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, 2013, 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/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_ioctl.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 struct kmem_cache *ll_file_data_slab;
  61
  62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(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         ENTRY;
  78
  79         OBD_ALLOC_PTR(sbi);
  80         if (sbi == NULL)
  81                 RETURN(NULL);
  82
  83         spin_lock_init(&sbi->ll_lock);
  84         mutex_init(&sbi->ll_lco.lco_lock);
  85         spin_lock_init(&sbi->ll_pp_extent_lock);
  86         spin_lock_init(&sbi->ll_process_lock);
  87         sbi->ll_rw_stats_on = 0;
  88
  89         si_meminfo(&si);
  90         pages = si.totalram - si.totalhigh;
  91         lru_page_max = pages / 2;
  92
  93         /* initialize ll_cache data */
  94         atomic_set(&sbi->ll_cache.ccc_users, 0);
  95         sbi->ll_cache.ccc_lru_max = lru_page_max;
  96         atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
  97         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
  98         INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
  99
 100         /* turn unstable check off by default as it impacts performance */
 101         sbi->ll_cache.ccc_unstable_check = 0;
 102         atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 103         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 104
 105         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 106                                            SBI_DEFAULT_READAHEAD_MAX);
 107         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 108         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 109                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 110         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 111         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 112
 113         ll_generate_random_uuid(uuid);
 114         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 115         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 116
 117         spin_lock(&ll_sb_lock);
 118         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 119         spin_unlock(&ll_sb_lock);
 120
 121         sbi->ll_flags |= LL_SBI_VERBOSE;
 122 #ifdef ENABLE_CHECKSUM
 123         sbi->ll_flags |= LL_SBI_CHECKSUM;
 124 #endif
 125
 126 #ifdef HAVE_LRU_RESIZE_SUPPORT
 127         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 128 #endif
 129
 130         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 132                                pp_r_hist.oh_lock);
 133                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 134                                pp_w_hist.oh_lock);
 135         }
 136
 137         /* metadata statahead is enabled by default */
 138         sbi->ll_sa_max = LL_SA_RPC_DEF;
 139         atomic_set(&sbi->ll_sa_total, 0);
 140         atomic_set(&sbi->ll_sa_wrong, 0);
 141         atomic_set(&sbi->ll_agl_total, 0);
 142         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 143
 144         /* root squash */
 145         sbi->ll_squash.rsi_uid = 0;
 146         sbi->ll_squash.rsi_gid = 0;
 147         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 148         init_rwsem(&sbi->ll_squash.rsi_sem);
 149
 150         RETURN(sbi);
 151 }
 152
 153 static void ll_free_sbi(struct super_block *sb)
 154 {
 155         struct ll_sb_info *sbi = ll_s2sbi(sb);
 156         ENTRY;
 157
 158         if (sbi != NULL) {
 159                 spin_lock(&ll_sb_lock);
 160                 list_del(&sbi->ll_list);
 161                 spin_unlock(&ll_sb_lock);
 162                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 163                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 170                                     struct vfsmount *mnt)
 171 {
 172         struct inode *root = 0;
 173         struct ll_sb_info *sbi = ll_s2sbi(sb);
 174         struct obd_device *obd;
 175         struct obd_capa *oc = NULL;
 176         struct obd_statfs *osfs = NULL;
 177         struct ptlrpc_request *request = NULL;
 178         struct obd_connect_data *data = NULL;
 179         struct obd_uuid *uuid;
 180         struct md_op_data *op_data;
 181         struct lustre_md lmd;
 182         obd_valid valid;
 183         int size, err, checksum;
 184         ENTRY;
 185
 186         obd = class_name2obd(md);
 187         if (!obd) {
 188                 CERROR("MD %s: not setup or attached\n", md);
 189                 RETURN(-EINVAL);
 190         }
 191
 192         OBD_ALLOC_PTR(data);
 193         if (data == NULL)
 194                 RETURN(-ENOMEM);
 195
 196         OBD_ALLOC_PTR(osfs);
 197         if (osfs == NULL) {
 198                 OBD_FREE_PTR(data);
 199                 RETURN(-ENOMEM);
 200         }
 201
 202         /* indicate the features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_ATTRFID  |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 207                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 208                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 209                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 210                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 211                                   OBD_CONNECT_EINPROGRESS |
 212                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 213                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 214                                   OBD_CONNECT_MAX_EASIZE |
 215                                   OBD_CONNECT_FLOCK_DEAD |
 216                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 217                                   OBD_CONNECT_OPEN_BY_FID |
 218                                   OBD_CONNECT_DIR_STRIPE;
 219
 220         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 221                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 222
 223 #ifdef HAVE_LRU_RESIZE_SUPPORT
 224         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 225                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 226 #endif
 227 #ifdef CONFIG_FS_POSIX_ACL
 228         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 229 #endif
 230
 231         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 232                 /* flag mdc connection as lightweight, only used for test
 233                  * purpose, use with care */
 234                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 235
 236         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 237         data->ocd_version = LUSTRE_VERSION_CODE;
 238
 239         if (sb->s_flags & MS_RDONLY)
 240                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 241         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 242                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 243
 244 #ifdef HAVE_MS_FLOCK_LOCK
 245         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 246         sb->s_flags |= MS_FLOCK_LOCK;
 247 #endif
 248 #ifdef MS_HAS_NEW_AOPS
 249         sb->s_flags |= MS_HAS_NEW_AOPS;
 250 #endif
 251
 252         if (sbi->ll_flags & LL_SBI_FLOCK)
 253                 sbi->ll_fop = &ll_file_operations_flock;
 254         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 255                 sbi->ll_fop = &ll_file_operations;
 256         else
 257                 sbi->ll_fop = &ll_file_operations_noflock;
 258
 259         /* real client */
 260         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 261         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 262                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 263
 264         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 265
 266         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 267         if (err == -EBUSY) {
 268                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 269                                    "recovery, of which this client is not a "
 270                                    "part. Please wait for recovery to complete,"
 271                                    " abort, or time out.\n", md);
 272                 GOTO(out, err);
 273         } else if (err) {
 274                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 275                 GOTO(out, err);
 276         }
 277
 278         sbi->ll_md_exp->exp_connect_data = *data;
 279
 280         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 281                            LUSTRE_SEQ_METADATA);
 282         if (err) {
 283                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 284                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 285                 GOTO(out_md, err);
 286         }
 287
 288         /* For mount, we only need fs info from MDT0, and also in DNE, it
 289          * can make sure the client can be mounted as long as MDT0 is
 290          * avaible */
 291         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 292                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 293                         OBD_STATFS_FOR_MDT0);
 294         if (err)
 295                 GOTO(out_md_fid, err);
 296
 297         /* This needs to be after statfs to ensure connect has finished.
 298          * Note that "data" does NOT contain the valid connect reply.
 299          * If connecting to a 1.8 server there will be no LMV device, so
 300          * we can access the MDC export directly and exp_connect_flags will
 301          * be non-zero, but if accessing an upgraded 2.1 server it will
 302          * have the correct flags filled in.
 303          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 304         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 305         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 306             valid != CLIENT_CONNECT_MDT_REQD) {
 307                 char *buf;
 308
 309                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 310                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 311                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 312                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 313                                    "feature(s) needed for correct operation "
 314                                    "of this client (%s). Please upgrade "
 315                                    "server or downgrade client.\n",
 316                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 317                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 318                 GOTO(out_md_fid, err = -EPROTO);
 319         }
 320
 321         size = sizeof(*data);
 322         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 323                            KEY_CONN_DATA,  &size, data, NULL);
 324         if (err) {
 325                 CERROR("%s: Get connect data failed: rc = %d\n",
 326                        sbi->ll_md_exp->exp_obd->obd_name, err);
 327                 GOTO(out_md_fid, err);
 328         }
 329
 330         LASSERT(osfs->os_bsize);
 331         sb->s_blocksize = osfs->os_bsize;
 332         sb->s_blocksize_bits = log2(osfs->os_bsize);
 333         sb->s_magic = LL_SUPER_MAGIC;
 334         sb->s_maxbytes = MAX_LFS_FILESIZE;
 335         sbi->ll_namelen = osfs->os_namelen;
 336         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 337
 338         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 339             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 340                 LCONSOLE_INFO("Disabling user_xattr feature because "
 341                               "it is not supported on the server\n");
 342                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 343         }
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 346 #ifdef MS_POSIXACL
 347                 sb->s_flags |= MS_POSIXACL;
 348 #endif
 349                 sbi->ll_flags |= LL_SBI_ACL;
 350         } else {
 351                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 352 #ifdef MS_POSIXACL
 353                 sb->s_flags &= ~MS_POSIXACL;
 354 #endif
 355                 sbi->ll_flags &= ~LL_SBI_ACL;
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 359                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 360                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 361                         LCONSOLE_INFO("client is set as remote by default.\n");
 362                 }
 363         } else {
 364                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 365                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 366                         LCONSOLE_INFO("client claims to be remote, but server "
 367                                       "rejected, forced to be local.\n");
 368                 }
 369         }
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 372                 LCONSOLE_INFO("client enabled MDS capability!\n");
 373                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 374         }
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 377                 LCONSOLE_INFO("client enabled OSS capability!\n");
 378                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 379         }
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 382                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 385                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 386         else
 387                 sbi->ll_md_brw_pages = 1;
 388
 389         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 390                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 391
 392         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 393                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 394                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 395                                       "unknown maximum xattr size.\n", dt);
 396                 } else {
 397                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 398                         sbi->ll_xattr_cache_enabled = 1;
 399                 }
 400         }
 401
 402         obd = class_name2obd(dt);
 403         if (!obd) {
 404                 CERROR("DT %s: not setup or attached\n", dt);
 405                 GOTO(out_md_fid, err = -ENODEV);
 406         }
 407
 408         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 409                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 410                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 411                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 412                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 413                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 414                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 415                                   OBD_CONNECT_MAXBYTES |
 416                                   OBD_CONNECT_EINPROGRESS |
 417                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 418                                   OBD_CONNECT_LAYOUTLOCK |
 419                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 420
 421         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 422                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 423
 424         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 425                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 426                  * disabled by default, because it can still be enabled on the
 427                  * fly via /proc. As a consequence, we still need to come to an
 428                  * agreement on the supported algorithms at connect time */
 429                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 430
 431                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 432                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 433                 else
 434                         data->ocd_cksum_types = cksum_types_supported_client();
 435         }
 436
 437 #ifdef HAVE_LRU_RESIZE_SUPPORT
 438         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 439 #endif
 440         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 441                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 442
 443         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 444                "ocd_grant: %d\n", data->ocd_connect_flags,
 445                data->ocd_version, data->ocd_grant);
 446
 447         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 448         obd->obd_upcall.onu_upcall = cl_ocd_update;
 449
 450         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 451
 452         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 453                           NULL);
 454         if (err == -EBUSY) {
 455                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 456                                    "recovery, of which this client is not a "
 457                                    "part.  Please wait for recovery to "
 458                                    "complete, abort, or time out.\n", dt);
 459                 GOTO(out_md, err);
 460         } else if (err) {
 461                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 462                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 463                 GOTO(out_md, err);
 464         }
 465
 466         sbi->ll_dt_exp->exp_connect_data = *data;
 467
 468         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 469                            LUSTRE_SEQ_METADATA);
 470         if (err) {
 471                 CERROR("%s: Can't init data layer FID infrastructure, "
 472                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 473                 GOTO(out_dt, err);
 474         }
 475
 476         mutex_lock(&sbi->ll_lco.lco_lock);
 477         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 478         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 479         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 480         mutex_unlock(&sbi->ll_lco.lco_lock);
 481
 482         fid_zero(&sbi->ll_root_fid);
 483         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 484         if (err) {
 485                 CERROR("cannot mds_connect: rc = %d\n", err);
 486                 GOTO(out_lock_cn_cb, err);
 487         }
 488         if (!fid_is_sane(&sbi->ll_root_fid)) {
 489                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 490                        sbi->ll_md_exp->exp_obd->obd_name,
 491                        PFID(&sbi->ll_root_fid));
 492                 GOTO(out_lock_cn_cb, err = -EINVAL);
 493         }
 494         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 495
 496         sb->s_op = &lustre_super_operations;
 497 #if THREAD_SIZE >= 8192 /*b=17630*/
 498         sb->s_export_op = &lustre_export_operations;
 499 #endif
 500
 501         /* make root inode
 502          * XXX: move this to after cbd setup? */
 503         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 504                 OBD_MD_FLMODEASIZE;
 505         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 506                 valid |= OBD_MD_FLRMTPERM;
 507         else if (sbi->ll_flags & LL_SBI_ACL)
 508                 valid |= OBD_MD_FLACL;
 509
 510         OBD_ALLOC_PTR(op_data);
 511         if (op_data == NULL)
 512                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 513
 514         op_data->op_fid1 = sbi->ll_root_fid;
 515         op_data->op_mode = 0;
 516         op_data->op_capa1 = oc;
 517         op_data->op_valid = valid;
 518
 519         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 520         if (oc)
 521                 capa_put(oc);
 522         OBD_FREE_PTR(op_data);
 523         if (err) {
 524                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 525                        sbi->ll_md_exp->exp_obd->obd_name, err);
 526                 GOTO(out_lock_cn_cb, err);
 527         }
 528
 529         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 530                                sbi->ll_md_exp, &lmd);
 531         if (err) {
 532                 CERROR("failed to understand root inode md: rc = %d\n", err);
 533                 ptlrpc_req_finished(request);
 534                 GOTO(out_lock_cn_cb, err);
 535         }
 536
 537         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 538         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 539                                             sbi->ll_flags & LL_SBI_32BIT_API),
 540                        &lmd);
 541         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 542         ptlrpc_req_finished(request);
 543
 544         if (IS_ERR(root)) {
 545                 if (lmd.lsm)
 546                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 547 #ifdef CONFIG_FS_POSIX_ACL
 548                 if (lmd.posix_acl) {
 549                         posix_acl_release(lmd.posix_acl);
 550                         lmd.posix_acl = NULL;
 551                 }
 552 #endif
 553                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 554                 root = NULL;
 555                 CERROR("lustre_lite: bad iget4 for root\n");
 556                 GOTO(out_root, err);
 557         }
 558
 559         err = ll_close_thread_start(&sbi->ll_lcq);
 560         if (err) {
 561                 CERROR("cannot start close thread: rc %d\n", err);
 562                 GOTO(out_root, err);
 563         }
 564
 565 #ifdef CONFIG_FS_POSIX_ACL
 566         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 567                 rct_init(&sbi->ll_rct);
 568                 et_init(&sbi->ll_et);
 569         }
 570 #endif
 571
 572         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 573         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 574                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 575                                  NULL);
 576         cl_sb_init(sb);
 577
 578         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 579                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 580                                  &sbi->ll_cache, NULL);
 581
 582         sb->s_root = d_make_root(root);
 583         if (sb->s_root == NULL) {
 584                 CERROR("%s: can't make root dentry\n",
 585                         ll_get_fsname(sb, NULL, 0));
 586                 GOTO(out_root, err = -ENOMEM);
 587         }
 588 #ifdef HAVE_DCACHE_LOCK
 589         sb->s_root->d_op = &ll_d_ops;
 590 #endif
 591
 592         sbi->ll_sdev_orig = sb->s_dev;
 593
 594         /* We set sb->s_dev equal on all lustre clients in order to support
 595          * NFS export clustering.  NFSD requires that the FSID be the same
 596          * on all clients. */
 597         /* s_dev is also used in lt_compare() to compare two fs, but that is
 598          * only a node-local comparison. */
 599         uuid = obd_get_uuid(sbi->ll_md_exp);
 600         if (uuid != NULL)
 601                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 602
 603         if (data != NULL)
 604                 OBD_FREE_PTR(data);
 605         if (osfs != NULL)
 606                 OBD_FREE_PTR(osfs);
 607         if (proc_lustre_fs_root != NULL) {
 608                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 609                                                   dt, md);
 610                 if (err < 0) {
 611                         CERROR("%s: could not register mount in lprocfs: "
 612                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 613                         err = 0;
 614                 }
 615         }
 616
 617         RETURN(err);
 618 out_root:
 619         if (root)
 620                 iput(root);
 621 out_lock_cn_cb:
 622         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 623 out_dt:
 624         obd_disconnect(sbi->ll_dt_exp);
 625         sbi->ll_dt_exp = NULL;
 626         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 627         obd_zombie_barrier();
 628 out_md_fid:
 629         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 630 out_md:
 631         obd_disconnect(sbi->ll_md_exp);
 632         sbi->ll_md_exp = NULL;
 633 out:
 634         if (data != NULL)
 635                 OBD_FREE_PTR(data);
 636         if (osfs != NULL)
 637                 OBD_FREE_PTR(osfs);
 638         return err;
 639 }
 640
 641 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 642 {
 643         int size, rc;
 644
 645         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 646         size = sizeof(int);
 647         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 648                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 649         if (rc)
 650                 CERROR("Get max mdsize error rc %d\n", rc);
 651
 652         RETURN(rc);
 653 }
 654
 655 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 656 {
 657         int size, rc;
 658
 659         size = sizeof(int);
 660         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 661                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 662         if (rc)
 663                 CERROR("Get default mdsize error rc %d\n", rc);
 664
 665         RETURN(rc);
 666 }
 667
 668 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 669 {
 670         int size, rc;
 671
 672         size = sizeof(int);
 673         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 674                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 675         if (rc)
 676                 CERROR("Get max cookiesize error rc %d\n", rc);
 677
 678         RETURN(rc);
 679 }
 680
 681 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 682 {
 683         int size, rc;
 684
 685         size = sizeof(int);
 686         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 687                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 688         if (rc)
 689                 CERROR("Get default cookiesize error rc %d\n", rc);
 690
 691         RETURN(rc);
 692 }
 693
 694 static void ll_dump_inode(struct inode *inode)
 695 {
 696         struct ll_d_hlist_node *tmp;
 697         int dentry_count = 0;
 698
 699         LASSERT(inode != NULL);
 700
 701         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 702                 dentry_count++;
 703
 704         CERROR("%s: inode %p dump: dev=%s fid="DFID
 705                " mode=%o count=%u, %d dentries\n",
 706                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 707                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 708                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 709 }
 710
 711 void lustre_dump_dentry(struct dentry *dentry, int recur)
 712 {
 713         struct list_head *tmp;
 714         int subdirs = 0;
 715
 716         LASSERT(dentry != NULL);
 717
 718         list_for_each(tmp, &dentry->d_subdirs)
 719                 subdirs++;
 720
 721         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 722                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 723                dentry->d_name.len, dentry->d_name.name,
 724                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 725                dentry->d_parent, dentry->d_inode, d_count(dentry),
 726                dentry->d_flags, dentry->d_fsdata, subdirs);
 727         if (dentry->d_inode != NULL)
 728                 ll_dump_inode(dentry->d_inode);
 729
 730         if (recur == 0)
 731                 return;
 732
 733         list_for_each(tmp, &dentry->d_subdirs) {
 734                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 735                 lustre_dump_dentry(d, recur - 1);
 736         }
 737 }
 738
 739 static void client_common_put_super(struct super_block *sb)
 740 {
 741         struct ll_sb_info *sbi = ll_s2sbi(sb);
 742         ENTRY;
 743
 744 #ifdef CONFIG_FS_POSIX_ACL
 745         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 746                 et_fini(&sbi->ll_et);
 747                 rct_fini(&sbi->ll_rct);
 748         }
 749 #endif
 750
 751         ll_close_thread_shutdown(sbi->ll_lcq);
 752
 753         cl_sb_fini(sb);
 754
 755         list_del(&sbi->ll_conn_chain);
 756
 757         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 758         obd_disconnect(sbi->ll_dt_exp);
 759         sbi->ll_dt_exp = NULL;
 760         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 761          * see LU-2543. */
 762         obd_zombie_barrier();
 763
 764         lprocfs_unregister_mountpoint(sbi);
 765
 766         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 767         obd_disconnect(sbi->ll_md_exp);
 768         sbi->ll_md_exp = NULL;
 769
 770         EXIT;
 771 }
 772
 773 void ll_kill_super(struct super_block *sb)
 774 {
 775         struct ll_sb_info *sbi;
 776
 777         ENTRY;
 778
 779         /* not init sb ?*/
 780         if (!(sb->s_flags & MS_ACTIVE))
 781                 return;
 782
 783         sbi = ll_s2sbi(sb);
 784         /* we need restore s_dev from changed for clustred NFS before put_super
 785          * because new kernels have cached s_dev and change sb->s_dev in
 786          * put_super not affected real removing devices */
 787         if (sbi) {
 788                 sb->s_dev = sbi->ll_sdev_orig;
 789                 sbi->ll_umounting = 1;
 790         }
 791         EXIT;
 792 }
 793
 794 static inline int ll_set_opt(const char *opt, char *data, int fl)
 795 {
 796         if (strncmp(opt, data, strlen(opt)) != 0)
 797                 return(0);
 798         else
 799                 return(fl);
 800 }
 801
 802 /* non-client-specific mount options are parsed in lmd_parse */
 803 static int ll_options(char *options, int *flags)
 804 {
 805         int tmp;
 806         char *s1 = options, *s2;
 807         ENTRY;
 808
 809         if (!options)
 810                 RETURN(0);
 811
 812         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 813
 814         while (*s1) {
 815                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 816                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 817                 if (tmp) {
 818                         *flags |= tmp;
 819                         goto next;
 820                 }
 821                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 822                 if (tmp) {
 823                         *flags |= tmp;
 824                         goto next;
 825                 }
 826                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 832                 if (tmp) {
 833                         *flags &= ~tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 842                 if (tmp) {
 843                         *flags &= ~tmp;
 844                         goto next;
 845                 }
 846 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
 847                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 848                 if (tmp) {
 849                         /* Ignore deprecated mount option.  The client will
 850                          * always try to mount with ACL support, whether this
 851                          * is used depends on whether server supports it. */
 852                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 853                                                   "mount option 'acl'.\n");
 854                         goto next;
 855                 }
 856                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 857                 if (tmp) {
 858                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 859                                                   "mount option 'noacl'.\n");
 860                         goto next;
 861                 }
 862 #else
 863 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 864 #endif
 865                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 871                 if (tmp) {
 872                         *flags |= tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 876                 if (tmp) {
 877                         *flags &= ~tmp;
 878                         goto next;
 879                 }
 880
 881                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 882                 if (tmp) {
 883                         *flags |= tmp;
 884                         goto next;
 885                 }
 886                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 887                 if (tmp) {
 888                         *flags &= ~tmp;
 889                         goto next;
 890                 }
 891                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 892                 if (tmp) {
 893                         *flags |= tmp;
 894                         goto next;
 895                 }
 896                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 897                 if (tmp) {
 898                         *flags &= ~tmp;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 902                 if (tmp) {
 903                         *flags |= tmp;
 904                         goto next;
 905                 }
 906                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 907                 if (tmp) {
 908                         *flags &= ~tmp;
 909                         goto next;
 910                 }
 911                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 912                 if (tmp) {
 913                         *flags |= tmp;
 914                         goto next;
 915                 }
 916                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 917                 if (tmp) {
 918                         *flags |= tmp;
 919                         goto next;
 920                 }
 921                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 922                 if (tmp) {
 923                         *flags |= tmp;
 924                         goto next;
 925                 }
 926                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 927                 if (tmp) {
 928                         *flags &= ~tmp;
 929                         goto next;
 930                 }
 931                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 932                                    s1);
 933                 RETURN(-EINVAL);
 934
 935 next:
 936                 /* Find next opt */
 937                 s2 = strchr(s1, ',');
 938                 if (s2 == NULL)
 939                         break;
 940                 s1 = s2 + 1;
 941         }
 942         RETURN(0);
 943 }
 944
 945 void ll_lli_init(struct ll_inode_info *lli)
 946 {
 947         lli->lli_inode_magic = LLI_INODE_MAGIC;
 948         lli->lli_flags = 0;
 949         lli->lli_ioepoch = 0;
 950         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 951         spin_lock_init(&lli->lli_lock);
 952         lli->lli_posix_acl = NULL;
 953         lli->lli_remote_perms = NULL;
 954         mutex_init(&lli->lli_rmtperm_mutex);
 955         /* Do not set lli_fid, it has been initialized already. */
 956         fid_zero(&lli->lli_pfid);
 957         INIT_LIST_HEAD(&lli->lli_close_list);
 958         INIT_LIST_HEAD(&lli->lli_oss_capas);
 959         atomic_set(&lli->lli_open_count, 0);
 960         lli->lli_mds_capa = NULL;
 961         lli->lli_rmtperm_time = 0;
 962         lli->lli_pending_och = NULL;
 963         lli->lli_mds_read_och = NULL;
 964         lli->lli_mds_write_och = NULL;
 965         lli->lli_mds_exec_och = NULL;
 966         lli->lli_open_fd_read_count = 0;
 967         lli->lli_open_fd_write_count = 0;
 968         lli->lli_open_fd_exec_count = 0;
 969         mutex_init(&lli->lli_och_mutex);
 970         spin_lock_init(&lli->lli_agl_lock);
 971         lli->lli_has_smd = false;
 972         spin_lock_init(&lli->lli_layout_lock);
 973         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 974         lli->lli_clob = NULL;
 975
 976         init_rwsem(&lli->lli_xattrs_list_rwsem);
 977         mutex_init(&lli->lli_xattrs_enq_lock);
 978
 979         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 980         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 981                 mutex_init(&lli->lli_readdir_mutex);
 982                 lli->lli_opendir_key = NULL;
 983                 lli->lli_sai = NULL;
 984                 spin_lock_init(&lli->lli_sa_lock);
 985                 lli->lli_opendir_pid = 0;
 986         } else {
 987                 mutex_init(&lli->lli_size_mutex);
 988                 lli->lli_symlink_name = NULL;
 989                 init_rwsem(&lli->lli_trunc_sem);
 990                 mutex_init(&lli->lli_write_mutex);
 991                 init_rwsem(&lli->lli_glimpse_sem);
 992                 lli->lli_glimpse_time = 0;
 993                 INIT_LIST_HEAD(&lli->lli_agl_list);
 994                 lli->lli_agl_index = 0;
 995                 lli->lli_async_rc = 0;
 996         }
 997         mutex_init(&lli->lli_layout_mutex);
 998 }
 999
1000 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1001 {
1002         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1003
1004         bdi->name = "lustre";
1005         return bdi_register(bdi, NULL, "lustre-%d",
1006                             atomic_inc_return(&ll_bdi_num));
1007 }
1008
1009 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1010 {
1011         struct lustre_profile *lprof = NULL;
1012         struct lustre_sb_info *lsi = s2lsi(sb);
1013         struct ll_sb_info *sbi;
1014         char  *dt = NULL, *md = NULL;
1015         char  *profilenm = get_profile_name(sb);
1016         struct config_llog_instance *cfg;
1017         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1018         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1019         int    err;
1020         ENTRY;
1021
1022         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1023
1024         OBD_ALLOC_PTR(cfg);
1025         if (cfg == NULL)
1026                 RETURN(-ENOMEM);
1027
1028         try_module_get(THIS_MODULE);
1029
1030         /* client additional sb info */
1031         lsi->lsi_llsbi = sbi = ll_init_sbi();
1032         if (!sbi) {
1033                 module_put(THIS_MODULE);
1034                 OBD_FREE_PTR(cfg);
1035                 RETURN(-ENOMEM);
1036         }
1037
1038         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1039         if (err)
1040                 GOTO(out_free, err);
1041
1042         err = bdi_init(&lsi->lsi_bdi);
1043         if (err)
1044                 GOTO(out_free, err);
1045         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1046         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1047         err = ll_bdi_register(&lsi->lsi_bdi);
1048         if (err)
1049                 GOTO(out_free, err);
1050
1051         sb->s_bdi = &lsi->lsi_bdi;
1052 #ifndef HAVE_DCACHE_LOCK
1053         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1054         sb->s_d_op = &ll_d_ops;
1055 #endif
1056
1057         /* Generate a string unique to this super, in case some joker tries
1058            to mount the same fs at two mount points.
1059            Use the address of the super itself.*/
1060         cfg->cfg_instance = sb;
1061         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1062         cfg->cfg_callback = class_config_llog_handler;
1063         /* set up client obds */
1064         err = lustre_process_log(sb, profilenm, cfg);
1065         if (err < 0)
1066                 GOTO(out_free, err);
1067
1068         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1069         lprof = class_get_profile(profilenm);
1070         if (lprof == NULL) {
1071                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1072                                    " read from the MGS.  Does that filesystem "
1073                                    "exist?\n", profilenm);
1074                 GOTO(out_free, err = -EINVAL);
1075         }
1076         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1077                lprof->lp_md, lprof->lp_dt);
1078
1079         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1080         if (!dt)
1081                 GOTO(out_free, err = -ENOMEM);
1082         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1083
1084         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1085         if (!md)
1086                 GOTO(out_free, err = -ENOMEM);
1087         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1088
1089         /* connections, registrations, sb setup */
1090         err = client_common_fill_super(sb, md, dt, mnt);
1091
1092 out_free:
1093         if (md)
1094                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1095         if (dt)
1096                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1097         if (err)
1098                 ll_put_super(sb);
1099         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1100                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1101
1102         OBD_FREE_PTR(cfg);
1103         RETURN(err);
1104 } /* ll_fill_super */
1105
1106 void ll_put_super(struct super_block *sb)
1107 {
1108         struct config_llog_instance cfg, params_cfg;
1109         struct obd_device *obd;
1110         struct lustre_sb_info *lsi = s2lsi(sb);
1111         struct ll_sb_info *sbi = ll_s2sbi(sb);
1112         char *profilenm = get_profile_name(sb);
1113         int ccc_count, next, force = 1, rc = 0;
1114         ENTRY;
1115
1116         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1117
1118         ll_print_capa_stat(sbi);
1119
1120         cfg.cfg_instance = sb;
1121         lustre_end_log(sb, profilenm, &cfg);
1122
1123         params_cfg.cfg_instance = sb;
1124         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1125
1126         if (sbi->ll_md_exp) {
1127                 obd = class_exp2obd(sbi->ll_md_exp);
1128                 if (obd)
1129                         force = obd->obd_force;
1130         }
1131
1132         /* Wait for unstable pages to be committed to stable storage */
1133         if (force == 0) {
1134                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1135                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1136                         atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1137                         &lwi);
1138         }
1139
1140         ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1141         if (force == 0 && rc != -EINTR)
1142                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1143
1144
1145         /* We need to set force before the lov_disconnect in
1146            lustre_common_put_super, since l_d cleans up osc's as well. */
1147         if (force) {
1148                 next = 0;
1149                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1150                                                      &next)) != NULL) {
1151                         obd->obd_force = force;
1152                 }
1153         }
1154
1155         if (sbi->ll_lcq) {
1156                 /* Only if client_common_fill_super succeeded */
1157                 client_common_put_super(sb);
1158         }
1159
1160         next = 0;
1161         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1162                 class_manual_cleanup(obd);
1163         }
1164
1165         if (sbi->ll_flags & LL_SBI_VERBOSE)
1166                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1167
1168         if (profilenm)
1169                 class_del_profile(profilenm);
1170
1171         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1172                 bdi_destroy(&lsi->lsi_bdi);
1173                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1174         }
1175
1176         ll_free_sbi(sb);
1177         lsi->lsi_llsbi = NULL;
1178
1179         lustre_common_put_super(sb);
1180
1181         cl_env_cache_purge(~0);
1182
1183         module_put(THIS_MODULE);
1184
1185         EXIT;
1186 } /* client_put_super */
1187
1188 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1189 {
1190         struct inode *inode = NULL;
1191
1192         /* NOTE: we depend on atomic igrab() -bzzz */
1193         lock_res_and_lock(lock);
1194         if (lock->l_resource->lr_lvb_inode) {
1195                 struct ll_inode_info * lli;
1196                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1197                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1198                         inode = igrab(lock->l_resource->lr_lvb_inode);
1199                 } else {
1200                         inode = lock->l_resource->lr_lvb_inode;
1201                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1202                                          D_WARNING, lock, "lr_lvb_inode %p is "
1203                                          "bogus: magic %08x",
1204                                          lock->l_resource->lr_lvb_inode,
1205                                          lli->lli_inode_magic);
1206                         inode = NULL;
1207                 }
1208         }
1209         unlock_res_and_lock(lock);
1210         return inode;
1211 }
1212
1213 static void ll_dir_clear_lsm_md(struct inode *inode)
1214 {
1215         struct ll_inode_info *lli = ll_i2info(inode);
1216
1217         LASSERT(S_ISDIR(inode->i_mode));
1218
1219         if (lli->lli_lsm_md != NULL) {
1220                 lmv_free_memmd(lli->lli_lsm_md);
1221                 lli->lli_lsm_md = NULL;
1222         }
1223 }
1224
1225 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1226                                       const struct lu_fid *fid,
1227                                       struct lustre_md *md)
1228 {
1229         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1230         struct mdt_body         *body = md->body;
1231         struct inode            *inode;
1232         ino_t                   ino;
1233         ENTRY;
1234
1235         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1236         inode = iget_locked(sb, ino);
1237         if (inode == NULL) {
1238                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1239                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1240                 RETURN(ERR_PTR(-ENOENT));
1241         }
1242
1243         if (inode->i_state & I_NEW) {
1244                 struct ll_inode_info *lli = ll_i2info(inode);
1245                 struct lmv_stripe_md *lsm = md->lmv;
1246
1247                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1248                                 (body->mbo_mode & S_IFMT);
1249                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1250                          PFID(fid));
1251
1252                 LTIME_S(inode->i_mtime) = 0;
1253                 LTIME_S(inode->i_atime) = 0;
1254                 LTIME_S(inode->i_ctime) = 0;
1255                 inode->i_rdev = 0;
1256
1257                 /* initializing backing dev info. */
1258                 inode->i_mapping->backing_dev_info =
1259                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1260                 inode->i_op = &ll_dir_inode_operations;
1261                 inode->i_fop = &ll_dir_operations;
1262                 lli->lli_fid = *fid;
1263                 ll_lli_init(lli);
1264
1265                 LASSERT(lsm != NULL);
1266                 /* master object FID */
1267                 lli->lli_pfid = body->mbo_fid1;
1268                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1269                        lli, PFID(fid), PFID(&lli->lli_pfid));
1270                 unlock_new_inode(inode);
1271         }
1272
1273         RETURN(inode);
1274 }
1275
1276 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1277 {
1278         struct lu_fid *fid;
1279         struct lmv_stripe_md *lsm = md->lmv;
1280         int i;
1281
1282         LASSERT(lsm != NULL);
1283         /* XXX sigh, this lsm_root initialization should be in
1284          * LMV layer, but it needs ll_iget right now, so we
1285          * put this here right now. */
1286         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1287                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1288                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1289                 /* Unfortunately ll_iget will call ll_update_inode,
1290                  * where the initialization of slave inode is slightly
1291                  * different, so it reset lsm_md to NULL to avoid
1292                  * initializing lsm for slave inode. */
1293                 /* For migrating inode, master stripe and master object will
1294                  * be same, so we only need assign this inode */
1295                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1296                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1297                 else
1298                         lsm->lsm_md_oinfo[i].lmo_root =
1299                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1300
1301                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1302                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1303
1304                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1305                         return rc;
1306                 }
1307         }
1308
1309         /* Here is where the lsm is being initialized(fill lmo_info) after
1310          * client retrieve MD stripe information from MDT. */
1311         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1312                                 ll_md_blocking_ast);
1313 }
1314
1315 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1316                                 const struct lmv_stripe_md *lsm_md2)
1317 {
1318         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1319                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1320                lsm_md1->lsm_md_master_mdt_index ==
1321                                         lsm_md2->lsm_md_master_mdt_index &&
1322                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1323                lsm_md1->lsm_md_layout_version ==
1324                                         lsm_md2->lsm_md_layout_version &&
1325                strcmp(lsm_md1->lsm_md_pool_name,
1326                       lsm_md2->lsm_md_pool_name) == 0;
1327 }
1328
1329 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1330 {
1331         struct ll_inode_info *lli = ll_i2info(inode);
1332         struct lmv_stripe_md *lsm = md->lmv;
1333         int     rc;
1334         ENTRY;
1335
1336         LASSERT(S_ISDIR(inode->i_mode));
1337         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1338                PFID(ll_inode2fid(inode)));
1339
1340         /* no striped information from request. */
1341         if (lsm == NULL) {
1342                 if (lli->lli_lsm_md == NULL) {
1343                         RETURN(0);
1344                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1345                                                 LMV_HASH_FLAG_MIGRATION) {
1346                         /* migration is done, the temporay MIGRATE layout has
1347                          * been removed */
1348                         CDEBUG(D_INODE, DFID" finish migration.\n",
1349                                PFID(ll_inode2fid(inode)));
1350                         lmv_free_memmd(lli->lli_lsm_md);
1351                         lli->lli_lsm_md = NULL;
1352                         RETURN(0);
1353                 } else {
1354                         /* The lustre_md from req does not include stripeEA,
1355                          * see ll_md_setattr */
1356                         RETURN(0);
1357                 }
1358         }
1359
1360         /* set the directory layout */
1361         if (lli->lli_lsm_md == NULL) {
1362
1363                 rc = ll_init_lsm_md(inode, md);
1364                 if (rc != 0)
1365                         RETURN(rc);
1366
1367                 lli->lli_lsm_md = lsm;
1368                 /* set lsm_md to NULL, so the following free lustre_md
1369                  * will not free this lsm */
1370                 md->lmv = NULL;
1371                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1372                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1373                 RETURN(0);
1374         }
1375
1376         /* Compare the old and new stripe information */
1377         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1378                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1379                 int                     idx;
1380
1381                 CERROR("%s: lmv layout mismatch "DFID"(%p)/"DFID"(%p)"
1382                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1383                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1384                        ll_get_fsname(inode->i_sb, NULL, 0),
1385                        PFID(&lsm->lsm_md_master_fid), lsm,
1386                        PFID(&old_lsm->lsm_md_master_fid), old_lsm,
1387                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1388                        lsm->lsm_md_stripe_count,
1389                        old_lsm->lsm_md_stripe_count,
1390                        lsm->lsm_md_master_mdt_index,
1391                        old_lsm->lsm_md_master_mdt_index,
1392                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1393                        lsm->lsm_md_layout_version,
1394                        old_lsm->lsm_md_layout_version,
1395                        lsm->lsm_md_pool_name,
1396                        old_lsm->lsm_md_pool_name);
1397
1398                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1399                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1400                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1401                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1402                 }
1403
1404                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1405                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1406                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1407                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1408                 }
1409
1410                 RETURN(-EIO);
1411         }
1412
1413         rc = md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1414                               md->body, ll_md_blocking_ast);
1415
1416         RETURN(rc);
1417 }
1418
1419 void ll_clear_inode(struct inode *inode)
1420 {
1421         struct ll_inode_info *lli = ll_i2info(inode);
1422         struct ll_sb_info *sbi = ll_i2sbi(inode);
1423         ENTRY;
1424
1425         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1426                PFID(ll_inode2fid(inode)), inode);
1427
1428         if (S_ISDIR(inode->i_mode)) {
1429                 /* these should have been cleared in ll_file_release */
1430                 LASSERT(lli->lli_opendir_key == NULL);
1431                 LASSERT(lli->lli_sai == NULL);
1432                 LASSERT(lli->lli_opendir_pid == 0);
1433         }
1434
1435         spin_lock(&lli->lli_lock);
1436         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1437         spin_unlock(&lli->lli_lock);
1438         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1439
1440         LASSERT(!lli->lli_open_fd_write_count);
1441         LASSERT(!lli->lli_open_fd_read_count);
1442         LASSERT(!lli->lli_open_fd_exec_count);
1443
1444         if (lli->lli_mds_write_och)
1445                 ll_md_real_close(inode, FMODE_WRITE);
1446         if (lli->lli_mds_exec_och)
1447                 ll_md_real_close(inode, FMODE_EXEC);
1448         if (lli->lli_mds_read_och)
1449                 ll_md_real_close(inode, FMODE_READ);
1450
1451         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1452                 OBD_FREE(lli->lli_symlink_name,
1453                          strlen(lli->lli_symlink_name) + 1);
1454                 lli->lli_symlink_name = NULL;
1455         }
1456
1457         ll_xattr_cache_destroy(inode);
1458
1459         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1460                 LASSERT(lli->lli_posix_acl == NULL);
1461                 if (lli->lli_remote_perms) {
1462                         free_rmtperm_hash(lli->lli_remote_perms);
1463                         lli->lli_remote_perms = NULL;
1464                 }
1465         }
1466 #ifdef CONFIG_FS_POSIX_ACL
1467         else if (lli->lli_posix_acl) {
1468                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1469                 LASSERT(lli->lli_remote_perms == NULL);
1470                 posix_acl_release(lli->lli_posix_acl);
1471                 lli->lli_posix_acl = NULL;
1472         }
1473 #endif
1474         lli->lli_inode_magic = LLI_INODE_DEAD;
1475
1476         ll_clear_inode_capas(inode);
1477         if (S_ISDIR(inode->i_mode))
1478                 ll_dir_clear_lsm_md(inode);
1479         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1480                 LASSERT(list_empty(&lli->lli_agl_list));
1481
1482         /*
1483          * XXX This has to be done before lsm is freed below, because
1484          * cl_object still uses inode lsm.
1485          */
1486         cl_inode_fini(inode);
1487         lli->lli_has_smd = false;
1488
1489         EXIT;
1490 }
1491
1492 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1493                   struct md_open_data **mod)
1494 {
1495         struct lustre_md md;
1496         struct inode *inode = dentry->d_inode;
1497         struct ll_sb_info *sbi = ll_i2sbi(inode);
1498         struct ptlrpc_request *request = NULL;
1499         int rc, ia_valid;
1500         ENTRY;
1501
1502         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1503                                      LUSTRE_OPC_ANY, NULL);
1504         if (IS_ERR(op_data))
1505                 RETURN(PTR_ERR(op_data));
1506
1507         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1508                         &request, mod);
1509         if (rc) {
1510                 ptlrpc_req_finished(request);
1511                 if (rc == -ENOENT) {
1512                         clear_nlink(inode);
1513                         /* Unlinked special device node? Or just a race?
1514                          * Pretend we done everything. */
1515                         if (!S_ISREG(inode->i_mode) &&
1516                             !S_ISDIR(inode->i_mode)) {
1517                                 ia_valid = op_data->op_attr.ia_valid;
1518                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1519                                 rc = simple_setattr(dentry, &op_data->op_attr);
1520                                 op_data->op_attr.ia_valid = ia_valid;
1521                         }
1522                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1523                         CERROR("md_setattr fails: rc = %d\n", rc);
1524                 }
1525                 RETURN(rc);
1526         }
1527
1528         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1529                               sbi->ll_md_exp, &md);
1530         if (rc) {
1531                 ptlrpc_req_finished(request);
1532                 RETURN(rc);
1533         }
1534
1535         ia_valid = op_data->op_attr.ia_valid;
1536         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1537          * cache is not cleared yet. */
1538         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1539         rc = simple_setattr(dentry, &op_data->op_attr);
1540         op_data->op_attr.ia_valid = ia_valid;
1541
1542         /* Extract epoch data if obtained. */
1543         op_data->op_handle = md.body->mbo_handle;
1544         op_data->op_ioepoch = md.body->mbo_ioepoch;
1545
1546         rc = ll_update_inode(inode, &md);
1547         ptlrpc_req_finished(request);
1548
1549         RETURN(rc);
1550 }
1551
1552 /* Close IO epoch and send Size-on-MDS attribute update. */
1553 static int ll_setattr_done_writing(struct inode *inode,
1554                                    struct md_op_data *op_data,
1555                                    struct md_open_data *mod)
1556 {
1557         struct ll_inode_info *lli = ll_i2info(inode);
1558         int rc = 0;
1559         ENTRY;
1560
1561         LASSERT(op_data != NULL);
1562         if (!S_ISREG(inode->i_mode))
1563                 RETURN(0);
1564
1565         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1566                op_data->op_ioepoch, PFID(&lli->lli_fid));
1567
1568         op_data->op_flags = MF_EPOCH_CLOSE;
1569         ll_done_writing_attr(inode, op_data);
1570         ll_pack_inode2opdata(inode, op_data, NULL);
1571
1572         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1573         if (rc == -EAGAIN) {
1574                 /* MDS has instructed us to obtain Size-on-MDS attribute
1575                  * from OSTs and send setattr to back to MDS. */
1576                 rc = ll_som_update(inode, op_data);
1577         } else if (rc) {
1578                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1579                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1580                        PFID(ll_inode2fid(inode)), rc);
1581         }
1582         RETURN(rc);
1583 }
1584
1585 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1586 {
1587         struct obd_capa *capa;
1588         int rc;
1589
1590         if (attr->ia_valid & ATTR_SIZE)
1591                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1592         else
1593                 capa = ll_mdscapa_get(inode);
1594
1595         rc = cl_setattr_ost(inode, attr, capa);
1596
1597         if (attr->ia_valid & ATTR_SIZE)
1598                 ll_truncate_free_capa(capa);
1599         else
1600                 capa_put(capa);
1601
1602         return rc;
1603 }
1604
1605 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1606  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1607  * keep these values until such a time that objects are allocated for it.
1608  * We do the MDS operations first, as it is checking permissions for us.
1609  * We don't to the MDS RPC if there is nothing that we want to store there,
1610  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1611  * going to do an RPC anyways.
1612  *
1613  * If we are doing a truncate, we will send the mtime and ctime updates
1614  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1615  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1616  * at the same time.
1617  *
1618  * In case of HSMimport, we only set attr on MDS.
1619  */
1620 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1621 {
1622         struct inode *inode = dentry->d_inode;
1623         struct ll_inode_info *lli = ll_i2info(inode);
1624         struct md_op_data *op_data = NULL;
1625         struct md_open_data *mod = NULL;
1626         bool file_is_released = false;
1627         int rc = 0, rc1 = 0;
1628         ENTRY;
1629
1630         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1631                "valid %x, hsm_import %d\n",
1632                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1633                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1634                hsm_import);
1635
1636         if (attr->ia_valid & ATTR_SIZE) {
1637                 /* Check new size against VFS/VM file size limit and rlimit */
1638                 rc = inode_newsize_ok(inode, attr->ia_size);
1639                 if (rc)
1640                         RETURN(rc);
1641
1642                 /* The maximum Lustre file size is variable, based on the
1643                  * OST maximum object size and number of stripes.  This
1644                  * needs another check in addition to the VFS check above. */
1645                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1646                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1647                                PFID(&lli->lli_fid), attr->ia_size,
1648                                ll_file_maxbytes(inode));
1649                         RETURN(-EFBIG);
1650                 }
1651
1652                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1653         }
1654
1655         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1656         if (attr->ia_valid & TIMES_SET_FLAGS) {
1657                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1658                     !cfs_capable(CFS_CAP_FOWNER))
1659                         RETURN(-EPERM);
1660         }
1661
1662         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1663         if (attr->ia_valid & ATTR_CTIME) {
1664                 attr->ia_ctime = CFS_CURRENT_TIME;
1665                 attr->ia_valid |= ATTR_CTIME_SET;
1666         }
1667         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1668             (attr->ia_valid & ATTR_ATIME)) {
1669                 attr->ia_atime = CFS_CURRENT_TIME;
1670                 attr->ia_valid |= ATTR_ATIME_SET;
1671         }
1672         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1673             (attr->ia_valid & ATTR_MTIME)) {
1674                 attr->ia_mtime = CFS_CURRENT_TIME;
1675                 attr->ia_valid |= ATTR_MTIME_SET;
1676         }
1677
1678         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1679                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1680                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1681                        cfs_time_current_sec());
1682
1683         /* We always do an MDS RPC, even if we're only changing the size;
1684          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1685
1686         OBD_ALLOC_PTR(op_data);
1687         if (op_data == NULL)
1688                 RETURN(-ENOMEM);
1689
1690         if (!S_ISDIR(inode->i_mode)) {
1691                 if (attr->ia_valid & ATTR_SIZE)
1692                         inode_dio_write_done(inode);
1693                 mutex_unlock(&inode->i_mutex);
1694         }
1695
1696         /* truncate on a released file must failed with -ENODATA,
1697          * so size must not be set on MDS for released file
1698          * but other attributes must be set
1699          */
1700         if (S_ISREG(inode->i_mode)) {
1701                 struct lov_stripe_md *lsm;
1702                 __u32 gen;
1703
1704                 ll_layout_refresh(inode, &gen);
1705                 lsm = ccc_inode_lsm_get(inode);
1706                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1707                         file_is_released = true;
1708                 ccc_inode_lsm_put(inode, lsm);
1709
1710                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1711                         if (file_is_released) {
1712                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1713                                 if (rc < 0)
1714                                         GOTO(out, rc);
1715
1716                                 file_is_released = false;
1717                                 ll_layout_refresh(inode, &gen);
1718                         }
1719
1720                         /* If we are changing file size, file content is
1721                          * modified, flag it. */
1722                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1723                         spin_lock(&lli->lli_lock);
1724                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1725                         spin_unlock(&lli->lli_lock);
1726                         op_data->op_bias |= MDS_DATA_MODIFIED;
1727                 }
1728         }
1729
1730         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1731
1732         /* Open epoch for truncate. */
1733         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1734             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1735                 op_data->op_flags = MF_EPOCH_OPEN;
1736
1737         rc = ll_md_setattr(dentry, op_data, &mod);
1738         if (rc)
1739                 GOTO(out, rc);
1740
1741         /* RPC to MDT is sent, cancel data modification flag */
1742         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1743                 spin_lock(&lli->lli_lock);
1744                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1745                 spin_unlock(&lli->lli_lock);
1746         }
1747
1748         ll_ioepoch_open(lli, op_data->op_ioepoch);
1749         if (!S_ISREG(inode->i_mode) || file_is_released)
1750                 GOTO(out, rc = 0);
1751
1752         if (attr->ia_valid & (ATTR_SIZE |
1753                               ATTR_ATIME | ATTR_ATIME_SET |
1754                               ATTR_MTIME | ATTR_MTIME_SET)) {
1755                 /* For truncate and utimes sending attributes to OSTs, setting
1756                  * mtime/atime to the past will be performed under PW [0:EOF]
1757                  * extent lock (new_size:EOF for truncate).  It may seem
1758                  * excessive to send mtime/atime updates to OSTs when not
1759                  * setting times to past, but it is necessary due to possible
1760                  * time de-synchronization between MDT inode and OST objects */
1761                 if (attr->ia_valid & ATTR_SIZE)
1762                         down_write(&lli->lli_trunc_sem);
1763                 rc = ll_setattr_ost(inode, attr);
1764                 if (attr->ia_valid & ATTR_SIZE)
1765                         up_write(&lli->lli_trunc_sem);
1766         }
1767         EXIT;
1768 out:
1769         if (op_data) {
1770                 if (op_data->op_ioepoch) {
1771                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1772                         if (!rc)
1773                                 rc = rc1;
1774                 }
1775                 ll_finish_md_op_data(op_data);
1776         }
1777         if (!S_ISDIR(inode->i_mode)) {
1778                 mutex_lock(&inode->i_mutex);
1779                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1780                         inode_dio_wait(inode);
1781         }
1782
1783         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1784                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1785
1786         return rc;
1787 }
1788
1789 int ll_setattr(struct dentry *de, struct iattr *attr)
1790 {
1791         int mode = de->d_inode->i_mode;
1792
1793         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1794                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1795                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1796
1797         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1798                                (ATTR_SIZE|ATTR_MODE)) &&
1799             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1800              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1801               !(attr->ia_mode & S_ISGID))))
1802                 attr->ia_valid |= ATTR_FORCE;
1803
1804         if ((attr->ia_valid & ATTR_MODE) &&
1805             (mode & S_ISUID) &&
1806             !(attr->ia_mode & S_ISUID) &&
1807             !(attr->ia_valid & ATTR_KILL_SUID))
1808                 attr->ia_valid |= ATTR_KILL_SUID;
1809
1810         if ((attr->ia_valid & ATTR_MODE) &&
1811             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1812             !(attr->ia_mode & S_ISGID) &&
1813             !(attr->ia_valid & ATTR_KILL_SGID))
1814                 attr->ia_valid |= ATTR_KILL_SGID;
1815
1816         return ll_setattr_raw(de, attr, false);
1817 }
1818
1819 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1820                        __u64 max_age, __u32 flags)
1821 {
1822         struct ll_sb_info *sbi = ll_s2sbi(sb);
1823         struct obd_statfs obd_osfs;
1824         int rc;
1825         ENTRY;
1826
1827         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1828         if (rc) {
1829                 CERROR("md_statfs fails: rc = %d\n", rc);
1830                 RETURN(rc);
1831         }
1832
1833         osfs->os_type = sb->s_magic;
1834
1835         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1836                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1837
1838         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1839                 flags |= OBD_STATFS_NODELAY;
1840
1841         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1842         if (rc) {
1843                 CERROR("obd_statfs fails: rc = %d\n", rc);
1844                 RETURN(rc);
1845         }
1846
1847         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1848                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1849                obd_osfs.os_files);
1850
1851         osfs->os_bsize = obd_osfs.os_bsize;
1852         osfs->os_blocks = obd_osfs.os_blocks;
1853         osfs->os_bfree = obd_osfs.os_bfree;
1854         osfs->os_bavail = obd_osfs.os_bavail;
1855
1856         /* If we don't have as many objects free on the OST as inodes
1857          * on the MDS, we reduce the total number of inodes to
1858          * compensate, so that the "inodes in use" number is correct.
1859          */
1860         if (obd_osfs.os_ffree < osfs->os_ffree) {
1861                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1862                         obd_osfs.os_ffree;
1863                 osfs->os_ffree = obd_osfs.os_ffree;
1864         }
1865
1866         RETURN(rc);
1867 }
1868 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1869 {
1870         struct super_block *sb = de->d_sb;
1871         struct obd_statfs osfs;
1872         __u64 fsid = huge_encode_dev(sb->s_dev);
1873         int rc;
1874
1875         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1876         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1877
1878         /* Some amount of caching on the client is allowed */
1879         rc = ll_statfs_internal(sb, &osfs,
1880                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1881                                 0);
1882         if (rc)
1883                 return rc;
1884
1885         statfs_unpack(sfs, &osfs);
1886
1887         /* We need to downshift for all 32-bit kernels, because we can't
1888          * tell if the kernel is being called via sys_statfs64() or not.
1889          * Stop before overflowing f_bsize - in which case it is better
1890          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1891         if (sizeof(long) < 8) {
1892                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1893                         sfs->f_bsize <<= 1;
1894
1895                         osfs.os_blocks >>= 1;
1896                         osfs.os_bfree >>= 1;
1897                         osfs.os_bavail >>= 1;
1898                 }
1899         }
1900
1901         sfs->f_blocks = osfs.os_blocks;
1902         sfs->f_bfree = osfs.os_bfree;
1903         sfs->f_bavail = osfs.os_bavail;
1904         sfs->f_fsid.val[0] = (__u32)fsid;
1905         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1906         return 0;
1907 }
1908
1909 void ll_inode_size_lock(struct inode *inode)
1910 {
1911         struct ll_inode_info *lli;
1912
1913         LASSERT(!S_ISDIR(inode->i_mode));
1914
1915         lli = ll_i2info(inode);
1916         mutex_lock(&lli->lli_size_mutex);
1917 }
1918
1919 void ll_inode_size_unlock(struct inode *inode)
1920 {
1921         struct ll_inode_info *lli;
1922
1923         lli = ll_i2info(inode);
1924         mutex_unlock(&lli->lli_size_mutex);
1925 }
1926
1927 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1928 {
1929         struct ll_inode_info *lli = ll_i2info(inode);
1930         struct mdt_body *body = md->body;
1931         struct lov_stripe_md *lsm = md->lsm;
1932         struct ll_sb_info *sbi = ll_i2sbi(inode);
1933
1934         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1935         if (lsm != NULL) {
1936                 if (!lli->lli_has_smd &&
1937                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1938                         cl_file_inode_init(inode, md);
1939
1940                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1941                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1942                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1943         }
1944
1945         if (S_ISDIR(inode->i_mode)) {
1946                 int     rc;
1947
1948                 rc = ll_update_lsm_md(inode, md);
1949                 if (rc != 0)
1950                         return rc;
1951         }
1952
1953         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1954                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1955                         ll_update_remote_perm(inode, md->remote_perm);
1956         }
1957 #ifdef CONFIG_FS_POSIX_ACL
1958         else if (body->mbo_valid & OBD_MD_FLACL) {
1959                 spin_lock(&lli->lli_lock);
1960                 if (lli->lli_posix_acl)
1961                         posix_acl_release(lli->lli_posix_acl);
1962                 lli->lli_posix_acl = md->posix_acl;
1963                 spin_unlock(&lli->lli_lock);
1964         }
1965 #endif
1966         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1967                                         sbi->ll_flags & LL_SBI_32BIT_API);
1968         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1969
1970         if (body->mbo_valid & OBD_MD_FLATIME) {
1971                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1972                         LTIME_S(inode->i_atime) = body->mbo_atime;
1973                 lli->lli_lvb.lvb_atime = body->mbo_atime;
1974         }
1975
1976         if (body->mbo_valid & OBD_MD_FLMTIME) {
1977                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1978                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1979                                "to "LPU64"\n", inode->i_ino,
1980                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1981                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1982                 }
1983                 lli->lli_lvb.lvb_mtime = body->mbo_mtime;
1984         }
1985
1986         if (body->mbo_valid & OBD_MD_FLCTIME) {
1987                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1988                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1989                 lli->lli_lvb.lvb_ctime = body->mbo_ctime;
1990         }
1991
1992         if (body->mbo_valid & OBD_MD_FLMODE)
1993                 inode->i_mode = (inode->i_mode & S_IFMT) |
1994                                 (body->mbo_mode & ~S_IFMT);
1995
1996         if (body->mbo_valid & OBD_MD_FLTYPE)
1997                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1998                                 (body->mbo_mode & S_IFMT);
1999
2000         LASSERT(inode->i_mode != 0);
2001         if (S_ISREG(inode->i_mode))
2002                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2003                                        LL_MAX_BLKSIZE_BITS);
2004         else
2005                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2006
2007         if (body->mbo_valid & OBD_MD_FLUID)
2008                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2009         if (body->mbo_valid & OBD_MD_FLGID)
2010                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2011         if (body->mbo_valid & OBD_MD_FLFLAGS)
2012                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
2013         if (body->mbo_valid & OBD_MD_FLNLINK)
2014                 set_nlink(inode, body->mbo_nlink);
2015         if (body->mbo_valid & OBD_MD_FLRDEV)
2016                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2017
2018         if (body->mbo_valid & OBD_MD_FLID) {
2019                 /* FID shouldn't be changed! */
2020                 if (fid_is_sane(&lli->lli_fid)) {
2021                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2022                                  "Trying to change FID "DFID
2023                                  " to the "DFID", inode "DFID"(%p)\n",
2024                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2025                                  PFID(ll_inode2fid(inode)), inode);
2026                 } else {
2027                         lli->lli_fid = body->mbo_fid1;
2028                 }
2029         }
2030
2031         LASSERT(fid_seq(&lli->lli_fid) != 0);
2032
2033         if (body->mbo_valid & OBD_MD_FLSIZE) {
2034                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2035                     S_ISREG(inode->i_mode)) {
2036                         struct lustre_handle lockh;
2037                         ldlm_mode_t mode;
2038
2039                         /* As it is possible a blocking ast has been processed
2040                          * by this time, we need to check there is an UPDATE
2041                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2042                          * it. */
2043                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2044                                                &lockh, LDLM_FL_CBPENDING,
2045                                                LCK_CR | LCK_CW |
2046                                                LCK_PR | LCK_PW);
2047                         if (mode) {
2048                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2049                                                       LLIF_EPOCH_PENDING |
2050                                                       LLIF_SOM_DIRTY)) {
2051                                         CERROR("%s: inode "DFID" flags %u still"
2052                                                " has size authority! do not "
2053                                                "trust the size from MDS\n",
2054                                                sbi->ll_md_exp->exp_obd->obd_name,
2055                                                PFID(ll_inode2fid(inode)),
2056                                                lli->lli_flags);
2057                                 } else {
2058                                         /* Use old size assignment to avoid
2059                                          * deadlock bz14138 & bz14326 */
2060                                         i_size_write(inode, body->mbo_size);
2061                                         spin_lock(&lli->lli_lock);
2062                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2063                                         spin_unlock(&lli->lli_lock);
2064                                 }
2065                                 ldlm_lock_decref(&lockh, mode);
2066                         }
2067                 } else {
2068                         /* Use old size assignment to avoid
2069                          * deadlock bz14138 & bz14326 */
2070                         i_size_write(inode, body->mbo_size);
2071
2072                         CDEBUG(D_VFSTRACE,
2073                                "inode="DFID", updating i_size %llu\n",
2074                                PFID(ll_inode2fid(inode)),
2075                                (unsigned long long)body->mbo_size);
2076                 }
2077
2078                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2079                         inode->i_blocks = body->mbo_blocks;
2080         }
2081
2082         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2083                 LASSERT(md->mds_capa);
2084                 ll_add_capa(inode, md->mds_capa);
2085         }
2086
2087         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2088                 LASSERT(md->oss_capa);
2089                 ll_add_capa(inode, md->oss_capa);
2090         }
2091
2092         if (body->mbo_valid & OBD_MD_TSTATE) {
2093                 if (body->mbo_t_state & MS_RESTORE)
2094                         lli->lli_flags |= LLIF_FILE_RESTORING;
2095         }
2096
2097         return 0;
2098 }
2099
2100 int ll_read_inode2(struct inode *inode, void *opaque)
2101 {
2102         struct lustre_md *md = opaque;
2103         struct ll_inode_info *lli = ll_i2info(inode);
2104         int     rc;
2105         ENTRY;
2106
2107         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2108                PFID(&lli->lli_fid), inode);
2109
2110         LASSERT(!lli->lli_has_smd);
2111
2112         /* Core attributes from the MDS first.  This is a new inode, and
2113          * the VFS doesn't zero times in the core inode so we have to do
2114          * it ourselves.  They will be overwritten by either MDS or OST
2115          * attributes - we just need to make sure they aren't newer. */
2116         LTIME_S(inode->i_mtime) = 0;
2117         LTIME_S(inode->i_atime) = 0;
2118         LTIME_S(inode->i_ctime) = 0;
2119         inode->i_rdev = 0;
2120         rc = ll_update_inode(inode, md);
2121         if (rc != 0)
2122                 RETURN(rc);
2123
2124         /* OIDEBUG(inode); */
2125
2126         /* initializing backing dev info. */
2127         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2128
2129
2130         if (S_ISREG(inode->i_mode)) {
2131                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2132                 inode->i_op = &ll_file_inode_operations;
2133                 inode->i_fop = sbi->ll_fop;
2134                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2135                 EXIT;
2136         } else if (S_ISDIR(inode->i_mode)) {
2137                 inode->i_op = &ll_dir_inode_operations;
2138                 inode->i_fop = &ll_dir_operations;
2139                 EXIT;
2140         } else if (S_ISLNK(inode->i_mode)) {
2141                 inode->i_op = &ll_fast_symlink_inode_operations;
2142                 EXIT;
2143         } else {
2144                 inode->i_op = &ll_special_inode_operations;
2145
2146                 init_special_inode(inode, inode->i_mode,
2147                                    inode->i_rdev);
2148
2149                 EXIT;
2150         }
2151
2152         return 0;
2153 }
2154
2155 void ll_delete_inode(struct inode *inode)
2156 {
2157         struct cl_inode_info *lli = cl_i2info(inode);
2158         ENTRY;
2159
2160         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2161                 /* discard all dirty pages before truncating them, required by
2162                  * osc_extent implementation at LU-1030. */
2163                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2164                                    CL_FSYNC_DISCARD, 1);
2165
2166         truncate_inode_pages(&inode->i_data, 0);
2167
2168         /* Workaround for LU-118 */
2169         if (inode->i_data.nrpages) {
2170                 spin_lock_irq(&inode->i_data.tree_lock);
2171                 spin_unlock_irq(&inode->i_data.tree_lock);
2172                 LASSERTF(inode->i_data.nrpages == 0,
2173                          "inode="DFID"(%p) nrpages=%lu, see "
2174                          "http://jira.whamcloud.com/browse/LU-118\n",
2175                          PFID(ll_inode2fid(inode)), inode,
2176                          inode->i_data.nrpages);
2177         }
2178         /* Workaround end */
2179
2180 #ifdef HAVE_SBOPS_EVICT_INODE
2181         ll_clear_inode(inode);
2182 #endif
2183         clear_inode(inode);
2184
2185         EXIT;
2186 }
2187
2188 int ll_iocontrol(struct inode *inode, struct file *file,
2189                  unsigned int cmd, unsigned long arg)
2190 {
2191         struct ll_sb_info *sbi = ll_i2sbi(inode);
2192         struct ptlrpc_request *req = NULL;
2193         int rc, flags = 0;
2194         ENTRY;
2195
2196         switch(cmd) {
2197         case FSFILT_IOC_GETFLAGS: {
2198                 struct mdt_body *body;
2199                 struct md_op_data *op_data;
2200
2201                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2202                                              0, 0, LUSTRE_OPC_ANY,
2203                                              NULL);
2204                 if (IS_ERR(op_data))
2205                         RETURN(PTR_ERR(op_data));
2206
2207                 op_data->op_valid = OBD_MD_FLFLAGS;
2208                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2209                 ll_finish_md_op_data(op_data);
2210                 if (rc) {
2211                         CERROR("%s: failure inode "DFID": rc = %d\n",
2212                                sbi->ll_md_exp->exp_obd->obd_name,
2213                                PFID(ll_inode2fid(inode)), rc);
2214                         RETURN(-abs(rc));
2215                 }
2216
2217                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2218
2219                 flags = body->mbo_flags;
2220
2221                 ptlrpc_req_finished(req);
2222
2223                 RETURN(put_user(flags, (int *)arg));
2224         }
2225         case FSFILT_IOC_SETFLAGS: {
2226                 struct lov_stripe_md *lsm;
2227                 struct obd_info oinfo = { { { 0 } } };
2228                 struct md_op_data *op_data;
2229
2230                 if (get_user(flags, (int *)arg))
2231                         RETURN(-EFAULT);
2232
2233                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2234                                              LUSTRE_OPC_ANY, NULL);
2235                 if (IS_ERR(op_data))
2236                         RETURN(PTR_ERR(op_data));
2237
2238                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2239                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2240                 rc = md_setattr(sbi->ll_md_exp, op_data,
2241                                 NULL, 0, NULL, 0, &req, NULL);
2242                 ll_finish_md_op_data(op_data);
2243                 ptlrpc_req_finished(req);
2244                 if (rc)
2245                         RETURN(rc);
2246
2247                 inode->i_flags = ll_ext_to_inode_flags(flags);
2248
2249                 lsm = ccc_inode_lsm_get(inode);
2250                 if (!lsm_has_objects(lsm)) {
2251                         ccc_inode_lsm_put(inode, lsm);
2252                         RETURN(0);
2253                 }
2254
2255                 OBDO_ALLOC(oinfo.oi_oa);
2256                 if (!oinfo.oi_oa) {
2257                         ccc_inode_lsm_put(inode, lsm);
2258                         RETURN(-ENOMEM);
2259                 }
2260                 oinfo.oi_md = lsm;
2261                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2262                 oinfo.oi_oa->o_flags = flags;
2263                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2264                                        OBD_MD_FLGROUP;
2265                 oinfo.oi_capa = ll_mdscapa_get(inode);
2266                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2267                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2268                 capa_put(oinfo.oi_capa);
2269                 OBDO_FREE(oinfo.oi_oa);
2270                 ccc_inode_lsm_put(inode, lsm);
2271
2272                 if (rc && rc != -EPERM && rc != -EACCES)
2273                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2274
2275                 RETURN(rc);
2276         }
2277         default:
2278                 RETURN(-ENOSYS);
2279         }
2280
2281         RETURN(0);
2282 }
2283
2284 int ll_flush_ctx(struct inode *inode)
2285 {
2286         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2287
2288         CDEBUG(D_SEC, "flush context for user %d\n",
2289                from_kuid(&init_user_ns, current_uid()));
2290
2291         obd_set_info_async(NULL, sbi->ll_md_exp,
2292                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2293                            0, NULL, NULL);
2294         obd_set_info_async(NULL, sbi->ll_dt_exp,
2295                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2296                            0, NULL, NULL);
2297         return 0;
2298 }
2299
2300 /* umount -f client means force down, don't save state */
2301 void ll_umount_begin(struct super_block *sb)
2302 {
2303         struct ll_sb_info *sbi = ll_s2sbi(sb);
2304         struct obd_device *obd;
2305         struct obd_ioctl_data *ioc_data;
2306         ENTRY;
2307
2308         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2309                sb->s_count, atomic_read(&sb->s_active));
2310
2311         obd = class_exp2obd(sbi->ll_md_exp);
2312         if (obd == NULL) {
2313                 CERROR("Invalid MDC connection handle "LPX64"\n",
2314                        sbi->ll_md_exp->exp_handle.h_cookie);
2315                 EXIT;
2316                 return;
2317         }
2318         obd->obd_force = 1;
2319
2320         obd = class_exp2obd(sbi->ll_dt_exp);
2321         if (obd == NULL) {
2322                 CERROR("Invalid LOV connection handle "LPX64"\n",
2323                        sbi->ll_dt_exp->exp_handle.h_cookie);
2324                 EXIT;
2325                 return;
2326         }
2327         obd->obd_force = 1;
2328
2329         OBD_ALLOC_PTR(ioc_data);
2330         if (ioc_data) {
2331                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2332                               sizeof *ioc_data, ioc_data, NULL);
2333
2334                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2335                               sizeof *ioc_data, ioc_data, NULL);
2336
2337                 OBD_FREE_PTR(ioc_data);
2338         }
2339
2340         /* Really, we'd like to wait until there are no requests outstanding,
2341          * and then continue.  For now, we just invalidate the requests,
2342          * schedule() and sleep one second if needed, and hope.
2343          */
2344         schedule();
2345         EXIT;
2346 }
2347
2348 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2349 {
2350         struct ll_sb_info *sbi = ll_s2sbi(sb);
2351         char *profilenm = get_profile_name(sb);
2352         int err;
2353         __u32 read_only;
2354
2355         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2356                 read_only = *flags & MS_RDONLY;
2357                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2358                                          sizeof(KEY_READ_ONLY),
2359                                          KEY_READ_ONLY, sizeof(read_only),
2360                                          &read_only, NULL);
2361                 if (err) {
2362                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2363                                       profilenm, read_only ?
2364                                       "read-only" : "read-write", err);
2365                         return err;
2366                 }
2367
2368                 if (read_only)
2369                         sb->s_flags |= MS_RDONLY;
2370                 else
2371                         sb->s_flags &= ~MS_RDONLY;
2372
2373                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2374                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2375                                       read_only ?  "read-only" : "read-write");
2376         }
2377         return 0;
2378 }
2379
2380 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2381                   struct super_block *sb, struct lookup_intent *it)
2382 {
2383         struct ll_sb_info *sbi = NULL;
2384         struct lustre_md md = { 0 };
2385         int rc;
2386         ENTRY;
2387
2388         LASSERT(*inode || sb);
2389         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2390         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2391                               sbi->ll_md_exp, &md);
2392         if (rc)
2393                 RETURN(rc);
2394
2395         if (*inode) {
2396                 rc = ll_update_inode(*inode, &md);
2397                 if (rc != 0)
2398                         GOTO(out, rc);
2399         } else {
2400                 LASSERT(sb != NULL);
2401
2402                 /*
2403                  * At this point server returns to client's same fid as client
2404                  * generated for creating. So using ->fid1 is okay here.
2405                  */
2406                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2407
2408                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2409                                              sbi->ll_flags & LL_SBI_32BIT_API),
2410                                  &md);
2411                 if (IS_ERR(*inode)) {
2412 #ifdef CONFIG_FS_POSIX_ACL
2413                         if (md.posix_acl) {
2414                                 posix_acl_release(md.posix_acl);
2415                                 md.posix_acl = NULL;
2416                         }
2417 #endif
2418                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2419                         *inode = NULL;
2420                         CERROR("new_inode -fatal: rc %d\n", rc);
2421                         GOTO(out, rc);
2422                 }
2423         }
2424
2425         /* Handling piggyback layout lock.
2426          * Layout lock can be piggybacked by getattr and open request.
2427          * The lsm can be applied to inode only if it comes with a layout lock
2428          * otherwise correct layout may be overwritten, for example:
2429          * 1. proc1: mdt returns a lsm but not granting layout
2430          * 2. layout was changed by another client
2431          * 3. proc2: refresh layout and layout lock granted
2432          * 4. proc1: to apply a stale layout */
2433         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2434                 struct lustre_handle lockh;
2435                 struct ldlm_lock *lock;
2436
2437                 lockh.cookie = it->d.lustre.it_lock_handle;
2438                 lock = ldlm_handle2lock(&lockh);
2439                 LASSERT(lock != NULL);
2440                 if (ldlm_has_layout(lock)) {
2441                         struct cl_object_conf conf;
2442
2443                         memset(&conf, 0, sizeof(conf));
2444                         conf.coc_opc = OBJECT_CONF_SET;
2445                         conf.coc_inode = *inode;
2446                         conf.coc_lock = lock;
2447                         conf.u.coc_md = &md;
2448                         (void)ll_layout_conf(*inode, &conf);
2449                 }
2450                 LDLM_LOCK_PUT(lock);
2451         }
2452
2453 out:
2454         if (md.lsm != NULL)
2455                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2456         md_free_lustre_md(sbi->ll_md_exp, &md);
2457         RETURN(rc);
2458 }
2459
2460 int ll_obd_statfs(struct inode *inode, void *arg)
2461 {
2462         struct ll_sb_info *sbi = NULL;
2463         struct obd_export *exp;
2464         char *buf = NULL;
2465         struct obd_ioctl_data *data = NULL;
2466         __u32 type;
2467         __u32 flags;
2468         int len = 0, rc;
2469
2470         if (!inode || !(sbi = ll_i2sbi(inode)))
2471                 GOTO(out_statfs, rc = -EINVAL);
2472
2473         rc = obd_ioctl_getdata(&buf, &len, arg);
2474         if (rc)
2475                 GOTO(out_statfs, rc);
2476
2477         data = (void*)buf;
2478         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2479             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2480                 GOTO(out_statfs, rc = -EINVAL);
2481
2482         if (data->ioc_inllen1 != sizeof(__u32) ||
2483             data->ioc_inllen2 != sizeof(__u32) ||
2484             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2485             data->ioc_plen2 != sizeof(struct obd_uuid))
2486                 GOTO(out_statfs, rc = -EINVAL);
2487
2488         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2489         if (type & LL_STATFS_LMV)
2490                 exp = sbi->ll_md_exp;
2491         else if (type & LL_STATFS_LOV)
2492                 exp = sbi->ll_dt_exp;
2493         else
2494                 GOTO(out_statfs, rc = -ENODEV);
2495
2496         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2497         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2498         if (rc)
2499                 GOTO(out_statfs, rc);
2500 out_statfs:
2501         if (buf)
2502                 obd_ioctl_freedata(buf, len);
2503         return rc;
2504 }
2505
2506 int ll_process_config(struct lustre_cfg *lcfg)
2507 {
2508         struct super_block *sb;
2509         unsigned long x;
2510         int rc = 0;
2511         char *ptr;
2512
2513         /* The instance name contains the sb: lustre-client-aacfe000 */
2514         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2515         if (!ptr || !*(++ptr))
2516                 return -EINVAL;
2517         if (sscanf(ptr, "%lx", &x) != 1)
2518                 return -EINVAL;
2519         sb = (struct super_block *)x;
2520         /* This better be a real Lustre superblock! */
2521         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2522
2523         /* Note we have not called client_common_fill_super yet, so
2524            proc fns must be able to handle that! */
2525         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2526                                           lcfg, sb);
2527         if (rc > 0)
2528                 rc = 0;
2529         return rc;
2530 }
2531
2532 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2533 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2534                                        struct inode *i1, struct inode *i2,
2535                                        const char *name, int namelen,
2536                                        int mode, __u32 opc, void *data)
2537 {
2538         LASSERT(i1 != NULL);
2539
2540         if (name == NULL) {
2541                 /* Do not reuse namelen for something else. */
2542                 if (namelen != 0)
2543                         return ERR_PTR(-EINVAL);
2544         } else {
2545                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2546                         return ERR_PTR(-ENAMETOOLONG);
2547
2548                 if (!lu_name_is_valid_2(name, namelen))
2549                         return ERR_PTR(-EINVAL);
2550         }
2551
2552         if (op_data == NULL)
2553                 OBD_ALLOC_PTR(op_data);
2554
2555         if (op_data == NULL)
2556                 return ERR_PTR(-ENOMEM);
2557
2558         ll_i2gids(op_data->op_suppgids, i1, i2);
2559         op_data->op_fid1 = *ll_inode2fid(i1);
2560         op_data->op_capa1 = ll_mdscapa_get(i1);
2561         if (S_ISDIR(i1->i_mode))
2562                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2563
2564         if (i2) {
2565                 op_data->op_fid2 = *ll_inode2fid(i2);
2566                 op_data->op_capa2 = ll_mdscapa_get(i2);
2567                 if (S_ISDIR(i2->i_mode))
2568                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2569         } else {
2570                 fid_zero(&op_data->op_fid2);
2571                 op_data->op_capa2 = NULL;
2572         }
2573
2574         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2575                 op_data->op_cli_flags |= CLI_HASH64;
2576
2577         if (ll_need_32bit_api(ll_i2sbi(i1)))
2578                 op_data->op_cli_flags |= CLI_API32;
2579
2580         op_data->op_name = name;
2581         op_data->op_namelen = namelen;
2582         op_data->op_mode = mode;
2583         op_data->op_mod_time = cfs_time_current_sec();
2584         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2585         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2586         op_data->op_cap = cfs_curproc_cap_pack();
2587         op_data->op_bias = 0;
2588         op_data->op_cli_flags = 0;
2589         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2590              filename_is_volatile(name, namelen, NULL))
2591                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2592         op_data->op_mds = 0;
2593         op_data->op_data = data;
2594
2595         /* When called by ll_setattr_raw, file is i1. */
2596         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2597                 op_data->op_bias |= MDS_DATA_MODIFIED;
2598
2599         return op_data;
2600 }
2601
2602 void ll_finish_md_op_data(struct md_op_data *op_data)
2603 {
2604         capa_put(op_data->op_capa1);
2605         capa_put(op_data->op_capa2);
2606         OBD_FREE_PTR(op_data);
2607 }
2608
2609 #ifdef HAVE_SUPEROPS_USE_DENTRY
2610 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2611 #else
2612 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2613 #endif
2614 {
2615         struct ll_sb_info *sbi;
2616
2617 #ifdef HAVE_SUPEROPS_USE_DENTRY
2618         LASSERT((seq != NULL) && (dentry != NULL));
2619         sbi = ll_s2sbi(dentry->d_sb);
2620 #else
2621         LASSERT((seq != NULL) && (vfs != NULL));
2622         sbi = ll_s2sbi(vfs->mnt_sb);
2623 #endif
2624
2625         if (sbi->ll_flags & LL_SBI_NOLCK)
2626                 seq_puts(seq, ",nolock");
2627
2628         if (sbi->ll_flags & LL_SBI_FLOCK)
2629                 seq_puts(seq, ",flock");
2630
2631         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2632                 seq_puts(seq, ",localflock");
2633
2634         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2635                 seq_puts(seq, ",user_xattr");
2636
2637         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2638                 seq_puts(seq, ",lazystatfs");
2639
2640         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2641                 seq_puts(seq, ",user_fid2path");
2642
2643         RETURN(0);
2644 }
2645
2646 /**
2647  * Get obd name by cmd, and copy out to user space
2648  */
2649 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2650 {
2651         struct ll_sb_info *sbi = ll_i2sbi(inode);
2652         struct obd_device *obd;
2653         ENTRY;
2654
2655         if (cmd == OBD_IOC_GETDTNAME)
2656                 obd = class_exp2obd(sbi->ll_dt_exp);
2657         else if (cmd == OBD_IOC_GETMDNAME)
2658                 obd = class_exp2obd(sbi->ll_md_exp);
2659         else
2660                 RETURN(-EINVAL);
2661
2662         if (!obd)
2663                 RETURN(-ENOENT);
2664
2665         if (copy_to_user((void *)arg, obd->obd_name,
2666                          strlen(obd->obd_name) + 1))
2667                 RETURN(-EFAULT);
2668
2669         RETURN(0);
2670 }
2671
2672 /**
2673  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2674  * fsname will be returned in this buffer; otherwise, a static buffer will be
2675  * used to store the fsname and returned to caller.
2676  */
2677 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2678 {
2679         static char fsname_static[MTI_NAME_MAXLEN];
2680         struct lustre_sb_info *lsi = s2lsi(sb);
2681         char *ptr;
2682         int len;
2683
2684         if (buf == NULL) {
2685                 /* this means the caller wants to use static buffer
2686                  * and it doesn't care about race. Usually this is
2687                  * in error reporting path */
2688                 buf = fsname_static;
2689                 buflen = sizeof(fsname_static);
2690         }
2691
2692         len = strlen(lsi->lsi_lmd->lmd_profile);
2693         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2694         if (ptr && (strcmp(ptr, "-client") == 0))
2695                 len -= 7;
2696
2697         if (unlikely(len >= buflen))
2698                 len = buflen - 1;
2699         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2700         buf[len] = '\0';
2701
2702         return buf;
2703 }
2704
2705 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2706 {
2707         char *path = NULL;
2708
2709         struct path p;
2710
2711         p.dentry = dentry;
2712         p.mnt = current->fs->root.mnt;
2713         path_get(&p);
2714         path = d_path(&p, buf, bufsize);
2715         path_put(&p);
2716         return path;
2717 }
2718
2719 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2720 {
2721         char *buf, *path = NULL;
2722         struct dentry *dentry = NULL;
2723         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2724
2725         /* this can be called inside spin lock so use GFP_ATOMIC. */
2726         buf = (char *)__get_free_page(GFP_ATOMIC);
2727         if (buf != NULL) {
2728                 dentry = d_find_alias(page->mapping->host);
2729                 if (dentry != NULL)
2730                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2731         }
2732
2733         CDEBUG(D_WARNING,
2734                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2735                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2736                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2737                PFID(&obj->cob_header.coh_lu.loh_fid),
2738                (path && !IS_ERR(path)) ? path : "", ioret);
2739
2740         if (dentry != NULL)
2741                 dput(dentry);
2742
2743         if (buf != NULL)
2744                 free_page((unsigned long)buf);
2745 }
2746
2747 /*
2748  * Compute llite root squash state after a change of root squash
2749  * configuration setting or add/remove of a lnet nid
2750  */
2751 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2752 {
2753         struct root_squash_info *squash = &sbi->ll_squash;
2754         int i;
2755         bool matched;
2756         lnet_process_id_t id;
2757
2758         /* Update norootsquash flag */
2759         down_write(&squash->rsi_sem);
2760         if (list_empty(&squash->rsi_nosquash_nids))
2761                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2762         else {
2763                 /* Do not apply root squash as soon as one of our NIDs is
2764                  * in the nosquash_nids list */
2765                 matched = false;
2766                 i = 0;
2767                 while (LNetGetId(i++, &id) != -ENOENT) {
2768                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2769                                 continue;
2770                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2771                                 matched = true;
2772                                 break;
2773                         }
2774                 }
2775                 if (matched)
2776                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2777                 else
2778                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2779         }
2780         up_write(&squash->rsi_sem);
2781 }
2782
2783