined(HAVE_SHRINKER_CACHE)
1178
1179 #if defined(HAVE_CACHE_RETURN_INT)
1180 static int
1181 #else
1182 static void
1183 #endif
1184 ll_shrink_cache(int priority, unsigned int gfp_mask)
1185 {
1186         struct ll_sb_info *sbi;
1187         int count = 0;
1188
1189         list_for_each_entry(sbi, &ll_super_blocks, ll_list)
1190                 count += llap_shrink_cache(sbi, priority);
1191
1192 #if defined(HAVE_CACHE_RETURN_INT)
1193         return count;
1194 #endif
1195 }
1196
1197 struct cache_definition ll_cache_definition = {
1198         .name = "llap_cache",
1199         .shrink = ll_shrink_cache
1200 };
1201 #endif /* HAVE_REGISTER_CACHE || HAVE_SHRINKER_CACHE */
1202
1203 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1204 {
1205         struct inode *inode = NULL;
1206         /* NOTE: we depend on atomic igrab() -bzzz */
1207         lock_res_and_lock(lock);
1208         if (lock->l_ast_data) {
1209                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1210                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1211                         inode = igrab(lock->l_ast_data);
1212                 } else {
1213                         inode = lock->l_ast_data;
1214                         ldlm_lock_debug(NULL, inode->i_state & I_FREEING ?
1215                                                 D_INFO : D_WARNING,
1216                                         lock, __FILE__, __func__, __LINE__,
1217                                         "l_ast_data %p is bogus: magic %08x",
1218                                         lock->l_ast_data, lli->lli_inode_magic);
1219                         inode = NULL;
1220                 }
1221         }
1222         unlock_res_and_lock(lock);
1223         return inode;
1224 }
1225
1226 static int null_if_equal(struct ldlm_lock *lock, void *data)
1227 {
1228         if (data == lock->l_ast_data) {
1229                 lock->l_ast_data = NULL;
1230
1231                 if (lock->l_req_mode != lock->l_granted_mode)
1232                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1233         }
1234
1235         return LDLM_ITER_CONTINUE;
1236 }
1237
1238 void ll_clear_inode(struct inode *inode)
1239 {
1240         struct ll_inode_info *lli = ll_i2info(inode);
1241         struct ll_sb_info *sbi = ll_i2sbi(inode);
1242         ENTRY;
1243
1244         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1245                inode->i_generation, inode);
1246
1247         if (S_ISDIR(inode->i_mode)) {
1248                 /* these should have been cleared in ll_file_release */
1249                 LASSERT(lli->lli_sai == NULL);
1250                 LASSERT(lli->lli_opendir_key == NULL);
1251                 LASSERT(lli->lli_opendir_pid == 0);
1252         }
1253
1254         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1255         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1256                          null_if_equal, inode);
1257
1258         LASSERT(!lli->lli_open_fd_write_count);
1259         LASSERT(!lli->lli_open_fd_read_count);
1260         LASSERT(!lli->lli_open_fd_exec_count);
1261
1262         if (lli->lli_mds_write_och)
1263                 ll_md_real_close(inode, FMODE_WRITE);
1264         if (lli->lli_mds_exec_och)
1265                 ll_md_real_close(inode, FMODE_EXEC);
1266         if (lli->lli_mds_read_och)
1267                 ll_md_real_close(inode, FMODE_READ);
1268
1269         if (lli->lli_smd) {
1270                 obd_change_cbdata(sbi->ll_dt_exp, lli->lli_smd,
1271                                   null_if_equal, inode);
1272
1273                 obd_free_memmd(sbi->ll_dt_exp, &lli->lli_smd);
1274                 lli->lli_smd = NULL;
1275         }
1276
1277         if (lli->lli_symlink_name) {
1278                 OBD_FREE(lli->lli_symlink_name,
1279                          strlen(lli->lli_symlink_name) + 1);
1280                 lli->lli_symlink_name = NULL;
1281         }
1282
1283         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1284                 LASSERT(lli->lli_posix_acl == NULL);
1285                 if (lli->lli_remote_perms) {
1286                         free_rmtperm_hash(lli->lli_remote_perms);
1287                         lli->lli_remote_perms = NULL;
1288                 }
1289         }
1290 #ifdef CONFIG_FS_POSIX_ACL
1291         else if (lli->lli_posix_acl) {
1292                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1293                 LASSERT(lli->lli_remote_perms == NULL);
1294                 posix_acl_release(lli->lli_posix_acl);
1295                 lli->lli_posix_acl = NULL;
1296         }
1297 #endif
1298         lli->lli_inode_magic = LLI_INODE_DEAD;
1299
1300 #ifdef HAVE_EXPORT___IGET
1301         spin_lock(&sbi->ll_deathrow_lock);
1302         list_del_init(&lli->lli_dead_list);
1303         spin_unlock(&sbi->ll_deathrow_lock);
1304 #endif
1305         ll_clear_inode_capas(inode);
1306
1307         EXIT;
1308 }
1309
1310 int ll_md_setattr(struct inode *inode, struct md_op_data *op_data,
1311                   struct md_open_data **mod)
1312 {
1313         struct lustre_md md;
1314         struct ll_sb_info *sbi = ll_i2sbi(inode);
1315         struct ptlrpc_request *request = NULL;
1316         int rc;
1317         ENTRY;
1318
1319         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1320                                      LUSTRE_OPC_ANY, NULL);
1321         if (IS_ERR(op_data))
1322                 RETURN(PTR_ERR(op_data));
1323
1324         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1325                         &request, mod);
1326         if (rc) {
1327                 ptlrpc_req_finished(request);
1328                 if (rc == -ENOENT) {
1329                         inode->i_nlink = 0;
1330                         /* Unlinked special device node? Or just a race?
1331                          * Pretend we done everything. */
1332                         if (!S_ISREG(inode->i_mode) &&
1333                             !S_ISDIR(inode->i_mode))
1334                                 rc = inode_setattr(inode, &op_data->op_attr);
1335                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1336                         CERROR("md_setattr fails: rc = %d\n", rc);
1337                 }
1338                 RETURN(rc);
1339         }
1340
1341         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1342                               sbi->ll_md_exp, &md);
1343         if (rc) {
1344                 ptlrpc_req_finished(request);
1345                 RETURN(rc);
1346         }
1347
1348         /* We call inode_setattr to adjust timestamps.
1349          * If there is at least some data in file, we cleared ATTR_SIZE
1350          * above to avoid invoking vmtruncate, otherwise it is important
1351          * to call vmtruncate in inode_setattr to update inode->i_size
1352          * (bug 6196) */
1353         rc = inode_setattr(inode, &op_data->op_attr);
1354
1355         /* Extract epoch data if obtained. */
1356         op_data->op_handle = md.body->handle;
1357         op_data->op_ioepoch = md.body->ioepoch;
1358
1359         ll_update_inode(inode, &md);
1360         ptlrpc_req_finished(request);
1361
1362         RETURN(rc);
1363 }
1364
1365 /* Close IO epoch and send Size-on-MDS attribute update. */
1366 static int ll_setattr_done_writing(struct inode *inode,
1367                                    struct md_op_data *op_data,
1368                                    struct md_open_data *mod)
1369 {
1370         struct ll_inode_info *lli = ll_i2info(inode);
1371         int rc = 0;
1372         ENTRY;
1373
1374         LASSERT(op_data != NULL);
1375         if (!S_ISREG(inode->i_mode))
1376                 RETURN(0);
1377
1378         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1379                op_data->op_ioepoch, PFID(&lli->lli_fid));
1380
1381         op_data->op_flags = MF_EPOCH_CLOSE | MF_SOM_CHANGE;
1382         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1383         if (rc == -EAGAIN) {
1384                 /* MDS has instructed us to obtain Size-on-MDS attribute
1385                  * from OSTs and send setattr to back to MDS. */
1386                 rc = ll_sizeonmds_update(inode, mod, &op_data->op_handle,
1387                                          op_data->op_ioepoch);
1388         } else if (rc) {
1389                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1390                        inode->i_ino, rc);
1391         }
1392         RETURN(rc);
1393 }
1394
1395 static int ll_setattr_do_truncate(struct inode *inode, loff_t new_size)
1396 {
1397         struct ll_sb_info *sbi = ll_i2sbi(inode);
1398         struct ll_inode_info *lli = ll_i2info(inode);
1399         struct lov_stripe_md *lsm = lli->lli_smd;
1400         int rc;
1401         ldlm_policy_data_t policy = { .l_extent = {new_size,
1402                                                    OBD_OBJECT_EOF } };
1403         struct lustre_handle lockh = { 0 };
1404         int local_lock = 0; /* 0 - no local lock;
1405                              * 1 - lock taken by lock_extent;
1406                              * 2 - by obd_match*/
1407         int ast_flags;
1408         int err;
1409         ENTRY;
1410
1411         UNLOCK_INODE_MUTEX(inode);
1412         UP_WRITE_I_ALLOC_SEM(inode);
1413
1414         if (sbi->ll_lockless_truncate_enable &&
1415             (sbi->ll_lco.lco_flags & OBD_CONNECT_TRUNCLOCK)) {
1416                 ast_flags = LDLM_FL_BLOCK_GRANTED;
1417                 rc = obd_match(sbi->ll_dt_exp, lsm, LDLM_EXTENT,
1418                                &policy, LCK_PW, &ast_flags, inode, &lockh);
1419                 if (rc > 0) {
1420                         local_lock = 2;
1421                         rc = 0;
1422                 } else if (rc == 0) {
1423                         rc = ll_file_punch(inode, new_size, 1);
1424                 }
1425         } else {
1426                 /* XXX when we fix the AST intents to pass the discard-range
1427                  * XXX extent, make ast_flags always LDLM_AST_DISCARD_DATA
1428                  * XXX here. */
1429                 ast_flags = (new_size == 0) ? LDLM_AST_DISCARD_DATA : 0;
1430                 rc = ll_extent_lock(NULL, inode, lsm, LCK_PW, &policy,
1431                                     &lockh, ast_flags);
1432                 if (likely(rc == 0))
1433                         local_lock = 1;
1434         }
1435
1436         LOCK_INODE_MUTEX(inode);
1437         DOWN_WRITE_I_ALLOC_SEM(inode);
1438
1439         if (likely(rc == 0)) {
1440                 /* Only ll_inode_size_lock is taken at this level.
1441                  * lov_stripe_lock() is grabbed by ll_truncate() only over
1442                  * call to obd_adjust_kms().  If vmtruncate returns 0, then
1443                  * ll_truncate dropped ll_inode_size_lock() */
1444                 ll_inode_size_lock(inode, 0);
1445                 if (!local_lock) {
1446                         spin_lock(&lli->lli_lock);
1447                         lli->lli_flags |= LLIF_SRVLOCK;
1448                         spin_unlock(&lli->lli_lock);
1449                 }
1450                 rc = vmtruncate(inode, new_size);
1451                 if (!local_lock) {
1452                         spin_lock(&lli->lli_lock);
1453                         lli->lli_flags &= ~LLIF_SRVLOCK;
1454                         spin_unlock(&lli->lli_lock);
1455                 }
1456                 if (rc != 0) {
1457                         LASSERT(atomic_read(&lli->lli_size_sem.count) <= 0);
1458                         ll_inode_size_unlock(inode, 0);
1459                 }
1460         }
1461
1462         if (local_lock) {
1463                 if (local_lock == 2)
1464                         err = obd_cancel(sbi->ll_dt_exp, lsm, LCK_PW, &lockh);
1465                 else
1466                         err = ll_extent_unlock(NULL, inode, lsm, LCK_PW, &lockh);
1467                 if (unlikely(err != 0)){
1468                         CERROR("extent unlock failed: err=%d,"
1469                                " unlock method =%d\n", err, local_lock);
1470                         if (rc == 0)
1471                                 rc = err;
1472                 }
1473         }
1474         RETURN(rc);
1475 }
1476
1477 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1478  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1479  * keep these values until such a time that objects are allocated for it.
1480  * We do the MDS operations first, as it is checking permissions for us.
1481  * We don't to the MDS RPC if there is nothing that we want to store there,
1482  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1483  * going to do an RPC anyways.
1484  *
1485  * If we are doing a truncate, we will send the mtime and ctime updates
1486  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1487  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1488  * at the same time.
1489  */
1490 int ll_setattr_raw(struct inode *inode, struct iattr *attr)
1491 {
1492         struct ll_inode_info *lli = ll_i2info(inode);
1493         struct lov_stripe_md *lsm = lli->lli_smd;
1494         struct ll_sb_info *sbi = ll_i2sbi(inode);
1495         struct md_op_data *op_data = NULL;
1496         struct md_open_data *mod = NULL;
1497         int ia_valid = attr->ia_valid;
1498         int rc = 0, rc1 = 0;
1499         ENTRY;
1500
1501         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu valid %x\n", inode->i_ino,
1502                attr->ia_valid);
1503         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_SETATTR, 1);
1504
1505         if (ia_valid & ATTR_SIZE) {
1506                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1507                         CDEBUG(D_INODE, "file too large %llu > "LPU64"\n",
1508                                attr->ia_size, ll_file_maxbytes(inode));
1509                         RETURN(-EFBIG);
1510                 }
1511
1512                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1513         }
1514
1515         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1516         if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET)) {
1517                 if (current->fsuid != inode->i_uid &&
1518                     !cfs_capable(CFS_CAP_FOWNER))
1519                         RETURN(-EPERM);
1520         }
1521
1522         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1523         if (attr->ia_valid & ATTR_CTIME) {
1524                 attr->ia_ctime = CURRENT_TIME;
1525                 attr->ia_valid |= ATTR_CTIME_SET;
1526         }
1527         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1528                 attr->ia_atime = CURRENT_TIME;
1529                 attr->ia_valid |= ATTR_ATIME_SET;
1530         }
1531         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1532                 attr->ia_mtime = CURRENT_TIME;
1533                 attr->ia_valid |= ATTR_MTIME_SET;
1534         }
1535         if ((attr->ia_valid & ATTR_CTIME) && !(attr->ia_valid & ATTR_MTIME)) {
1536                 /* To avoid stale mtime on mds, obtain it from ost and send
1537                    to mds. */
1538                 rc = ll_glimpse_size(inode, 0);
1539                 if (rc)
1540                         RETURN(rc);
1541
1542                 attr->ia_valid |= ATTR_MTIME_SET | ATTR_MTIME;
1543                 attr->ia_mtime = inode->i_mtime;
1544         }
1545
1546         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1547                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1548                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1549                        cfs_time_current_sec());
1550
1551         /* NB: ATTR_SIZE will only be set after this point if the size
1552          * resides on the MDS, ie, this file has no objects. */
1553         if (lsm)
1554                 attr->ia_valid &= ~ATTR_SIZE;
1555
1556         /* We always do an MDS RPC, even if we're only changing the size;
1557          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1558
1559         OBD_ALLOC_PTR(op_data);
1560         if (op_data == NULL)
1561                 RETURN(-ENOMEM);
1562
1563         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1564
1565         /* Open epoch for truncate. */
1566         if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1567             (ia_valid & ATTR_SIZE))
1568                 op_data->op_flags = MF_EPOCH_OPEN;
1569
1570         rc = ll_md_setattr(inode, op_data, &mod);
1571         if (rc)
1572                 GOTO(out, rc);
1573
1574         if (op_data->op_ioepoch)
1575                 CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for "
1576                        "truncate\n", op_data->op_ioepoch, PFID(&lli->lli_fid));
1577
1578         if (!lsm || !S_ISREG(inode->i_mode)) {
1579                 CDEBUG(D_INODE, "no lsm: not setting attrs on OST\n");
1580                 GOTO(out, rc = 0);
1581         }
1582
1583         /* We really need to get our PW lock before we change inode->i_size.
1584          * If we don't we can race with other i_size updaters on our node, like
1585          * ll_file_read.  We can also race with i_size propogation to other
1586          * nodes through dirtying and writeback of final cached pages.  This
1587          * last one is especially bad for racing o_append users on other
1588          * nodes. */
1589         if (ia_valid & ATTR_SIZE) {
1590                 rc = ll_setattr_do_truncate(inode, attr->ia_size);
1591         } else if (ia_valid & (ATTR_MTIME | ATTR_MTIME_SET)) {
1592                 obd_flag flags;
1593                 struct obd_info oinfo = { { { 0 } } };
1594                 struct obdo *oa;
1595
1596                 CDEBUG(D_INODE, "set mtime on OST inode %lu to %lu\n",
1597                        inode->i_ino, LTIME_S(attr->ia_mtime));
1598
1599                 OBDO_ALLOC(oa);
1600                 if (oa) {
1601                         oa->o_id = lsm->lsm_object_id;
1602                         oa->o_gr = lsm->lsm_object_gr;
1603                         oa->o_valid = OBD_MD_FLID | OBD_MD_FLGROUP;
1604
1605                         flags = OBD_MD_FLTYPE | OBD_MD_FLATIME |
1606                                 OBD_MD_FLMTIME | OBD_MD_FLCTIME |
1607                                 OBD_MD_FLFID | OBD_MD_FLGENER |
1608                                 OBD_MD_FLGROUP;
1609
1610                         obdo_from_inode(oa, inode, flags);
1611
1612                         oinfo.oi_oa = oa;
1613                         oinfo.oi_md = lsm;
1614                         oinfo.oi_capa = ll_mdscapa_get(inode);
1615
1616                         /* XXX: this looks unnecessary now. */
1617                         rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1618                         capa_put(oinfo.oi_capa);
1619                         if (rc)
1620                                 CERROR("obd_setattr_async fails: rc=%d\n", rc);
1621                         OBDO_FREE(oa);
1622                 } else {
1623                         rc = -ENOMEM;
1624                 }
1625         }
1626         EXIT;
1627 out:
1628         if (op_data) {
1629                 if (op_data->op_ioepoch)
1630                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1631                 ll_finish_md_op_data(op_data);
1632         }
1633         return rc ? rc : rc1;
1634 }
1635
1636 int ll_setattr(struct dentry *de, struct iattr *attr)
1637 {
1638         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1639             (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1640                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1641
1642         if ((de->d_inode->i_mode & S_ISUID) &&
1643             !(attr->ia_mode & S_ISUID) &&
1644             !(attr->ia_valid & ATTR_KILL_SUID))
1645                 attr->ia_valid |= ATTR_KILL_SUID;
1646
1647         if (((de->d_inode->i_mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1648             !(attr->ia_mode & S_ISGID) &&
1649             !(attr->ia_valid & ATTR_KILL_SGID))
1650                 attr->ia_valid |= ATTR_KILL_SGID;
1651
1652         return ll_setattr_raw(de->d_inode, attr);
1653 }
1654
1655 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1656                        __u64 max_age, __u32 flags)
1657 {
1658         struct ll_sb_info *sbi = ll_s2sbi(sb);
1659         struct obd_statfs obd_osfs;
1660         int rc;
1661         ENTRY;
1662
1663         rc = obd_statfs(class_exp2obd(sbi->ll_md_exp), osfs, max_age, flags);
1664         if (rc) {
1665                 CERROR("md_statfs fails: rc = %d\n", rc);
1666                 RETURN(rc);
1667         }
1668
1669         osfs->os_type = sb->s_magic;
1670
1671         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1672                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1673
1674         rc = obd_statfs_rqset(class_exp2obd(sbi->ll_dt_exp),
1675                               &obd_osfs, max_age, flags);
1676         if (rc) {
1677                 CERROR("obd_statfs fails: rc = %d\n", rc);
1678                 RETURN(rc);
1679         }
1680
1681         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1682                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1683                obd_osfs.os_files);
1684
1685         osfs->os_bsize = obd_osfs.os_bsize;
1686         osfs->os_blocks = obd_osfs.os_blocks;
1687         osfs->os_bfree = obd_osfs.os_bfree;
1688         osfs->os_bavail = obd_osfs.os_bavail;
1689
1690         /* If we don't have as many objects free on the OST as inodes
1691          * on the MDS, we reduce the total number of inodes to
1692          * compensate, so that the "inodes in use" number is correct.
1693          */
1694         if (obd_osfs.os_ffree < osfs->os_ffree) {
1695                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1696                         obd_osfs.os_ffree;
1697                 osfs->os_ffree = obd_osfs.os_ffree;
1698         }
1699
1700         RETURN(rc);
1701 }
1702 #ifndef HAVE_STATFS_DENTRY_PARAM
1703 int ll_statfs(struct super_block *sb, struct kstatfs *sfs)
1704 {
1705 #else
1706 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1707 {
1708         struct super_block *sb = de->d_sb;
1709 #endif
1710         struct obd_statfs osfs;
1711         int rc;
1712
1713         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1714         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1715
1716         /* For now we will always get up-to-date statfs values, but in the
1717          * future we may allow some amount of caching on the client (e.g.
1718          * from QOS or lprocfs updates). */
1719         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - 1, 0);
1720         if (rc)
1721                 return rc;
1722
1723         statfs_unpack(sfs, &osfs);
1724
1725         /* We need to downshift for all 32-bit kernels, because we can't
1726          * tell if the kernel is being called via sys_statfs64() or not.
1727          * Stop before overflowing f_bsize - in which case it is better
1728          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1729         if (sizeof(long) < 8) {
1730                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1731                         sfs->f_bsize <<= 1;
1732
1733                         osfs.os_blocks >>= 1;
1734                         osfs.os_bfree >>= 1;
1735                         osfs.os_bavail >>= 1;
1736                 }
1737         }
1738
1739         sfs->f_blocks = osfs.os_blocks;
1740         sfs->f_bfree = osfs.os_bfree;
1741         sfs->f_bavail = osfs.os_bavail;
1742
1743         return 0;
1744 }
1745
1746 void ll_inode_size_lock(struct inode *inode, int lock_lsm)
1747 {
1748         struct ll_inode_info *lli;
1749         struct lov_stripe_md *lsm;
1750
1751         lli = ll_i2info(inode);
1752         LASSERT(lli->lli_size_sem_owner != current);
1753         down(&lli->lli_size_sem);
1754         LASSERT(lli->lli_size_sem_owner == NULL);
1755         lli->lli_size_sem_owner = current;
1756         lsm = lli->lli_smd;
1757         LASSERTF(lsm != NULL || lock_lsm == 0, "lsm %p, lock_lsm %d\n",
1758                  lsm, lock_lsm);
1759         if (lock_lsm)
1760                 lov_stripe_lock(lsm);
1761 }
1762
1763 void ll_inode_size_unlock(struct inode *inode, int unlock_lsm)
1764 {
1765         struct ll_inode_info *lli;
1766         struct lov_stripe_md *lsm;
1767
1768         lli = ll_i2info(inode);
1769         lsm = lli->lli_smd;
1770         LASSERTF(lsm != NULL || unlock_lsm == 0, "lsm %p, lock_lsm %d\n",
1771                  lsm, unlock_lsm);
1772         if (unlock_lsm)
1773                 lov_stripe_unlock(lsm);
1774         LASSERT(lli->lli_size_sem_owner == current);
1775         lli->lli_size_sem_owner = NULL;
1776         up(&lli->lli_size_sem);
1777 }
1778
1779 static void ll_replace_lsm(struct inode *inode, struct lov_stripe_md *lsm)
1780 {
1781         struct ll_inode_info *lli = ll_i2info(inode);
1782
1783         dump_lsm(D_INODE, lsm);
1784         dump_lsm(D_INODE, lli->lli_smd);
1785         LASSERTF(lsm->lsm_magic == LOV_MAGIC_JOIN,
1786                  "lsm must be joined lsm %p\n", lsm);
1787         obd_free_memmd(ll_i2dtexp(inode), &lli->lli_smd);
1788         CDEBUG(D_INODE, "replace lsm %p to lli_smd %p for inode %lu%u(%p)\n",
1789                lsm, lli->lli_smd, inode->i_ino, inode->i_generation, inode);
1790         lli->lli_smd = lsm;
1791         lli->lli_maxbytes = lsm->lsm_maxbytes;
1792         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1793                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1794 }
1795
1796 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1797 {
1798         struct ll_inode_info *lli = ll_i2info(inode);
1799         struct mdt_body *body = md->body;
1800         struct lov_stripe_md *lsm = md->lsm;
1801         struct ll_sb_info *sbi = ll_i2sbi(inode);
1802
1803         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1804         if (lsm != NULL) {
1805                 if (lli->lli_smd == NULL) {
1806                         if (lsm->lsm_magic != LOV_MAGIC_V1 &&
1807                             lsm->lsm_magic != LOV_MAGIC_V3 &&
1808                             lsm->lsm_magic != LOV_MAGIC_JOIN) {
1809                                 dump_lsm(D_ERROR, lsm);
1810                                 LBUG();
1811                         }
1812                         CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1813                                lsm, inode->i_ino, inode->i_generation, inode);
1814                         /* ll_inode_size_lock() requires it is only called
1815                          * with lli_smd != NULL or lock_lsm == 0 or we can
1816                          * race between lock/unlock.  bug 9547 */
1817                         lli->lli_smd = lsm;
1818                         lli->lli_maxbytes = lsm->lsm_maxbytes;
1819                         if (lli->lli_maxbytes > PAGE_CACHE_MAXBYTES)
1820                                 lli->lli_maxbytes = PAGE_CACHE_MAXBYTES;
1821                 } else {
1822                         if (lli->lli_smd->lsm_magic == lsm->lsm_magic &&
1823                              lli->lli_smd->lsm_stripe_count ==
1824                                         lsm->lsm_stripe_count) {
1825                                 if (lov_stripe_md_cmp(lli->lli_smd, lsm)) {
1826                                         CERROR("lsm mismatch for inode %ld\n",
1827                                                 inode->i_ino);
1828                                         CERROR("lli_smd:\n");
1829                                         dump_lsm(D_ERROR, lli->lli_smd);
1830                                         CERROR("lsm:\n");
1831                                         dump_lsm(D_ERROR, lsm);
1832                                         LBUG();
1833                                 }
1834                         } else
1835                                 ll_replace_lsm(inode, lsm);
1836                 }
1837                 if (lli->lli_smd != lsm)
1838                         obd_free_memmd(ll_i2dtexp(inode), &lsm);
1839         }
1840
1841         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1842                 if (body->valid & OBD_MD_FLRMTPERM)
1843                         ll_update_remote_perm(inode, md->remote_perm);
1844         }
1845 #ifdef CONFIG_FS_POSIX_ACL
1846         else if (body->valid & OBD_MD_FLACL) {
1847                 spin_lock(&lli->lli_lock);
1848                 if (lli->lli_posix_acl)
1849                         posix_acl_release(lli->lli_posix_acl);
1850                 lli->lli_posix_acl = md->posix_acl;
1851                 spin_unlock(&lli->lli_lock);
1852         }
1853 #endif
1854         inode->i_ino = ll_fid_build_ino(sbi, &body->fid1);
1855
1856         if (body->valid & OBD_MD_FLATIME &&
1857             body->atime > LTIME_S(inode->i_atime))
1858                 LTIME_S(inode->i_atime) = body->atime;
1859
1860         /* mtime is always updated with ctime, but can be set in past.
1861            As write and utime(2) may happen within 1 second, and utime's
1862            mtime has a priority over write's one, so take mtime from mds
1863            for the same ctimes. */
1864         if (body->valid & OBD_MD_FLCTIME &&
1865             body->ctime >= LTIME_S(inode->i_ctime)) {
1866                 LTIME_S(inode->i_ctime) = body->ctime;
1867                 if (body->valid & OBD_MD_FLMTIME) {
1868                         CDEBUG(D_INODE, "setting ino %lu mtime "
1869                                "from %lu to "LPU64"\n", inode->i_ino,
1870                                LTIME_S(inode->i_mtime), body->mtime);
1871                         LTIME_S(inode->i_mtime) = body->mtime;
1872                 }
1873         }
1874         if (body->valid & OBD_MD_FLMODE)
1875                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1876         if (body->valid & OBD_MD_FLTYPE)
1877                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1878         if (S_ISREG(inode->i_mode)) {
1879                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1880         } else {
1881                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1882         }
1883 #ifdef HAVE_INODE_BLKSIZE
1884         inode->i_blksize = 1<<inode->i_blkbits;
1885 #endif
1886         if (body->valid & OBD_MD_FLUID)
1887                 inode->i_uid = body->uid;
1888         if (body->valid & OBD_MD_FLGID)
1889                 inode->i_gid = body->gid;
1890         if (body->valid & OBD_MD_FLFLAGS)
1891                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1892         if (body->valid & OBD_MD_FLNLINK)
1893                 inode->i_nlink = body->nlink;
1894         if (body->valid & OBD_MD_FLRDEV)
1895                 inode->i_rdev = old_decode_dev(body->rdev);
1896
1897         if (body->valid & OBD_MD_FLID) {
1898                 /* FID shouldn't be changed! */
1899                 if (fid_is_sane(&lli->lli_fid)) {
1900                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1901                                  "Trying to change FID "DFID
1902                                  " to the "DFID", inode %lu/%u(%p)\n",
1903                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1904                                  inode->i_ino, inode->i_generation, inode);
1905                 } else
1906                         lli->lli_fid = body->fid1;
1907         }
1908
1909         LASSERT(fid_seq(&lli->lli_fid) != 0);
1910
1911         if (body->valid & OBD_MD_FLSIZE) {
1912                 if ((ll_i2mdexp(inode)->exp_connect_flags & OBD_CONNECT_SOM) &&
1913                     S_ISREG(inode->i_mode) && lli->lli_smd) {
1914                         struct lustre_handle lockh;
1915                         ldlm_mode_t mode;
1916
1917                         /* As it is possible a blocking ast has been processed
1918                          * by this time, we need to check there is an UPDATE
1919                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1920                          * it. */
1921                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1922                                                &lockh);
1923                         if (mode) {
1924                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1925                                                       LLIF_EPOCH_PENDING |
1926                                                       LLIF_SOM_DIRTY)) {
1927                                         CERROR("ino %lu flags %lu still has "
1928                                                "size authority! do not trust "
1929                                                "the size got from MDS\n",
1930                                                inode->i_ino, lli->lli_flags);
1931                                 } else {
1932                                         /* Use old size assignment to avoid
1933                                          * deadlock bz14138 & bz14326 */
1934                                         inode->i_size = body->size;
1935                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1936                                 }
1937                                 ldlm_lock_decref(&lockh, mode);
1938                         }
1939                 } else {
1940                         /* Use old size assignment to avoid
1941                          * deadlock bz14138 & bz14326 */
1942                         inode->i_size = body->size;
1943                 }
1944
1945                 if (body->valid & OBD_MD_FLBLOCKS)
1946                         inode->i_blocks = body->blocks;
1947         }
1948
1949         if (body->valid & OBD_MD_FLMDSCAPA) {
1950                 LASSERT(md->mds_capa);
1951                 ll_add_capa(inode, md->mds_capa);
1952         }
1953         if (body->valid & OBD_MD_FLOSSCAPA) {
1954                 LASSERT(md->oss_capa);
1955                 ll_add_capa(inode, md->oss_capa);
1956         }
1957 }
1958
1959 static struct backing_dev_info ll_backing_dev_info = {
1960         .ra_pages       = 0,    /* No readahead */
1961 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,12))
1962         .capabilities   = 0,    /* Does contribute to dirty memory */
1963 #else
1964         .memory_backed  = 0,    /* Does contribute to dirty memory */
1965 #endif
1966 };
1967
1968 void ll_read_inode2(struct inode *inode, void *opaque)
1969 {
1970         struct lustre_md *md = opaque;
1971         struct ll_inode_info *lli = ll_i2info(inode);
1972         ENTRY;
1973
1974         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n",
1975                inode->i_ino, inode->i_generation, inode);
1976
1977         ll_lli_init(lli);
1978
1979         LASSERT(!lli->lli_smd);
1980
1981         /* Core attributes from the MDS first.  This is a new inode, and
1982          * the VFS doesn't zero times in the core inode so we have to do
1983          * it ourselves.  They will be overwritten by either MDS or OST
1984          * attributes - we just need to make sure they aren't newer. */
1985         LTIME_S(inode->i_mtime) = 0;
1986         LTIME_S(inode->i_atime) = 0;
1987         LTIME_S(inode->i_ctime) = 0;
1988         inode->i_rdev = 0;
1989         ll_update_inode(inode, md);
1990
1991         /* OIDEBUG(inode); */
1992
1993         if (S_ISREG(inode->i_mode)) {
1994                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1995                 inode->i_op = &ll_file_inode_operations;
1996                 inode->i_fop = sbi->ll_fop;
1997                 inode->i_mapping->a_ops = &ll_aops;
1998                 EXIT;
1999         } else if (S_ISDIR(inode->i_mode)) {
2000                 inode->i_op = &ll_dir_inode_operations;
2001                 inode->i_fop = &ll_dir_operations;
2002                 inode->i_mapping->a_ops = &ll_dir_aops;
2003                 EXIT;
2004         } else if (S_ISLNK(inode->i_mode)) {
2005                 inode->i_op = &ll_fast_symlink_inode_operations;
2006                 EXIT;
2007         } else {
2008                 inode->i_op = &ll_special_inode_operations;
2009
2010                 init_special_inode(inode, inode->i_mode,
2011                                    kdev_t_to_nr(inode->i_rdev));
2012
2013                 /* initializing backing dev info. */
2014                 inode->i_mapping->backing_dev_info = &ll_backing_dev_info;
2015
2016                 EXIT;
2017         }
2018 }
2019
2020 void ll_delete_inode(struct inode *inode)
2021 {
2022         struct ll_sb_info *sbi = ll_i2sbi(inode);
2023         int rc;
2024         ENTRY;
2025
2026         rc = obd_fid_delete(sbi->ll_md_exp, ll_inode2fid(inode));
2027         if (rc) {
2028                 CERROR("fid_delete() failed, rc %d\n", rc);
2029         }
2030         truncate_inode_pages(&inode->i_data, 0);
2031         clear_inode(inode);
2032
2033         EXIT;
2034 }
2035
2036 int ll_iocontrol(struct inode *inode, struct file *file,
2037                  unsigned int cmd, unsigned long arg)
2038 {
2039         struct ll_sb_info *sbi = ll_i2sbi(inode);
2040         struct ptlrpc_request *req = NULL;
2041         int rc, flags = 0;
2042         ENTRY;
2043
2044         switch(cmd) {
2045         case EXT3_IOC_GETFLAGS: {
2046                 struct mdt_body *body;
2047                 struct obd_capa *oc;
2048
2049                 oc = ll_mdscapa_get(inode);
2050                 rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode), oc,
2051                                 OBD_MD_FLFLAGS, 0, &req);
2052                 capa_put(oc);
2053                 if (rc) {
2054                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
2055                         RETURN(-abs(rc));
2056                 }
2057
2058                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2059
2060                 flags = body->flags;
2061
2062                 ptlrpc_req_finished(req);
2063
2064                 RETURN(put_user(flags, (int *)arg));
2065         }
2066         case EXT3_IOC_SETFLAGS: {
2067                 struct lov_stripe_md *lsm = ll_i2info(inode)->lli_smd;
2068                 struct obd_info oinfo = { { { 0 } } };
2069                 struct md_op_data *op_data;
2070
2071                 if (get_user(flags, (int *)arg))
2072                         RETURN(-EFAULT);
2073
2074                 oinfo.oi_md = lsm;
2075                 OBDO_ALLOC(oinfo.oi_oa);
2076                 if (!oinfo.oi_oa)
2077                         RETURN(-ENOMEM);
2078
2079                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2080                                              LUSTRE_OPC_ANY, NULL);
2081                 if (IS_ERR(op_data))
2082                         RETURN(PTR_ERR(op_data));
2083
2084                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2085                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2086                 rc = md_setattr(sbi->ll_md_exp, op_data,
2087                                 NULL, 0, NULL, 0, &req, NULL);
2088                 ll_finish_md_op_data(op_data);
2089                 ptlrpc_req_finished(req);
2090                 if (rc || lsm == NULL) {
2091                         OBDO_FREE(oinfo.oi_oa);
2092                         RETURN(rc);
2093                 }
2094
2095                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
2096                 oinfo.oi_oa->o_gr = lsm->lsm_object_gr;
2097                 oinfo.oi_oa->o_flags = flags;
2098                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2099                                        OBD_MD_FLGROUP;
2100                 oinfo.oi_capa = ll_mdscapa_get(inode);
2101
2102                 obdo_from_inode(oinfo.oi_oa, inode,
2103                                 OBD_MD_FLFID | OBD_MD_FLGENER);
2104                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2105                 capa_put(oinfo.oi_capa);
2106                 OBDO_FREE(oinfo.oi_oa);
2107                 if (rc) {
2108                         if (rc != -EPERM && rc != -EACCES)
2109                                 CERROR("md_setattr_async fails: rc = %d\n", rc);
2110                         RETURN(rc);
2111                 }
2112
2113                 inode->i_flags = ll_ext_to_inode_flags(flags |
2114                                                        MDS_BFLAG_EXT_FLAGS);
2115                 RETURN(0);
2116         }
2117         default:
2118                 RETURN(-ENOSYS);
2119         }
2120
2121         RETURN(0);
2122 }
2123
2124 int ll_flush_ctx(struct inode *inode)
2125 {
2126         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2127
2128         CDEBUG(D_SEC, "flush context for user %d\n", current->uid);
2129
2130         obd_set_info_async(sbi->ll_md_exp,
2131                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2132                            0, NULL, NULL);
2133         obd_set_info_async(sbi->ll_dt_exp,
2134                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2135                            0, NULL, NULL);
2136         return 0;
2137 }
2138
2139 /* umount -f client means force down, don't save state */
2140 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2141 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2142 {
2143         struct super_block *sb = vfsmnt->mnt_sb;
2144 #else
2145 void ll_umount_begin(struct super_block *sb)
2146 {
2147 #endif
2148         struct lustre_sb_info *lsi = s2lsi(sb);
2149         struct ll_sb_info *sbi = ll_s2sbi(sb);
2150         struct obd_device *obd;
2151         struct obd_ioctl_data ioc_data = { 0 };
2152         ENTRY;
2153
2154 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2155         if (!(flags & MNT_FORCE)) {
2156                 EXIT;
2157                 return;
2158         }
2159 #endif
2160
2161         /* Tell the MGC we got umount -f */
2162         lsi->lsi_flags |= LSI_UMOUNT_FORCE;
2163
2164         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2165                sb->s_count, atomic_read(&sb->s_active));
2166
2167         obd = class_exp2obd(sbi->ll_md_exp);
2168         if (obd == NULL) {
2169                 CERROR("Invalid MDC connection handle "LPX64"\n",
2170                        sbi->ll_md_exp->exp_handle.h_cookie);
2171                 EXIT;
2172                 return;
2173         }
2174         obd->obd_force = 1;
2175         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp, sizeof ioc_data,
2176                       &ioc_data, NULL);
2177
2178         obd = class_exp2obd(sbi->ll_dt_exp);
2179         if (obd == NULL) {
2180                 CERROR("Invalid LOV connection handle "LPX64"\n",
2181                        sbi->ll_dt_exp->exp_handle.h_cookie);
2182                 EXIT;
2183                 return;
2184         }
2185
2186         obd->obd_force = 1;
2187         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp, sizeof ioc_data,
2188                       &ioc_data, NULL);
2189
2190         /* Really, we'd like to wait until there are no requests outstanding,
2191          * and then continue.  For now, we just invalidate the requests,
2192          * schedule, and hope.
2193          */
2194         schedule();
2195
2196         EXIT;
2197 }
2198
2199 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2200 {
2201         struct ll_sb_info *sbi = ll_s2sbi(sb);
2202         int err;
2203         __u32 read_only;
2204
2205         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2206                 read_only = *flags & MS_RDONLY;
2207                 err = obd_set_info_async(sbi->ll_md_exp,
2208                                          sizeof(KEY_READ_ONLY),
2209                                          KEY_READ_ONLY, sizeof(read_only),
2210                                          &read_only, NULL);
2211                 if (err) {
2212                         CERROR("Failed to change the read-only flag during "
2213                                "remount: %d\n", err);
2214                         return err;
2215                 }
2216
2217                 if (read_only)
2218                         sb->s_flags |= MS_RDONLY;
2219                 else
2220                         sb->s_flags &= ~MS_RDONLY;
2221         }
2222         return 0;
2223 }
2224
2225 int ll_prep_inode(struct inode **inode,
2226                   struct ptlrpc_request *req,
2227                   struct super_block *sb)
2228 {
2229         struct ll_sb_info *sbi = NULL;
2230         struct lustre_md md;
2231         int rc = 0;
2232         ENTRY;
2233
2234         LASSERT(*inode || sb);
2235         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2236         prune_deathrow(sbi, 1);
2237         memset(&md, 0, sizeof(struct lustre_md));
2238
2239         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2240                               sbi->ll_md_exp, &md);
2241         if (rc)
2242                 RETURN(rc);
2243
2244         if (*inode) {
2245                 ll_update_inode(*inode, &md);
2246         } else {
2247                 LASSERT(sb != NULL);
2248
2249                 /*
2250                  * At this point server returns to client's same fid as client
2251                  * generated for creating. So using ->fid1 is okay here.
2252                  */
2253                 LASSERT(fid_is_sane(&md.body->fid1));
2254
2255                 *inode = ll_iget(sb, ll_fid_build_ino(sbi, &md.body->fid1), &md);
2256                 if (*inode == NULL || is_bad_inode(*inode)) {
2257                         if (md.lsm)
2258                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2259 #ifdef CONFIG_FS_POSIX_ACL
2260                         if (md.posix_acl) {
2261                                 posix_acl_release(md.posix_acl);
2262                                 md.posix_acl = NULL;
2263                         }
2264 #endif
2265                         rc = -ENOMEM;
2266                         CERROR("new_inode -fatal: rc %d\n", rc);
2267                         GOTO(out, rc);
2268                 }
2269         }
2270
2271         rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp,
2272                          ll_i2info(*inode)->lli_smd);
2273 out:
2274         md_free_lustre_md(sbi->ll_md_exp, &md);
2275         RETURN(rc);
2276 }
2277
2278 char *llap_origins[] = {
2279         [LLAP_ORIGIN_UNKNOWN] = "--",
2280         [LLAP_ORIGIN_READPAGE] = "rp",
2281         [LLAP_ORIGIN_READAHEAD] = "ra",
2282         [LLAP_ORIGIN_COMMIT_WRITE] = "cw",
2283         [LLAP_ORIGIN_WRITEPAGE] = "wp",
2284         [LLAP_ORIGIN_LOCKLESS_IO] = "ls"
2285 };
2286
2287 struct ll_async_page *llite_pglist_next_llap(struct list_head *head,
2288                                              struct list_head *list)
2289 {
2290         struct ll_async_page *llap;
2291         struct list_head *pos;
2292
2293         list_for_each(pos, list) {
2294                 if (pos == head)
2295                         return NULL;
2296                 llap = list_entry(pos, struct ll_async_page, llap_pglist_item);
2297                 if (llap->llap_page == NULL)
2298                         continue;
2299                 return llap;
2300         }
2301         LBUG();
2302         return NULL;
2303 }
2304
2305 int ll_obd_statfs(struct inode *inode, void *arg)
2306 {
2307         struct ll_sb_info *sbi = NULL;
2308         struct obd_export *exp;
2309         char *buf = NULL;
2310         struct obd_ioctl_data *data = NULL;
2311         __u32 type;
2312         int len = 0, rc;
2313
2314         if (!inode || !(sbi = ll_i2sbi(inode)))
2315                 GOTO(out_statfs, rc = -EINVAL);
2316
2317         rc = obd_ioctl_getdata(&buf, &len, arg);
2318         if (rc)
2319                 GOTO(out_statfs, rc);
2320
2321         data = (void*)buf;
2322         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2323             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2324                 GOTO(out_statfs, rc = -EINVAL);
2325
2326         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2327         if (type == LL_STATFS_MDC)
2328                 exp = sbi->ll_md_exp;
2329         else if (type == LL_STATFS_LOV)
2330                 exp = sbi->ll_dt_exp;
2331         else
2332                 GOTO(out_statfs, rc = -ENODEV);
2333
2334         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2335         if (rc)
2336                 GOTO(out_statfs, rc);
2337 out_statfs:
2338         if (buf)
2339                 obd_ioctl_freedata(buf, len);
2340         return rc;
2341 }
2342
2343 int ll_process_config(struct lustre_cfg *lcfg)
2344 {
2345         char *ptr;
2346         void *sb;
2347         struct lprocfs_static_vars lvars;
2348         unsigned long x;
2349         int rc = 0;
2350
2351         lprocfs_llite_init_vars(&lvars);
2352
2353         /* The instance name contains the sb: lustre-client-aacfe000 */
2354         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2355         if (!ptr || !*(++ptr))
2356                 return -EINVAL;
2357         if (sscanf(ptr, "%lx", &x) != 1)
2358                 return -EINVAL;
2359         sb = (void *)x;
2360         /* This better be a real Lustre superblock! */
2361         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2362
2363         /* Note we have not called client_common_fill_super yet, so
2364            proc fns must be able to handle that! */
2365         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2366                                       lcfg, sb);
2367         return(rc);
2368 }
2369
2370 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2371 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2372                                        struct inode *i1, struct inode *i2,
2373                                        const char *name, int namelen,
2374                                        int mode, __u32 opc, void *data)
2375 {
2376         LASSERT(i1 != NULL);
2377
2378         if (namelen > ll_i2sbi(i1)->ll_namelen)
2379                 return ERR_PTR(-ENAMETOOLONG);
2380
2381         if (op_data == NULL)
2382                 OBD_ALLOC_PTR(op_data);
2383
2384         if (op_data == NULL)
2385                 return ERR_PTR(-ENOMEM);
2386
2387         ll_i2gids(op_data->op_suppgids, i1, i2);
2388         op_data->op_fid1 = *ll_inode2fid(i1);
2389         op_data->op_capa1 = ll_mdscapa_get(i1);
2390
2391         if (i2) {
2392                 op_data->op_fid2 = *ll_inode2fid(i2);
2393                 op_data->op_capa2 = ll_mdscapa_get(i2);
2394         } else {
2395                 fid_zero(&op_data->op_fid2);
2396                 op_data->op_capa2 = NULL;
2397         }
2398
2399         op_data->op_name = name;
2400         op_data->op_namelen = namelen;
2401         op_data->op_mode = mode;
2402         op_data->op_mod_time = cfs_time_current_sec();
2403         op_data->op_fsuid = current->fsuid;
2404         op_data->op_fsgid = current->fsgid;
2405         op_data->op_cap = cfs_curproc_cap_pack();
2406         op_data->op_bias = MDS_CHECK_SPLIT;
2407         op_data->op_opc = opc;
2408         op_data->op_mds = 0;
2409         op_data->op_data = data;
2410
2411         return op_data;
2412 }
2413
2414 void ll_finish_md_op_data(struct md_op_data *op_data)
2415 {
2416         capa_put(op_data->op_capa1);
2417         capa_put(op_data->op_capa2);
2418         OBD_FREE_PTR(op_data);
2419 }
2420
2421 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2422 {
2423         struct ll_sb_info *sbi;
2424
2425         LASSERT((seq != NULL) && (vfs != NULL));
2426         sbi = ll_s2sbi(vfs->mnt_sb);
2427
2428         if (sbi->ll_flags & LL_SBI_NOLCK)
2429                 seq_puts(seq, ",nolock");
2430
2431         if (sbi->ll_flags & LL_SBI_FLOCK)
2432                 seq_puts(seq, ",flock");
2433
2434         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2435                 seq_puts(seq, ",localflock");
2436
2437         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2438                 seq_puts(seq, ",user_xattr");
2439
2440         if (sbi->ll_flags & LL_SBI_ACL)
2441                 seq_puts(seq, ",acl");
2442
2443         RETURN(0);
2444 }