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