drivers/staging/lustre/lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  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/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/mm.h>
  47
  48 #include "../include/lustre_lite.h"
  49 #include "../include/lustre_ha.h"
  50 #include "../include/lustre_dlm.h"
  51 #include "../include/lprocfs_status.h"
  52 #include "../include/lustre_disk.h"
  53 #include "../include/lustre_param.h"
  54 #include "../include/lustre_log.h"
  55 #include "../include/cl_object.h"
  56 #include "../include/obd_cksum.h"
  57 #include "llite_internal.h"
  58
  59 struct kmem_cache *ll_file_data_slab;
  60 struct proc_dir_entry *proc_lustre_fs_root;
  61
  62 static LIST_HEAD(ll_super_blocks);
  63 static DEFINE_SPINLOCK(ll_sb_lock);
  64
  65 #ifndef log2
  66 #define log2(n) ffz(~(n))
  67 #endif
  68
  69 static struct ll_sb_info *ll_init_sbi(void)
  70 {
  71         struct ll_sb_info *sbi = NULL;
  72         unsigned long pages;
  73         unsigned long lru_page_max;
  74         struct sysinfo si;
  75         class_uuid_t uuid;
  76         int i;
  77
  78         sbi = kzalloc(sizeof(*sbi), GFP_NOFS);
  79         if (!sbi)
  80                 return NULL;
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
  91                 lru_page_max = pages / 2;
  92         } else {
  93                 lru_page_max = (pages / 4) * 3;
  94         }
  95
  96         /* initialize lru data */
  97         atomic_set(&sbi->ll_cache.ccc_users, 0);
  98         sbi->ll_cache.ccc_lru_max = lru_page_max;
  99         atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 100         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 101         INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         spin_lock(&ll_sb_lock);
 116         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         spin_unlock(&ll_sb_lock);
 118
 119         sbi->ll_flags |= LL_SBI_VERBOSE;
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121
 122         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 123
 124         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_r_hist.oh_lock);
 127                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 128                                pp_w_hist.oh_lock);
 129         }
 130
 131         /* metadata statahead is enabled by default */
 132         sbi->ll_sa_max = LL_SA_RPC_DEF;
 133         atomic_set(&sbi->ll_sa_total, 0);
 134         atomic_set(&sbi->ll_sa_wrong, 0);
 135         atomic_set(&sbi->ll_agl_total, 0);
 136         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 137
 138         return sbi;
 139 }
 140
 141 static void ll_free_sbi(struct super_block *sb)
 142 {
 143         struct ll_sb_info *sbi = ll_s2sbi(sb);
 144
 145         if (sbi != NULL) {
 146                 spin_lock(&ll_sb_lock);
 147                 list_del(&sbi->ll_list);
 148                 spin_unlock(&ll_sb_lock);
 149                 OBD_FREE(sbi, sizeof(*sbi));
 150         }
 151 }
 152
 153 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 154                                     struct vfsmount *mnt)
 155 {
 156         struct inode *root = NULL;
 157         struct ll_sb_info *sbi = ll_s2sbi(sb);
 158         struct obd_device *obd;
 159         struct obd_capa *oc = NULL;
 160         struct obd_statfs *osfs = NULL;
 161         struct ptlrpc_request *request = NULL;
 162         struct obd_connect_data *data = NULL;
 163         struct obd_uuid *uuid;
 164         struct md_op_data *op_data;
 165         struct lustre_md lmd;
 166         u64 valid;
 167         int size, err, checksum;
 168
 169         obd = class_name2obd(md);
 170         if (!obd) {
 171                 CERROR("MD %s: not setup or attached\n", md);
 172                 return -EINVAL;
 173         }
 174
 175         data = kzalloc(sizeof(*data), GFP_NOFS);
 176         if (!data)
 177                 return -ENOMEM;
 178
 179         osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
 180         if (!osfs) {
 181                 OBD_FREE_PTR(data);
 182                 return -ENOMEM;
 183         }
 184
 185         if (proc_lustre_fs_root) {
 186                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 187                                                   dt, md);
 188                 if (err < 0)
 189                         CERROR("could not register mount in /proc/fs/lustre\n");
 190         }
 191
 192         /* indicate the features supported by this client */
 193         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 194                                   OBD_CONNECT_ATTRFID  |
 195                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 196                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 197                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 198                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 199                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 200                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 201                                   OBD_CONNECT_EINPROGRESS |
 202                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 203                                   OBD_CONNECT_LAYOUTLOCK |
 204                                   OBD_CONNECT_PINGLESS |
 205                                   OBD_CONNECT_MAX_EASIZE |
 206                                   OBD_CONNECT_FLOCK_DEAD |
 207                                   OBD_CONNECT_DISP_STRIPE;
 208
 209         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 210                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 211
 212         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 213                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 214 #ifdef CONFIG_FS_POSIX_ACL
 215         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 216 #endif
 217
 218         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 219                 /* flag mdc connection as lightweight, only used for test
 220                  * purpose, use with care */
 221                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 222
 223         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 224         data->ocd_version = LUSTRE_VERSION_CODE;
 225
 226         if (sb->s_flags & MS_RDONLY)
 227                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 228         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 229                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 230
 231 #ifdef HAVE_MS_FLOCK_LOCK
 232         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 233         sb->s_flags |= MS_FLOCK_LOCK;
 234 #endif
 235 #ifdef MS_HAS_NEW_AOPS
 236         sb->s_flags |= MS_HAS_NEW_AOPS;
 237 #endif
 238
 239         if (sbi->ll_flags & LL_SBI_FLOCK)
 240                 sbi->ll_fop = &ll_file_operations_flock;
 241         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 242                 sbi->ll_fop = &ll_file_operations;
 243         else
 244                 sbi->ll_fop = &ll_file_operations_noflock;
 245
 246         /* real client */
 247         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 248         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 249                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 250
 251         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 252
 253         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 254         if (err == -EBUSY) {
 255                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 256                                    "recovery, of which this client is not a "
 257                                    "part. Please wait for recovery to complete,"
 258                                    " abort, or time out.\n", md);
 259                 goto out;
 260         } else if (err) {
 261                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 262                 goto out;
 263         }
 264
 265         sbi->ll_md_exp->exp_connect_data = *data;
 266
 267         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 268                            LUSTRE_SEQ_METADATA);
 269         if (err) {
 270                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 271                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 272                 goto out_md;
 273         }
 274
 275         /* For mount, we only need fs info from MDT0, and also in DNE, it
 276          * can make sure the client can be mounted as long as MDT0 is
 277          * available */
 278         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 279                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 280                         OBD_STATFS_FOR_MDT0);
 281         if (err)
 282                 goto out_md_fid;
 283
 284         /* This needs to be after statfs to ensure connect has finished.
 285          * Note that "data" does NOT contain the valid connect reply.
 286          * If connecting to a 1.8 server there will be no LMV device, so
 287          * we can access the MDC export directly and exp_connect_flags will
 288          * be non-zero, but if accessing an upgraded 2.1 server it will
 289          * have the correct flags filled in.
 290          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 291         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 292         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 293             valid != CLIENT_CONNECT_MDT_REQD) {
 294                 char *buf;
 295
 296                 buf = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
 297                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 298                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 299                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 300                                    "feature(s) needed for correct operation "
 301                                    "of this client (%s). Please upgrade "
 302                                    "server or downgrade client.\n",
 303                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 304                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 305                 err = -EPROTO;
 306                 goto out_md_fid;
 307         }
 308
 309         size = sizeof(*data);
 310         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 311                            KEY_CONN_DATA,  &size, data, NULL);
 312         if (err) {
 313                 CERROR("%s: Get connect data failed: rc = %d\n",
 314                        sbi->ll_md_exp->exp_obd->obd_name, err);
 315                 goto out_md_fid;
 316         }
 317
 318         LASSERT(osfs->os_bsize);
 319         sb->s_blocksize = osfs->os_bsize;
 320         sb->s_blocksize_bits = log2(osfs->os_bsize);
 321         sb->s_magic = LL_SUPER_MAGIC;
 322         sb->s_maxbytes = MAX_LFS_FILESIZE;
 323         sbi->ll_namelen = osfs->os_namelen;
 324         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 325
 326         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 327             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 328                 LCONSOLE_INFO("Disabling user_xattr feature because "
 329                               "it is not supported on the server\n");
 330                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 331         }
 332
 333         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 334 #ifdef MS_POSIXACL
 335                 sb->s_flags |= MS_POSIXACL;
 336 #endif
 337                 sbi->ll_flags |= LL_SBI_ACL;
 338         } else {
 339                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 340 #ifdef MS_POSIXACL
 341                 sb->s_flags &= ~MS_POSIXACL;
 342 #endif
 343                 sbi->ll_flags &= ~LL_SBI_ACL;
 344         }
 345
 346         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 347                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 348                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 349                         LCONSOLE_INFO("client is set as remote by default.\n");
 350                 }
 351         } else {
 352                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 353                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 354                         LCONSOLE_INFO("client claims to be remote, but server "
 355                                       "rejected, forced to be local.\n");
 356                 }
 357         }
 358
 359         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 360                 LCONSOLE_INFO("client enabled MDS capability!\n");
 361                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 362         }
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 365                 LCONSOLE_INFO("client enabled OSS capability!\n");
 366                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 370                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 373                 sbi->ll_md_brw_size = data->ocd_brw_size;
 374         else
 375                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 378                 LCONSOLE_INFO("Layout lock feature supported.\n");
 379                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 380         }
 381
 382         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 383                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 384                         LCONSOLE_INFO(
 385                                 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
 386                                 dt);
 387                 } else {
 388                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 389                         sbi->ll_xattr_cache_enabled = 1;
 390                 }
 391         }
 392
 393         obd = class_name2obd(dt);
 394         if (!obd) {
 395                 CERROR("DT %s: not setup or attached\n", dt);
 396                 err = -ENODEV;
 397                 goto out_md_fid;
 398         }
 399
 400         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 401                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 402                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 403                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 404                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 405                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 406                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 407                                   OBD_CONNECT_MAXBYTES |
 408                                   OBD_CONNECT_EINPROGRESS |
 409                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 410                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 411
 412         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 413                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 414
 415         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 416                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 417                  * disabled by default, because it can still be enabled on the
 418                  * fly via /proc. As a consequence, we still need to come to an
 419                  * agreement on the supported algorithms at connect time */
 420                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 421
 422                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 423                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 424                 else
 425                         data->ocd_cksum_types = cksum_types_supported_client();
 426         }
 427
 428         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 429         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 430                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 431
 432         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 433                "ocd_grant: %d\n", data->ocd_connect_flags,
 434                data->ocd_version, data->ocd_grant);
 435
 436         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 437         obd->obd_upcall.onu_upcall = cl_ocd_update;
 438
 439         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 440
 441         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 442                           NULL);
 443         if (err == -EBUSY) {
 444                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 445                                    "recovery, of which this client is not a "
 446                                    "part.  Please wait for recovery to "
 447                                    "complete, abort, or time out.\n", dt);
 448                 goto out_md;
 449         } else if (err) {
 450                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 451                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 452                 goto out_md;
 453         }
 454
 455         sbi->ll_dt_exp->exp_connect_data = *data;
 456
 457         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 458                            LUSTRE_SEQ_METADATA);
 459         if (err) {
 460                 CERROR("%s: Can't init data layer FID infrastructure, "
 461                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 462                 goto out_dt;
 463         }
 464
 465         mutex_lock(&sbi->ll_lco.lco_lock);
 466         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 467         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 468         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 469         mutex_unlock(&sbi->ll_lco.lco_lock);
 470
 471         fid_zero(&sbi->ll_root_fid);
 472         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 473         if (err) {
 474                 CERROR("cannot mds_connect: rc = %d\n", err);
 475                 goto out_lock_cn_cb;
 476         }
 477         if (!fid_is_sane(&sbi->ll_root_fid)) {
 478                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 479                        sbi->ll_md_exp->exp_obd->obd_name,
 480                        PFID(&sbi->ll_root_fid));
 481                 err = -EINVAL;
 482                 goto out_lock_cn_cb;
 483         }
 484         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 485
 486         sb->s_op = &lustre_super_operations;
 487 #if THREAD_SIZE >= 8192 /*b=17630*/
 488         sb->s_export_op = &lustre_export_operations;
 489 #endif
 490
 491         /* make root inode
 492          * XXX: move this to after cbd setup? */
 493         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 494         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 495                 valid |= OBD_MD_FLRMTPERM;
 496         else if (sbi->ll_flags & LL_SBI_ACL)
 497                 valid |= OBD_MD_FLACL;
 498
 499         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
 500         if (!op_data) {
 501                 err = -ENOMEM;
 502                 goto out_lock_cn_cb;
 503         }
 504
 505         op_data->op_fid1 = sbi->ll_root_fid;
 506         op_data->op_mode = 0;
 507         op_data->op_capa1 = oc;
 508         op_data->op_valid = valid;
 509
 510         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 511         if (oc)
 512                 capa_put(oc);
 513         OBD_FREE_PTR(op_data);
 514         if (err) {
 515                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 516                        sbi->ll_md_exp->exp_obd->obd_name, err);
 517                 goto out_lock_cn_cb;
 518         }
 519
 520         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 521                                sbi->ll_md_exp, &lmd);
 522         if (err) {
 523                 CERROR("failed to understand root inode md: rc = %d\n", err);
 524                 ptlrpc_req_finished(request);
 525                 goto out_lock_cn_cb;
 526         }
 527
 528         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 529         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 530                                             sbi->ll_flags & LL_SBI_32BIT_API),
 531                        &lmd);
 532         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 533         ptlrpc_req_finished(request);
 534
 535         if (root == NULL || IS_ERR(root)) {
 536                 if (lmd.lsm)
 537                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 538 #ifdef CONFIG_FS_POSIX_ACL
 539                 if (lmd.posix_acl) {
 540                         posix_acl_release(lmd.posix_acl);
 541                         lmd.posix_acl = NULL;
 542                 }
 543 #endif
 544                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 545                 root = NULL;
 546                 CERROR("lustre_lite: bad iget4 for root\n");
 547                 goto out_root;
 548         }
 549
 550         err = ll_close_thread_start(&sbi->ll_lcq);
 551         if (err) {
 552                 CERROR("cannot start close thread: rc %d\n", err);
 553                 goto out_root;
 554         }
 555
 556 #ifdef CONFIG_FS_POSIX_ACL
 557         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 558                 rct_init(&sbi->ll_rct);
 559                 et_init(&sbi->ll_et);
 560         }
 561 #endif
 562
 563         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 564         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 565                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 566                                  NULL);
 567         cl_sb_init(sb);
 568
 569         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 570                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 571                                  &sbi->ll_cache, NULL);
 572
 573         sb->s_root = d_make_root(root);
 574         if (sb->s_root == NULL) {
 575                 CERROR("%s: can't make root dentry\n",
 576                         ll_get_fsname(sb, NULL, 0));
 577                 err = -ENOMEM;
 578                 goto out_lock_cn_cb;
 579         }
 580
 581         sbi->ll_sdev_orig = sb->s_dev;
 582
 583         /* We set sb->s_dev equal on all lustre clients in order to support
 584          * NFS export clustering.  NFSD requires that the FSID be the same
 585          * on all clients. */
 586         /* s_dev is also used in lt_compare() to compare two fs, but that is
 587          * only a node-local comparison. */
 588         uuid = obd_get_uuid(sbi->ll_md_exp);
 589         if (uuid != NULL) {
 590                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 591                 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
 592         }
 593
 594         if (data != NULL)
 595                 OBD_FREE_PTR(data);
 596         if (osfs != NULL)
 597                 OBD_FREE_PTR(osfs);
 598
 599         return err;
 600 out_root:
 601         iput(root);
 602 out_lock_cn_cb:
 603         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 604 out_dt:
 605         obd_disconnect(sbi->ll_dt_exp);
 606         sbi->ll_dt_exp = NULL;
 607         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 608         obd_zombie_barrier();
 609 out_md_fid:
 610         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 611 out_md:
 612         obd_disconnect(sbi->ll_md_exp);
 613         sbi->ll_md_exp = NULL;
 614 out:
 615         if (data != NULL)
 616                 OBD_FREE_PTR(data);
 617         if (osfs != NULL)
 618                 OBD_FREE_PTR(osfs);
 619         lprocfs_unregister_mountpoint(sbi);
 620         return err;
 621 }
 622
 623 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 624 {
 625         int size, rc;
 626
 627         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 628         size = sizeof(int);
 629         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 630                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 631         if (rc)
 632                 CERROR("Get max mdsize error rc %d \n", rc);
 633
 634         return rc;
 635 }
 636
 637 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 638 {
 639         int size, rc;
 640
 641         size = sizeof(int);
 642         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 643                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 644         if (rc)
 645                 CERROR("Get default mdsize error rc %d\n", rc);
 646
 647         return rc;
 648 }
 649
 650 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 651 {
 652         int size, rc;
 653
 654         size = sizeof(int);
 655         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 656                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 657         if (rc)
 658                 CERROR("Get max cookiesize error rc %d\n", rc);
 659
 660         return rc;
 661 }
 662
 663 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 664 {
 665         int size, rc;
 666
 667         size = sizeof(int);
 668         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 669                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 670         if (rc)
 671                 CERROR("Get default cookiesize error rc %d\n", rc);
 672
 673         return rc;
 674 }
 675
 676 static void ll_dump_inode(struct inode *inode)
 677 {
 678         struct ll_d_hlist_node *tmp;
 679         int dentry_count = 0;
 680
 681         LASSERT(inode != NULL);
 682
 683         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 684                 dentry_count++;
 685
 686         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 687                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 688                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 689 }
 690
 691 void lustre_dump_dentry(struct dentry *dentry, int recur)
 692 {
 693         struct list_head *tmp;
 694         int subdirs = 0;
 695
 696         LASSERT(dentry != NULL);
 697
 698         list_for_each(tmp, &dentry->d_subdirs)
 699                 subdirs++;
 700
 701         CERROR("dentry %p dump: name=%pd parent=%p, inode=%p, count=%u,"
 702                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry, dentry,
 703                dentry->d_parent, dentry->d_inode, d_count(dentry),
 704                dentry->d_flags, dentry->d_fsdata, subdirs);
 705         if (dentry->d_inode != NULL)
 706                 ll_dump_inode(dentry->d_inode);
 707
 708         if (recur == 0)
 709                 return;
 710
 711         list_for_each(tmp, &dentry->d_subdirs) {
 712                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 713                 lustre_dump_dentry(d, recur - 1);
 714         }
 715 }
 716
 717 static void client_common_put_super(struct super_block *sb)
 718 {
 719         struct ll_sb_info *sbi = ll_s2sbi(sb);
 720
 721 #ifdef CONFIG_FS_POSIX_ACL
 722         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 723                 et_fini(&sbi->ll_et);
 724                 rct_fini(&sbi->ll_rct);
 725         }
 726 #endif
 727
 728         ll_close_thread_shutdown(sbi->ll_lcq);
 729
 730         cl_sb_fini(sb);
 731
 732         list_del(&sbi->ll_conn_chain);
 733
 734         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 735         obd_disconnect(sbi->ll_dt_exp);
 736         sbi->ll_dt_exp = NULL;
 737         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 738          * see LU-2543. */
 739         obd_zombie_barrier();
 740
 741         lprocfs_unregister_mountpoint(sbi);
 742
 743         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 744         obd_disconnect(sbi->ll_md_exp);
 745         sbi->ll_md_exp = NULL;
 746 }
 747
 748 void ll_kill_super(struct super_block *sb)
 749 {
 750         struct ll_sb_info *sbi;
 751
 752         /* not init sb ?*/
 753         if (!(sb->s_flags & MS_ACTIVE))
 754                 return;
 755
 756         sbi = ll_s2sbi(sb);
 757         /* we need to restore s_dev from changed for clustered NFS before put_super
 758          * because new kernels have cached s_dev and change sb->s_dev in
 759          * put_super not affected real removing devices */
 760         if (sbi) {
 761                 sb->s_dev = sbi->ll_sdev_orig;
 762                 sbi->ll_umounting = 1;
 763         }
 764 }
 765
 766 static inline int ll_set_opt(const char *opt, char *data, int fl)
 767 {
 768         if (strncmp(opt, data, strlen(opt)) != 0)
 769                 return 0;
 770         else
 771                 return fl;
 772 }
 773
 774 /* non-client-specific mount options are parsed in lmd_parse */
 775 static int ll_options(char *options, int *flags)
 776 {
 777         int tmp;
 778         char *s1 = options, *s2;
 779
 780         if (!options)
 781                 return 0;
 782
 783         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 784
 785         while (*s1) {
 786                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 787                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 788                 if (tmp) {
 789                         *flags |= tmp;
 790                         goto next;
 791                 }
 792                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 793                 if (tmp) {
 794                         *flags |= tmp;
 795                         goto next;
 796                 }
 797                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 798                 if (tmp) {
 799                         *flags |= tmp;
 800                         goto next;
 801                 }
 802                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 803                 if (tmp) {
 804                         *flags &= ~tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 808                 if (tmp) {
 809                         *flags |= tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 813                 if (tmp) {
 814                         *flags &= ~tmp;
 815                         goto next;
 816                 }
 817 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 818                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 819                 if (tmp) {
 820                         /* Ignore deprecated mount option.  The client will
 821                          * always try to mount with ACL support, whether this
 822                          * is used depends on whether server supports it. */
 823                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 824                                                   "mount option 'acl'.\n");
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 828                 if (tmp) {
 829                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 830                                                   "mount option 'noacl'.\n");
 831                         goto next;
 832                 }
 833 #else
 834 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 835 #endif
 836                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 842                 if (tmp) {
 843                         *flags |= tmp;
 844                         goto next;
 845                 }
 846                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 847                 if (tmp) {
 848                         *flags &= ~tmp;
 849                         goto next;
 850                 }
 851
 852                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 858                 if (tmp) {
 859                         *flags &= ~tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 863                 if (tmp) {
 864                         *flags |= tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 868                 if (tmp) {
 869                         *flags &= ~tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 873                 if (tmp) {
 874                         *flags |= tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 878                 if (tmp) {
 879                         *flags &= ~tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 883                 if (tmp) {
 884                         *flags |= tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 888                 if (tmp) {
 889                         *flags |= tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 893                 if (tmp) {
 894                         *flags |= tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 898                 if (tmp) {
 899                         *flags &= ~tmp;
 900                         goto next;
 901                 }
 902                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 903                                    s1);
 904                 return -EINVAL;
 905
 906 next:
 907                 /* Find next opt */
 908                 s2 = strchr(s1, ',');
 909                 if (s2 == NULL)
 910                         break;
 911                 s1 = s2 + 1;
 912         }
 913         return 0;
 914 }
 915
 916 void ll_lli_init(struct ll_inode_info *lli)
 917 {
 918         lli->lli_inode_magic = LLI_INODE_MAGIC;
 919         lli->lli_flags = 0;
 920         lli->lli_ioepoch = 0;
 921         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 922         spin_lock_init(&lli->lli_lock);
 923         lli->lli_posix_acl = NULL;
 924         lli->lli_remote_perms = NULL;
 925         mutex_init(&lli->lli_rmtperm_mutex);
 926         /* Do not set lli_fid, it has been initialized already. */
 927         fid_zero(&lli->lli_pfid);
 928         INIT_LIST_HEAD(&lli->lli_close_list);
 929         INIT_LIST_HEAD(&lli->lli_oss_capas);
 930         atomic_set(&lli->lli_open_count, 0);
 931         lli->lli_mds_capa = NULL;
 932         lli->lli_rmtperm_time = 0;
 933         lli->lli_pending_och = NULL;
 934         lli->lli_mds_read_och = NULL;
 935         lli->lli_mds_write_och = NULL;
 936         lli->lli_mds_exec_och = NULL;
 937         lli->lli_open_fd_read_count = 0;
 938         lli->lli_open_fd_write_count = 0;
 939         lli->lli_open_fd_exec_count = 0;
 940         mutex_init(&lli->lli_och_mutex);
 941         spin_lock_init(&lli->lli_agl_lock);
 942         lli->lli_has_smd = false;
 943         spin_lock_init(&lli->lli_layout_lock);
 944         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 945         lli->lli_clob = NULL;
 946
 947         init_rwsem(&lli->lli_xattrs_list_rwsem);
 948         mutex_init(&lli->lli_xattrs_enq_lock);
 949
 950         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 951         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 952                 mutex_init(&lli->lli_readdir_mutex);
 953                 lli->lli_opendir_key = NULL;
 954                 lli->lli_sai = NULL;
 955                 spin_lock_init(&lli->lli_sa_lock);
 956                 lli->lli_opendir_pid = 0;
 957         } else {
 958                 mutex_init(&lli->lli_size_mutex);
 959                 lli->lli_symlink_name = NULL;
 960                 init_rwsem(&lli->lli_trunc_sem);
 961                 mutex_init(&lli->lli_write_mutex);
 962                 init_rwsem(&lli->lli_glimpse_sem);
 963                 lli->lli_glimpse_time = 0;
 964                 INIT_LIST_HEAD(&lli->lli_agl_list);
 965                 lli->lli_agl_index = 0;
 966                 lli->lli_async_rc = 0;
 967         }
 968         mutex_init(&lli->lli_layout_mutex);
 969 }
 970
 971 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 972 {
 973         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 974
 975         bdi->name = "lustre";
 976         return bdi_register(bdi, NULL, "lustre-%d",
 977                             atomic_inc_return(&ll_bdi_num));
 978 }
 979
 980 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 981 {
 982         struct lustre_profile *lprof = NULL;
 983         struct lustre_sb_info *lsi = s2lsi(sb);
 984         struct ll_sb_info *sbi;
 985         char  *dt = NULL, *md = NULL;
 986         char  *profilenm = get_profile_name(sb);
 987         struct config_llog_instance *cfg;
 988         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 989         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 990         int    err;
 991
 992         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 993
 994         cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
 995         if (!cfg)
 996                 return -ENOMEM;
 997
 998         try_module_get(THIS_MODULE);
 999
1000         /* client additional sb info */
1001         lsi->lsi_llsbi = sbi = ll_init_sbi();
1002         if (!sbi) {
1003                 module_put(THIS_MODULE);
1004                 OBD_FREE_PTR(cfg);
1005                 return -ENOMEM;
1006         }
1007
1008         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1009         if (err)
1010                 goto out_free;
1011
1012         err = bdi_init(&lsi->lsi_bdi);
1013         if (err)
1014                 goto out_free;
1015         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1016         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1017         err = ll_bdi_register(&lsi->lsi_bdi);
1018         if (err)
1019                 goto out_free;
1020
1021         sb->s_bdi = &lsi->lsi_bdi;
1022         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1023         sb->s_d_op = &ll_d_ops;
1024
1025         /* Generate a string unique to this super, in case some joker tries
1026            to mount the same fs at two mount points.
1027            Use the address of the super itself.*/
1028         cfg->cfg_instance = sb;
1029         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1030         cfg->cfg_callback = class_config_llog_handler;
1031         /* set up client obds */
1032         err = lustre_process_log(sb, profilenm, cfg);
1033         if (err < 0) {
1034                 CERROR("Unable to process log: %d\n", err);
1035                 goto out_free;
1036         }
1037
1038         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1039         lprof = class_get_profile(profilenm);
1040         if (lprof == NULL) {
1041                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1042                                    " read from the MGS.  Does that filesystem "
1043                                    "exist?\n", profilenm);
1044                 err = -EINVAL;
1045                 goto out_free;
1046         }
1047         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1048                lprof->lp_md, lprof->lp_dt);
1049
1050         dt = kzalloc(strlen(lprof->lp_dt) + instlen + 2, GFP_NOFS);
1051         if (!dt) {
1052                 err = -ENOMEM;
1053                 goto out_free;
1054         }
1055         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1056
1057         md = kzalloc(strlen(lprof->lp_md) + instlen + 2, GFP_NOFS);
1058         if (!md) {
1059                 err = -ENOMEM;
1060                 goto out_free;
1061         }
1062         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1063
1064         /* connections, registrations, sb setup */
1065         err = client_common_fill_super(sb, md, dt, mnt);
1066
1067 out_free:
1068         if (md)
1069                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1070         if (dt)
1071                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1072         if (err)
1073                 ll_put_super(sb);
1074         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1075                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1076
1077         OBD_FREE_PTR(cfg);
1078         return err;
1079 } /* ll_fill_super */
1080
1081 void ll_put_super(struct super_block *sb)
1082 {
1083         struct config_llog_instance cfg, params_cfg;
1084         struct obd_device *obd;
1085         struct lustre_sb_info *lsi = s2lsi(sb);
1086         struct ll_sb_info *sbi = ll_s2sbi(sb);
1087         char *profilenm = get_profile_name(sb);
1088         int next, force = 1;
1089
1090         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1091
1092         ll_print_capa_stat(sbi);
1093
1094         cfg.cfg_instance = sb;
1095         lustre_end_log(sb, profilenm, &cfg);
1096
1097         params_cfg.cfg_instance = sb;
1098         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1099
1100         if (sbi->ll_md_exp) {
1101                 obd = class_exp2obd(sbi->ll_md_exp);
1102                 if (obd)
1103                         force = obd->obd_force;
1104         }
1105
1106         /* We need to set force before the lov_disconnect in
1107            lustre_common_put_super, since l_d cleans up osc's as well. */
1108         if (force) {
1109                 next = 0;
1110                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1111                                                      &next)) != NULL) {
1112                         obd->obd_force = force;
1113                 }
1114         }
1115
1116         if (sbi->ll_lcq) {
1117                 /* Only if client_common_fill_super succeeded */
1118                 client_common_put_super(sb);
1119         }
1120
1121         next = 0;
1122         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1123                 class_manual_cleanup(obd);
1124         }
1125
1126         if (sbi->ll_flags & LL_SBI_VERBOSE)
1127                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1128
1129         if (profilenm)
1130                 class_del_profile(profilenm);
1131
1132         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1133                 bdi_destroy(&lsi->lsi_bdi);
1134                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1135         }
1136
1137         ll_free_sbi(sb);
1138         lsi->lsi_llsbi = NULL;
1139
1140         lustre_common_put_super(sb);
1141
1142         module_put(THIS_MODULE);
1143 } /* client_put_super */
1144
1145 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1146 {
1147         struct inode *inode = NULL;
1148
1149         /* NOTE: we depend on atomic igrab() -bzzz */
1150         lock_res_and_lock(lock);
1151         if (lock->l_resource->lr_lvb_inode) {
1152                 struct ll_inode_info *lli;
1153                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1154                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1155                         inode = igrab(lock->l_resource->lr_lvb_inode);
1156                 } else {
1157                         inode = lock->l_resource->lr_lvb_inode;
1158                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1159                                          D_WARNING, lock, "lr_lvb_inode %p is "
1160                                          "bogus: magic %08x",
1161                                          lock->l_resource->lr_lvb_inode,
1162                                          lli->lli_inode_magic);
1163                         inode = NULL;
1164                 }
1165         }
1166         unlock_res_and_lock(lock);
1167         return inode;
1168 }
1169
1170 void ll_clear_inode(struct inode *inode)
1171 {
1172         struct ll_inode_info *lli = ll_i2info(inode);
1173         struct ll_sb_info *sbi = ll_i2sbi(inode);
1174
1175         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1176                inode->i_generation, inode);
1177
1178         if (S_ISDIR(inode->i_mode)) {
1179                 /* these should have been cleared in ll_file_release */
1180                 LASSERT(lli->lli_opendir_key == NULL);
1181                 LASSERT(lli->lli_sai == NULL);
1182                 LASSERT(lli->lli_opendir_pid == 0);
1183         }
1184
1185         spin_lock(&lli->lli_lock);
1186         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1187         spin_unlock(&lli->lli_lock);
1188         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1189
1190         LASSERT(!lli->lli_open_fd_write_count);
1191         LASSERT(!lli->lli_open_fd_read_count);
1192         LASSERT(!lli->lli_open_fd_exec_count);
1193
1194         if (lli->lli_mds_write_och)
1195                 ll_md_real_close(inode, FMODE_WRITE);
1196         if (lli->lli_mds_exec_och)
1197                 ll_md_real_close(inode, FMODE_EXEC);
1198         if (lli->lli_mds_read_och)
1199                 ll_md_real_close(inode, FMODE_READ);
1200
1201         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1202                 OBD_FREE(lli->lli_symlink_name,
1203                          strlen(lli->lli_symlink_name) + 1);
1204                 lli->lli_symlink_name = NULL;
1205         }
1206
1207         ll_xattr_cache_destroy(inode);
1208
1209         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1210                 LASSERT(lli->lli_posix_acl == NULL);
1211                 if (lli->lli_remote_perms) {
1212                         free_rmtperm_hash(lli->lli_remote_perms);
1213                         lli->lli_remote_perms = NULL;
1214                 }
1215         }
1216 #ifdef CONFIG_FS_POSIX_ACL
1217         else if (lli->lli_posix_acl) {
1218                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1219                 LASSERT(lli->lli_remote_perms == NULL);
1220                 posix_acl_release(lli->lli_posix_acl);
1221                 lli->lli_posix_acl = NULL;
1222         }
1223 #endif
1224         lli->lli_inode_magic = LLI_INODE_DEAD;
1225
1226         ll_clear_inode_capas(inode);
1227         if (!S_ISDIR(inode->i_mode))
1228                 LASSERT(list_empty(&lli->lli_agl_list));
1229
1230         /*
1231          * XXX This has to be done before lsm is freed below, because
1232          * cl_object still uses inode lsm.
1233          */
1234         cl_inode_fini(inode);
1235         lli->lli_has_smd = false;
1236 }
1237
1238 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1239                   struct md_open_data **mod)
1240 {
1241         struct lustre_md md;
1242         struct inode *inode = dentry->d_inode;
1243         struct ll_sb_info *sbi = ll_i2sbi(inode);
1244         struct ptlrpc_request *request = NULL;
1245         int rc, ia_valid;
1246
1247         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1248                                      LUSTRE_OPC_ANY, NULL);
1249         if (IS_ERR(op_data))
1250                 return PTR_ERR(op_data);
1251
1252         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1253                         &request, mod);
1254         if (rc) {
1255                 ptlrpc_req_finished(request);
1256                 if (rc == -ENOENT) {
1257                         clear_nlink(inode);
1258                         /* Unlinked special device node? Or just a race?
1259                          * Pretend we done everything. */
1260                         if (!S_ISREG(inode->i_mode) &&
1261                             !S_ISDIR(inode->i_mode)) {
1262                                 ia_valid = op_data->op_attr.ia_valid;
1263                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1264                                 rc = simple_setattr(dentry, &op_data->op_attr);
1265                                 op_data->op_attr.ia_valid = ia_valid;
1266                         }
1267                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1268                         CERROR("md_setattr fails: rc = %d\n", rc);
1269                 }
1270                 return rc;
1271         }
1272
1273         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1274                               sbi->ll_md_exp, &md);
1275         if (rc) {
1276                 ptlrpc_req_finished(request);
1277                 return rc;
1278         }
1279
1280         ia_valid = op_data->op_attr.ia_valid;
1281         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1282          * cache is not cleared yet. */
1283         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1284         rc = simple_setattr(dentry, &op_data->op_attr);
1285         op_data->op_attr.ia_valid = ia_valid;
1286
1287         /* Extract epoch data if obtained. */
1288         op_data->op_handle = md.body->handle;
1289         op_data->op_ioepoch = md.body->ioepoch;
1290
1291         ll_update_inode(inode, &md);
1292         ptlrpc_req_finished(request);
1293
1294         return rc;
1295 }
1296
1297 /* Close IO epoch and send Size-on-MDS attribute update. */
1298 static int ll_setattr_done_writing(struct inode *inode,
1299                                    struct md_op_data *op_data,
1300                                    struct md_open_data *mod)
1301 {
1302         struct ll_inode_info *lli = ll_i2info(inode);
1303         int rc = 0;
1304
1305         LASSERT(op_data != NULL);
1306         if (!S_ISREG(inode->i_mode))
1307                 return 0;
1308
1309         CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1310                op_data->op_ioepoch, PFID(&lli->lli_fid));
1311
1312         op_data->op_flags = MF_EPOCH_CLOSE;
1313         ll_done_writing_attr(inode, op_data);
1314         ll_pack_inode2opdata(inode, op_data, NULL);
1315
1316         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1317         if (rc == -EAGAIN) {
1318                 /* MDS has instructed us to obtain Size-on-MDS attribute
1319                  * from OSTs and send setattr to back to MDS. */
1320                 rc = ll_som_update(inode, op_data);
1321         } else if (rc) {
1322                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1323                        inode->i_ino, rc);
1324         }
1325         return rc;
1326 }
1327
1328 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1329 {
1330         struct obd_capa *capa;
1331         int rc;
1332
1333         if (attr->ia_valid & ATTR_SIZE)
1334                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1335         else
1336                 capa = ll_mdscapa_get(inode);
1337
1338         rc = cl_setattr_ost(inode, attr, capa);
1339
1340         if (attr->ia_valid & ATTR_SIZE)
1341                 ll_truncate_free_capa(capa);
1342         else
1343                 capa_put(capa);
1344
1345         return rc;
1346 }
1347
1348
1349 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1350  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1351  * keep these values until such a time that objects are allocated for it.
1352  * We do the MDS operations first, as it is checking permissions for us.
1353  * We don't to the MDS RPC if there is nothing that we want to store there,
1354  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1355  * going to do an RPC anyways.
1356  *
1357  * If we are doing a truncate, we will send the mtime and ctime updates
1358  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1359  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1360  * at the same time.
1361  *
1362  * In case of HSMimport, we only set attr on MDS.
1363  */
1364 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1365 {
1366         struct inode *inode = dentry->d_inode;
1367         struct ll_inode_info *lli = ll_i2info(inode);
1368         struct md_op_data *op_data = NULL;
1369         struct md_open_data *mod = NULL;
1370         bool file_is_released = false;
1371         int rc = 0, rc1 = 0;
1372
1373         CDEBUG(D_VFSTRACE,
1374                 "%s: setattr inode %p/fid:"DFID
1375                 " from %llu to %llu, valid %x, hsm_import %d\n",
1376                 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1377                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1378                 attr->ia_valid, hsm_import);
1379
1380         if (attr->ia_valid & ATTR_SIZE) {
1381                 /* Check new size against VFS/VM file size limit and rlimit */
1382                 rc = inode_newsize_ok(inode, attr->ia_size);
1383                 if (rc)
1384                         return rc;
1385
1386                 /* The maximum Lustre file size is variable, based on the
1387                  * OST maximum object size and number of stripes.  This
1388                  * needs another check in addition to the VFS check above. */
1389                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1390                         CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1391                                PFID(&lli->lli_fid), attr->ia_size,
1392                                ll_file_maxbytes(inode));
1393                         return -EFBIG;
1394                 }
1395
1396                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1397         }
1398
1399         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1400         if (attr->ia_valid & TIMES_SET_FLAGS) {
1401                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1402                     !capable(CFS_CAP_FOWNER))
1403                         return -EPERM;
1404         }
1405
1406         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1407         if (attr->ia_valid & ATTR_CTIME) {
1408                 attr->ia_ctime = CURRENT_TIME;
1409                 attr->ia_valid |= ATTR_CTIME_SET;
1410         }
1411         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1412             (attr->ia_valid & ATTR_ATIME)) {
1413                 attr->ia_atime = CURRENT_TIME;
1414                 attr->ia_valid |= ATTR_ATIME_SET;
1415         }
1416         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1417             (attr->ia_valid & ATTR_MTIME)) {
1418                 attr->ia_mtime = CURRENT_TIME;
1419                 attr->ia_valid |= ATTR_MTIME_SET;
1420         }
1421
1422         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1423                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1424                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1425                        get_seconds());
1426
1427         /* If we are changing file size, file content is modified, flag it. */
1428         if (attr->ia_valid & ATTR_SIZE) {
1429                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1430                 spin_lock(&lli->lli_lock);
1431                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1432                 spin_unlock(&lli->lli_lock);
1433         }
1434
1435         /* We always do an MDS RPC, even if we're only changing the size;
1436          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1437
1438         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1439         if (!op_data)
1440                 return -ENOMEM;
1441
1442         if (!S_ISDIR(inode->i_mode)) {
1443                 if (attr->ia_valid & ATTR_SIZE)
1444                         inode_dio_write_done(inode);
1445                 mutex_unlock(&inode->i_mutex);
1446         }
1447
1448         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1449
1450         /* Open epoch for truncate. */
1451         if (exp_connect_som(ll_i2mdexp(inode)) &&
1452             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1453                 op_data->op_flags = MF_EPOCH_OPEN;
1454
1455         /* truncate on a released file must failed with -ENODATA,
1456          * so size must not be set on MDS for released file
1457          * but other attributes must be set
1458          */
1459         if (S_ISREG(inode->i_mode)) {
1460                 struct lov_stripe_md *lsm;
1461                 __u32 gen;
1462
1463                 ll_layout_refresh(inode, &gen);
1464                 lsm = ccc_inode_lsm_get(inode);
1465                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1466                         file_is_released = true;
1467                 ccc_inode_lsm_put(inode, lsm);
1468         }
1469
1470         /* if not in HSM import mode, clear size attr for released file
1471          * we clear the attribute send to MDT in op_data, not the original
1472          * received from caller in attr which is used later to
1473          * decide return code */
1474         if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1475                 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1476
1477         rc = ll_md_setattr(dentry, op_data, &mod);
1478         if (rc)
1479                 goto out;
1480
1481         /* truncate failed (only when non HSM import), others succeed */
1482         if (file_is_released) {
1483                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1484                         rc = -ENODATA;
1485                 else
1486                         rc = 0;
1487                 goto out;
1488         }
1489
1490         /* RPC to MDT is sent, cancel data modification flag */
1491         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1492                 spin_lock(&lli->lli_lock);
1493                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1494                 spin_unlock(&lli->lli_lock);
1495         }
1496
1497         ll_ioepoch_open(lli, op_data->op_ioepoch);
1498         if (!S_ISREG(inode->i_mode)) {
1499                 rc = 0;
1500                 goto out;
1501         }
1502
1503         if (attr->ia_valid & (ATTR_SIZE |
1504                               ATTR_ATIME | ATTR_ATIME_SET |
1505                               ATTR_MTIME | ATTR_MTIME_SET))
1506                 /* For truncate and utimes sending attributes to OSTs, setting
1507                  * mtime/atime to the past will be performed under PW [0:EOF]
1508                  * extent lock (new_size:EOF for truncate).  It may seem
1509                  * excessive to send mtime/atime updates to OSTs when not
1510                  * setting times to past, but it is necessary due to possible
1511                  * time de-synchronization between MDT inode and OST objects */
1512                 if (attr->ia_valid & ATTR_SIZE)
1513                         down_write(&lli->lli_trunc_sem);
1514                 rc = ll_setattr_ost(inode, attr);
1515                 if (attr->ia_valid & ATTR_SIZE)
1516                         up_write(&lli->lli_trunc_sem);
1517 out:
1518         if (op_data) {
1519                 if (op_data->op_ioepoch) {
1520                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1521                         if (!rc)
1522                                 rc = rc1;
1523                 }
1524                 ll_finish_md_op_data(op_data);
1525         }
1526         if (!S_ISDIR(inode->i_mode)) {
1527                 mutex_lock(&inode->i_mutex);
1528                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1529                         inode_dio_wait(inode);
1530         }
1531
1532         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1533                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1534
1535         return rc;
1536 }
1537
1538 int ll_setattr(struct dentry *de, struct iattr *attr)
1539 {
1540         int mode = de->d_inode->i_mode;
1541
1542         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1543                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1544                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1545
1546         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1547                                (ATTR_SIZE|ATTR_MODE)) &&
1548             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1549              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1550               !(attr->ia_mode & S_ISGID))))
1551                 attr->ia_valid |= ATTR_FORCE;
1552
1553         if ((attr->ia_valid & ATTR_MODE) &&
1554             (mode & S_ISUID) &&
1555             !(attr->ia_mode & S_ISUID) &&
1556             !(attr->ia_valid & ATTR_KILL_SUID))
1557                 attr->ia_valid |= ATTR_KILL_SUID;
1558
1559         if ((attr->ia_valid & ATTR_MODE) &&
1560             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1561             !(attr->ia_mode & S_ISGID) &&
1562             !(attr->ia_valid & ATTR_KILL_SGID))
1563                 attr->ia_valid |= ATTR_KILL_SGID;
1564
1565         return ll_setattr_raw(de, attr, false);
1566 }
1567
1568 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1569                        __u64 max_age, __u32 flags)
1570 {
1571         struct ll_sb_info *sbi = ll_s2sbi(sb);
1572         struct obd_statfs obd_osfs;
1573         int rc;
1574
1575         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1576         if (rc) {
1577                 CERROR("md_statfs fails: rc = %d\n", rc);
1578                 return rc;
1579         }
1580
1581         osfs->os_type = sb->s_magic;
1582
1583         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1584                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1585                osfs->os_files);
1586
1587         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1588                 flags |= OBD_STATFS_NODELAY;
1589
1590         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1591         if (rc) {
1592                 CERROR("obd_statfs fails: rc = %d\n", rc);
1593                 return rc;
1594         }
1595
1596         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1597                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1598                obd_osfs.os_files);
1599
1600         osfs->os_bsize = obd_osfs.os_bsize;
1601         osfs->os_blocks = obd_osfs.os_blocks;
1602         osfs->os_bfree = obd_osfs.os_bfree;
1603         osfs->os_bavail = obd_osfs.os_bavail;
1604
1605         /* If we don't have as many objects free on the OST as inodes
1606          * on the MDS, we reduce the total number of inodes to
1607          * compensate, so that the "inodes in use" number is correct.
1608          */
1609         if (obd_osfs.os_ffree < osfs->os_ffree) {
1610                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1611                         obd_osfs.os_ffree;
1612                 osfs->os_ffree = obd_osfs.os_ffree;
1613         }
1614
1615         return rc;
1616 }
1617 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1618 {
1619         struct super_block *sb = de->d_sb;
1620         struct obd_statfs osfs;
1621         int rc;
1622
1623         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1624         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1625
1626         /* Some amount of caching on the client is allowed */
1627         rc = ll_statfs_internal(sb, &osfs,
1628                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1629                                 0);
1630         if (rc)
1631                 return rc;
1632
1633         statfs_unpack(sfs, &osfs);
1634
1635         /* We need to downshift for all 32-bit kernels, because we can't
1636          * tell if the kernel is being called via sys_statfs64() or not.
1637          * Stop before overflowing f_bsize - in which case it is better
1638          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1639         if (sizeof(long) < 8) {
1640                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1641                         sfs->f_bsize <<= 1;
1642
1643                         osfs.os_blocks >>= 1;
1644                         osfs.os_bfree >>= 1;
1645                         osfs.os_bavail >>= 1;
1646                 }
1647         }
1648
1649         sfs->f_blocks = osfs.os_blocks;
1650         sfs->f_bfree = osfs.os_bfree;
1651         sfs->f_bavail = osfs.os_bavail;
1652         sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1653         return 0;
1654 }
1655
1656 void ll_inode_size_lock(struct inode *inode)
1657 {
1658         struct ll_inode_info *lli;
1659
1660         LASSERT(!S_ISDIR(inode->i_mode));
1661
1662         lli = ll_i2info(inode);
1663         mutex_lock(&lli->lli_size_mutex);
1664 }
1665
1666 void ll_inode_size_unlock(struct inode *inode)
1667 {
1668         struct ll_inode_info *lli;
1669
1670         lli = ll_i2info(inode);
1671         mutex_unlock(&lli->lli_size_mutex);
1672 }
1673
1674 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1675 {
1676         struct ll_inode_info *lli = ll_i2info(inode);
1677         struct mdt_body *body = md->body;
1678         struct lov_stripe_md *lsm = md->lsm;
1679         struct ll_sb_info *sbi = ll_i2sbi(inode);
1680
1681         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1682         if (lsm != NULL) {
1683                 if (!lli->lli_has_smd &&
1684                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1685                         cl_file_inode_init(inode, md);
1686
1687                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1688                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1689                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1690         }
1691
1692         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1693                 if (body->valid & OBD_MD_FLRMTPERM)
1694                         ll_update_remote_perm(inode, md->remote_perm);
1695         }
1696 #ifdef CONFIG_FS_POSIX_ACL
1697         else if (body->valid & OBD_MD_FLACL) {
1698                 spin_lock(&lli->lli_lock);
1699                 if (lli->lli_posix_acl)
1700                         posix_acl_release(lli->lli_posix_acl);
1701                 lli->lli_posix_acl = md->posix_acl;
1702                 spin_unlock(&lli->lli_lock);
1703         }
1704 #endif
1705         inode->i_ino = cl_fid_build_ino(&body->fid1,
1706                                         sbi->ll_flags & LL_SBI_32BIT_API);
1707         inode->i_generation = cl_fid_build_gen(&body->fid1);
1708
1709         if (body->valid & OBD_MD_FLATIME) {
1710                 if (body->atime > LTIME_S(inode->i_atime))
1711                         LTIME_S(inode->i_atime) = body->atime;
1712                 lli->lli_lvb.lvb_atime = body->atime;
1713         }
1714         if (body->valid & OBD_MD_FLMTIME) {
1715                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1716                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1717                                inode->i_ino, LTIME_S(inode->i_mtime),
1718                                body->mtime);
1719                         LTIME_S(inode->i_mtime) = body->mtime;
1720                 }
1721                 lli->lli_lvb.lvb_mtime = body->mtime;
1722         }
1723         if (body->valid & OBD_MD_FLCTIME) {
1724                 if (body->ctime > LTIME_S(inode->i_ctime))
1725                         LTIME_S(inode->i_ctime) = body->ctime;
1726                 lli->lli_lvb.lvb_ctime = body->ctime;
1727         }
1728         if (body->valid & OBD_MD_FLMODE)
1729                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1730         if (body->valid & OBD_MD_FLTYPE)
1731                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1732         LASSERT(inode->i_mode != 0);
1733         if (S_ISREG(inode->i_mode)) {
1734                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1735         } else {
1736                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1737         }
1738         if (body->valid & OBD_MD_FLUID)
1739                 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1740         if (body->valid & OBD_MD_FLGID)
1741                 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1742         if (body->valid & OBD_MD_FLFLAGS)
1743                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1744         if (body->valid & OBD_MD_FLNLINK)
1745                 set_nlink(inode, body->nlink);
1746         if (body->valid & OBD_MD_FLRDEV)
1747                 inode->i_rdev = old_decode_dev(body->rdev);
1748
1749         if (body->valid & OBD_MD_FLID) {
1750                 /* FID shouldn't be changed! */
1751                 if (fid_is_sane(&lli->lli_fid)) {
1752                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1753                                  "Trying to change FID "DFID
1754                                  " to the "DFID", inode %lu/%u(%p)\n",
1755                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1756                                  inode->i_ino, inode->i_generation, inode);
1757                 } else
1758                         lli->lli_fid = body->fid1;
1759         }
1760
1761         LASSERT(fid_seq(&lli->lli_fid) != 0);
1762
1763         if (body->valid & OBD_MD_FLSIZE) {
1764                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1765                     S_ISREG(inode->i_mode)) {
1766                         struct lustre_handle lockh;
1767                         ldlm_mode_t mode;
1768
1769                         /* As it is possible a blocking ast has been processed
1770                          * by this time, we need to check there is an UPDATE
1771                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1772                          * it. */
1773                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1774                                                &lockh, LDLM_FL_CBPENDING,
1775                                                LCK_CR | LCK_CW |
1776                                                LCK_PR | LCK_PW);
1777                         if (mode) {
1778                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1779                                                       LLIF_EPOCH_PENDING |
1780                                                       LLIF_SOM_DIRTY)) {
1781                                         CERROR("ino %lu flags %u still has "
1782                                                "size authority! do not trust "
1783                                                "the size got from MDS\n",
1784                                                inode->i_ino, lli->lli_flags);
1785                                 } else {
1786                                         /* Use old size assignment to avoid
1787                                          * deadlock bz14138 & bz14326 */
1788                                         i_size_write(inode, body->size);
1789                                         spin_lock(&lli->lli_lock);
1790                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1791                                         spin_unlock(&lli->lli_lock);
1792                                 }
1793                                 ldlm_lock_decref(&lockh, mode);
1794                         }
1795                 } else {
1796                         /* Use old size assignment to avoid
1797                          * deadlock bz14138 & bz14326 */
1798                         i_size_write(inode, body->size);
1799
1800                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1801                                inode->i_ino, (unsigned long long)body->size);
1802                 }
1803
1804                 if (body->valid & OBD_MD_FLBLOCKS)
1805                         inode->i_blocks = body->blocks;
1806         }
1807
1808         if (body->valid & OBD_MD_FLMDSCAPA) {
1809                 LASSERT(md->mds_capa);
1810                 ll_add_capa(inode, md->mds_capa);
1811         }
1812         if (body->valid & OBD_MD_FLOSSCAPA) {
1813                 LASSERT(md->oss_capa);
1814                 ll_add_capa(inode, md->oss_capa);
1815         }
1816
1817         if (body->valid & OBD_MD_TSTATE) {
1818                 if (body->t_state & MS_RESTORE)
1819                         lli->lli_flags |= LLIF_FILE_RESTORING;
1820         }
1821 }
1822
1823 void ll_read_inode2(struct inode *inode, void *opaque)
1824 {
1825         struct lustre_md *md = opaque;
1826         struct ll_inode_info *lli = ll_i2info(inode);
1827
1828         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1829                PFID(&lli->lli_fid), inode);
1830
1831         LASSERT(!lli->lli_has_smd);
1832
1833         /* Core attributes from the MDS first.  This is a new inode, and
1834          * the VFS doesn't zero times in the core inode so we have to do
1835          * it ourselves.  They will be overwritten by either MDS or OST
1836          * attributes - we just need to make sure they aren't newer. */
1837         LTIME_S(inode->i_mtime) = 0;
1838         LTIME_S(inode->i_atime) = 0;
1839         LTIME_S(inode->i_ctime) = 0;
1840         inode->i_rdev = 0;
1841         ll_update_inode(inode, md);
1842
1843         /* OIDEBUG(inode); */
1844
1845         /* initializing backing dev info. */
1846         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1847
1848
1849         if (S_ISREG(inode->i_mode)) {
1850                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1851                 inode->i_op = &ll_file_inode_operations;
1852                 inode->i_fop = sbi->ll_fop;
1853                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1854         } else if (S_ISDIR(inode->i_mode)) {
1855                 inode->i_op = &ll_dir_inode_operations;
1856                 inode->i_fop = &ll_dir_operations;
1857         } else if (S_ISLNK(inode->i_mode)) {
1858                 inode->i_op = &ll_fast_symlink_inode_operations;
1859         } else {
1860                 inode->i_op = &ll_special_inode_operations;
1861
1862                 init_special_inode(inode, inode->i_mode,
1863                                    inode->i_rdev);
1864         }
1865 }
1866
1867 void ll_delete_inode(struct inode *inode)
1868 {
1869         struct cl_inode_info *lli = cl_i2info(inode);
1870
1871         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1872                 /* discard all dirty pages before truncating them, required by
1873                  * osc_extent implementation at LU-1030. */
1874                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1875                                    CL_FSYNC_DISCARD, 1);
1876
1877         truncate_inode_pages_final(&inode->i_data);
1878
1879         /* Workaround for LU-118 */
1880         if (inode->i_data.nrpages) {
1881                 TREE_READ_LOCK_IRQ(&inode->i_data);
1882                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1883                 LASSERTF(inode->i_data.nrpages == 0,
1884                          "inode=%lu/%u(%p) nrpages=%lu, see "
1885                          "http://jira.whamcloud.com/browse/LU-118\n",
1886                          inode->i_ino, inode->i_generation, inode,
1887                          inode->i_data.nrpages);
1888         }
1889         /* Workaround end */
1890
1891         ll_clear_inode(inode);
1892         clear_inode(inode);
1893 }
1894
1895 int ll_iocontrol(struct inode *inode, struct file *file,
1896                  unsigned int cmd, unsigned long arg)
1897 {
1898         struct ll_sb_info *sbi = ll_i2sbi(inode);
1899         struct ptlrpc_request *req = NULL;
1900         int rc, flags = 0;
1901
1902         switch (cmd) {
1903         case FSFILT_IOC_GETFLAGS: {
1904                 struct mdt_body *body;
1905                 struct md_op_data *op_data;
1906
1907                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1908                                              0, 0, LUSTRE_OPC_ANY,
1909                                              NULL);
1910                 if (IS_ERR(op_data))
1911                         return PTR_ERR(op_data);
1912
1913                 op_data->op_valid = OBD_MD_FLFLAGS;
1914                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1915                 ll_finish_md_op_data(op_data);
1916                 if (rc) {
1917                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1918                         return -abs(rc);
1919                 }
1920
1921                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1922
1923                 flags = body->flags;
1924
1925                 ptlrpc_req_finished(req);
1926
1927                 return put_user(flags, (int *)arg);
1928         }
1929         case FSFILT_IOC_SETFLAGS: {
1930                 struct lov_stripe_md *lsm;
1931                 struct obd_info oinfo = { { { 0 } } };
1932                 struct md_op_data *op_data;
1933
1934                 if (get_user(flags, (int *)arg))
1935                         return -EFAULT;
1936
1937                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1938                                              LUSTRE_OPC_ANY, NULL);
1939                 if (IS_ERR(op_data))
1940                         return PTR_ERR(op_data);
1941
1942                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1943                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1944                 rc = md_setattr(sbi->ll_md_exp, op_data,
1945                                 NULL, 0, NULL, 0, &req, NULL);
1946                 ll_finish_md_op_data(op_data);
1947                 ptlrpc_req_finished(req);
1948                 if (rc)
1949                         return rc;
1950
1951                 inode->i_flags = ll_ext_to_inode_flags(flags);
1952
1953                 lsm = ccc_inode_lsm_get(inode);
1954                 if (!lsm_has_objects(lsm)) {
1955                         ccc_inode_lsm_put(inode, lsm);
1956                         return 0;
1957                 }
1958
1959                 OBDO_ALLOC(oinfo.oi_oa);
1960                 if (!oinfo.oi_oa) {
1961                         ccc_inode_lsm_put(inode, lsm);
1962                         return -ENOMEM;
1963                 }
1964                 oinfo.oi_md = lsm;
1965                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1966                 oinfo.oi_oa->o_flags = flags;
1967                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1968                                        OBD_MD_FLGROUP;
1969                 oinfo.oi_capa = ll_mdscapa_get(inode);
1970                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1971                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1972                 capa_put(oinfo.oi_capa);
1973                 OBDO_FREE(oinfo.oi_oa);
1974                 ccc_inode_lsm_put(inode, lsm);
1975
1976                 if (rc && rc != -EPERM && rc != -EACCES)
1977                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1978
1979                 return rc;
1980         }
1981         default:
1982                 return -ENOSYS;
1983         }
1984
1985         return 0;
1986 }
1987
1988 int ll_flush_ctx(struct inode *inode)
1989 {
1990         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1991
1992         CDEBUG(D_SEC, "flush context for user %d\n",
1993                       from_kuid(&init_user_ns, current_uid()));
1994
1995         obd_set_info_async(NULL, sbi->ll_md_exp,
1996                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1997                            0, NULL, NULL);
1998         obd_set_info_async(NULL, sbi->ll_dt_exp,
1999                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2000                            0, NULL, NULL);
2001         return 0;
2002 }
2003
2004 /* umount -f client means force down, don't save state */
2005 void ll_umount_begin(struct super_block *sb)
2006 {
2007         struct ll_sb_info *sbi = ll_s2sbi(sb);
2008         struct obd_device *obd;
2009         struct obd_ioctl_data *ioc_data;
2010
2011         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2012                sb->s_count, atomic_read(&sb->s_active));
2013
2014         obd = class_exp2obd(sbi->ll_md_exp);
2015         if (obd == NULL) {
2016                 CERROR("Invalid MDC connection handle %#llx\n",
2017                        sbi->ll_md_exp->exp_handle.h_cookie);
2018                 return;
2019         }
2020         obd->obd_force = 1;
2021
2022         obd = class_exp2obd(sbi->ll_dt_exp);
2023         if (obd == NULL) {
2024                 CERROR("Invalid LOV connection handle %#llx\n",
2025                        sbi->ll_dt_exp->exp_handle.h_cookie);
2026                 return;
2027         }
2028         obd->obd_force = 1;
2029
2030         ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
2031         if (ioc_data) {
2032                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2033                               sizeof(*ioc_data), ioc_data, NULL);
2034
2035                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2036                               sizeof(*ioc_data), ioc_data, NULL);
2037
2038                 OBD_FREE_PTR(ioc_data);
2039         }
2040
2041         /* Really, we'd like to wait until there are no requests outstanding,
2042          * and then continue.  For now, we just invalidate the requests,
2043          * schedule() and sleep one second if needed, and hope.
2044          */
2045         schedule();
2046 }
2047
2048 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2049 {
2050         struct ll_sb_info *sbi = ll_s2sbi(sb);
2051         char *profilenm = get_profile_name(sb);
2052         int err;
2053         __u32 read_only;
2054
2055         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2056                 read_only = *flags & MS_RDONLY;
2057                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2058                                          sizeof(KEY_READ_ONLY),
2059                                          KEY_READ_ONLY, sizeof(read_only),
2060                                          &read_only, NULL);
2061                 if (err) {
2062                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2063                                       profilenm, read_only ?
2064                                       "read-only" : "read-write", err);
2065                         return err;
2066                 }
2067
2068                 if (read_only)
2069                         sb->s_flags |= MS_RDONLY;
2070                 else
2071                         sb->s_flags &= ~MS_RDONLY;
2072
2073                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2074                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2075                                       read_only ?  "read-only" : "read-write");
2076         }
2077         return 0;
2078 }
2079
2080 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2081                   struct super_block *sb, struct lookup_intent *it)
2082 {
2083         struct ll_sb_info *sbi = NULL;
2084         struct lustre_md md;
2085         int rc;
2086
2087         LASSERT(*inode || sb);
2088         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2089         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2090                               sbi->ll_md_exp, &md);
2091         if (rc)
2092                 return rc;
2093
2094         if (*inode) {
2095                 ll_update_inode(*inode, &md);
2096         } else {
2097                 LASSERT(sb != NULL);
2098
2099                 /*
2100                  * At this point server returns to client's same fid as client
2101                  * generated for creating. So using ->fid1 is okay here.
2102                  */
2103                 LASSERT(fid_is_sane(&md.body->fid1));
2104
2105                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2106                                              sbi->ll_flags & LL_SBI_32BIT_API),
2107                                  &md);
2108                 if (*inode == NULL || IS_ERR(*inode)) {
2109 #ifdef CONFIG_FS_POSIX_ACL
2110                         if (md.posix_acl) {
2111                                 posix_acl_release(md.posix_acl);
2112                                 md.posix_acl = NULL;
2113                         }
2114 #endif
2115                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2116                         *inode = NULL;
2117                         CERROR("new_inode -fatal: rc %d\n", rc);
2118                         goto out;
2119                 }
2120         }
2121
2122         /* Handling piggyback layout lock.
2123          * Layout lock can be piggybacked by getattr and open request.
2124          * The lsm can be applied to inode only if it comes with a layout lock
2125          * otherwise correct layout may be overwritten, for example:
2126          * 1. proc1: mdt returns a lsm but not granting layout
2127          * 2. layout was changed by another client
2128          * 3. proc2: refresh layout and layout lock granted
2129          * 4. proc1: to apply a stale layout */
2130         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2131                 struct lustre_handle lockh;
2132                 struct ldlm_lock *lock;
2133
2134                 lockh.cookie = it->d.lustre.it_lock_handle;
2135                 lock = ldlm_handle2lock(&lockh);
2136                 LASSERT(lock != NULL);
2137                 if (ldlm_has_layout(lock)) {
2138                         struct cl_object_conf conf;
2139
2140                         memset(&conf, 0, sizeof(conf));
2141                         conf.coc_opc = OBJECT_CONF_SET;
2142                         conf.coc_inode = *inode;
2143                         conf.coc_lock = lock;
2144                         conf.u.coc_md = &md;
2145                         (void)ll_layout_conf(*inode, &conf);
2146                 }
2147                 LDLM_LOCK_PUT(lock);
2148         }
2149
2150 out:
2151         if (md.lsm != NULL)
2152                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2153         md_free_lustre_md(sbi->ll_md_exp, &md);
2154         return rc;
2155 }
2156
2157 int ll_obd_statfs(struct inode *inode, void *arg)
2158 {
2159         struct ll_sb_info *sbi = NULL;
2160         struct obd_export *exp;
2161         char *buf = NULL;
2162         struct obd_ioctl_data *data = NULL;
2163         __u32 type;
2164         __u32 flags;
2165         int len = 0, rc;
2166
2167         if (!inode || !(sbi = ll_i2sbi(inode))) {
2168                 rc = -EINVAL;
2169                 goto out_statfs;
2170         }
2171
2172         rc = obd_ioctl_getdata(&buf, &len, arg);
2173         if (rc)
2174                 goto out_statfs;
2175
2176         data = (void *)buf;
2177         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2178             !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2179                 rc = -EINVAL;
2180                 goto out_statfs;
2181         }
2182
2183         if (data->ioc_inllen1 != sizeof(__u32) ||
2184             data->ioc_inllen2 != sizeof(__u32) ||
2185             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2186             data->ioc_plen2 != sizeof(struct obd_uuid)) {
2187                 rc = -EINVAL;
2188                 goto out_statfs;
2189         }
2190
2191         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2192         if (type & LL_STATFS_LMV)
2193                 exp = sbi->ll_md_exp;
2194         else if (type & LL_STATFS_LOV)
2195                 exp = sbi->ll_dt_exp;
2196         else {
2197                 rc = -ENODEV;
2198                 goto out_statfs;
2199         }
2200
2201         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2202         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2203         if (rc)
2204                 goto out_statfs;
2205 out_statfs:
2206         if (buf)
2207                 obd_ioctl_freedata(buf, len);
2208         return rc;
2209 }
2210
2211 int ll_process_config(struct lustre_cfg *lcfg)
2212 {
2213         char *ptr;
2214         void *sb;
2215         struct lprocfs_static_vars lvars;
2216         unsigned long x;
2217         int rc = 0;
2218
2219         lprocfs_llite_init_vars(&lvars);
2220
2221         /* The instance name contains the sb: lustre-client-aacfe000 */
2222         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2223         if (!ptr || !*(++ptr))
2224                 return -EINVAL;
2225         if (sscanf(ptr, "%lx", &x) != 1)
2226                 return -EINVAL;
2227         sb = (void *)x;
2228         /* This better be a real Lustre superblock! */
2229         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2230
2231         /* Note we have not called client_common_fill_super yet, so
2232            proc fns must be able to handle that! */
2233         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2234                                       lcfg, sb);
2235         if (rc > 0)
2236                 rc = 0;
2237         return rc;
2238 }
2239
2240 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2241 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2242                                        struct inode *i1, struct inode *i2,
2243                                        const char *name, int namelen,
2244                                        int mode, __u32 opc, void *data)
2245 {
2246         LASSERT(i1 != NULL);
2247
2248         if (namelen > ll_i2sbi(i1)->ll_namelen)
2249                 return ERR_PTR(-ENAMETOOLONG);
2250
2251         if (op_data == NULL)
2252                 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2253
2254         if (op_data == NULL)
2255                 return ERR_PTR(-ENOMEM);
2256
2257         ll_i2gids(op_data->op_suppgids, i1, i2);
2258         op_data->op_fid1 = *ll_inode2fid(i1);
2259         op_data->op_capa1 = ll_mdscapa_get(i1);
2260
2261         if (i2) {
2262                 op_data->op_fid2 = *ll_inode2fid(i2);
2263                 op_data->op_capa2 = ll_mdscapa_get(i2);
2264         } else {
2265                 fid_zero(&op_data->op_fid2);
2266                 op_data->op_capa2 = NULL;
2267         }
2268
2269         op_data->op_name = name;
2270         op_data->op_namelen = namelen;
2271         op_data->op_mode = mode;
2272         op_data->op_mod_time = get_seconds();
2273         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2274         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2275         op_data->op_cap = cfs_curproc_cap_pack();
2276         op_data->op_bias = 0;
2277         op_data->op_cli_flags = 0;
2278         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2279              filename_is_volatile(name, namelen, NULL))
2280                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2281         op_data->op_opc = opc;
2282         op_data->op_mds = 0;
2283         op_data->op_data = data;
2284
2285         /* If the file is being opened after mknod() (normally due to NFS)
2286          * try to use the default stripe data from parent directory for
2287          * allocating OST objects.  Try to pass the parent FID to MDS. */
2288         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2289             !ll_i2info(i2)->lli_has_smd) {
2290                 struct ll_inode_info *lli = ll_i2info(i2);
2291
2292                 spin_lock(&lli->lli_lock);
2293                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2294                         op_data->op_fid1 = lli->lli_pfid;
2295                 spin_unlock(&lli->lli_lock);
2296                 /** We ignore parent's capability temporary. */
2297         }
2298
2299         /* When called by ll_setattr_raw, file is i1. */
2300         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2301                 op_data->op_bias |= MDS_DATA_MODIFIED;
2302
2303         return op_data;
2304 }
2305
2306 void ll_finish_md_op_data(struct md_op_data *op_data)
2307 {
2308         capa_put(op_data->op_capa1);
2309         capa_put(op_data->op_capa2);
2310         OBD_FREE_PTR(op_data);
2311 }
2312
2313 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2314 {
2315         struct ll_sb_info *sbi;
2316
2317         LASSERT((seq != NULL) && (dentry != NULL));
2318         sbi = ll_s2sbi(dentry->d_sb);
2319
2320         if (sbi->ll_flags & LL_SBI_NOLCK)
2321                 seq_puts(seq, ",nolock");
2322
2323         if (sbi->ll_flags & LL_SBI_FLOCK)
2324                 seq_puts(seq, ",flock");
2325
2326         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2327                 seq_puts(seq, ",localflock");
2328
2329         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2330                 seq_puts(seq, ",user_xattr");
2331
2332         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2333                 seq_puts(seq, ",lazystatfs");
2334
2335         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2336                 seq_puts(seq, ",user_fid2path");
2337
2338         return 0;
2339 }
2340
2341 /**
2342  * Get obd name by cmd, and copy out to user space
2343  */
2344 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2345 {
2346         struct ll_sb_info *sbi = ll_i2sbi(inode);
2347         struct obd_device *obd;
2348
2349         if (cmd == OBD_IOC_GETDTNAME)
2350                 obd = class_exp2obd(sbi->ll_dt_exp);
2351         else if (cmd == OBD_IOC_GETMDNAME)
2352                 obd = class_exp2obd(sbi->ll_md_exp);
2353         else
2354                 return -EINVAL;
2355
2356         if (!obd)
2357                 return -ENOENT;
2358
2359         if (copy_to_user((void *)arg, obd->obd_name,
2360                              strlen(obd->obd_name) + 1))
2361                 return -EFAULT;
2362
2363         return 0;
2364 }
2365
2366 /**
2367  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2368  * fsname will be returned in this buffer; otherwise, a static buffer will be
2369  * used to store the fsname and returned to caller.
2370  */
2371 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2372 {
2373         static char fsname_static[MTI_NAME_MAXLEN];
2374         struct lustre_sb_info *lsi = s2lsi(sb);
2375         char *ptr;
2376         int len;
2377
2378         if (buf == NULL) {
2379                 /* this means the caller wants to use static buffer
2380                  * and it doesn't care about race. Usually this is
2381                  * in error reporting path */
2382                 buf = fsname_static;
2383                 buflen = sizeof(fsname_static);
2384         }
2385
2386         len = strlen(lsi->lsi_lmd->lmd_profile);
2387         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2388         if (ptr && (strcmp(ptr, "-client") == 0))
2389                 len -= 7;
2390
2391         if (unlikely(len >= buflen))
2392                 len = buflen - 1;
2393         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2394         buf[len] = '\0';
2395
2396         return buf;
2397 }
2398
2399 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2400 {
2401         char *path = NULL;
2402
2403         struct path p;
2404
2405         p.dentry = dentry;
2406         p.mnt = current->fs->root.mnt;
2407         path_get(&p);
2408         path = d_path(&p, buf, bufsize);
2409         path_put(&p);
2410
2411         return path;
2412 }
2413
2414 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2415 {
2416         char *buf, *path = NULL;
2417         struct dentry *dentry = NULL;
2418         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2419
2420         /* this can be called inside spin lock so use GFP_ATOMIC. */
2421         buf = (char *)__get_free_page(GFP_ATOMIC);
2422         if (buf != NULL) {
2423                 dentry = d_find_alias(page->mapping->host);
2424                 if (dentry != NULL)
2425                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2426         }
2427
2428         CDEBUG(D_WARNING,
2429                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2430                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2431                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2432                PFID(&obj->cob_header.coh_lu.loh_fid),
2433                (path && !IS_ERR(path)) ? path : "", ioret);
2434
2435         if (dentry != NULL)
2436                 dput(dentry);
2437
2438         if (buf != NULL)
2439                 free_page((unsigned long)buf);
2440 }