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/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_ioctl.h>
  50 #include <lustre_lite.h>
  51 #include <lustre_ha.h>
  52 #include <lustre_dlm.h>
  53 #include <lprocfs_status.h>
  54 #include <lustre_disk.h>
  55 #include <lustre_param.h>
  56 #include <lustre_log.h>
  57 #include <cl_object.h>
  58 #include <obd_cksum.h>
  59 #include "llite_internal.h"
  60
  61 struct kmem_cache *ll_file_data_slab;
  62
  63 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  64 static DEFINE_SPINLOCK(ll_sb_lock);
  65
  66 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static struct ll_sb_info *ll_init_sbi(void)
  71 {
  72         struct ll_sb_info *sbi = NULL;
  73         unsigned long pages;
  74         unsigned long lru_page_max;
  75         struct sysinfo si;
  76         class_uuid_t uuid;
  77         int i;
  78         ENTRY;
  79
  80         OBD_ALLOC_PTR(sbi);
  81         if (sbi == NULL)
  82                 RETURN(NULL);
  83
  84         spin_lock_init(&sbi->ll_lock);
  85         mutex_init(&sbi->ll_lco.lco_lock);
  86         spin_lock_init(&sbi->ll_pp_extent_lock);
  87         spin_lock_init(&sbi->ll_process_lock);
  88         sbi->ll_rw_stats_on = 0;
  89
  90         si_meminfo(&si);
  91         pages = si.totalram - si.totalhigh;
  92         lru_page_max = pages / 2;
  93
  94         /* initialize ll_cache data */
  95         atomic_set(&sbi->ll_cache.ccc_users, 0);
  96         sbi->ll_cache.ccc_lru_max = lru_page_max;
  97         atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
  98         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
  99         INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 100
 101         /* turn unstable check off by default as it impacts performance */
 102         sbi->ll_cache.ccc_unstable_check = 0;
 103         atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 104         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 105
 106         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 107                                            SBI_DEFAULT_READAHEAD_MAX);
 108         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 109         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 110                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 111         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 112         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 113
 114         ll_generate_random_uuid(uuid);
 115         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 116         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 117
 118         spin_lock(&ll_sb_lock);
 119         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 120         spin_unlock(&ll_sb_lock);
 121
 122         sbi->ll_flags |= LL_SBI_VERBOSE;
 123 #ifdef ENABLE_CHECKSUM
 124         sbi->ll_flags |= LL_SBI_CHECKSUM;
 125 #endif
 126
 127 #ifdef HAVE_LRU_RESIZE_SUPPORT
 128         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 129 #endif
 130
 131         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 133                                pp_r_hist.oh_lock);
 134                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 135                                pp_w_hist.oh_lock);
 136         }
 137
 138         /* metadata statahead is enabled by default */
 139         sbi->ll_sa_max = LL_SA_RPC_DEF;
 140         atomic_set(&sbi->ll_sa_total, 0);
 141         atomic_set(&sbi->ll_sa_wrong, 0);
 142         atomic_set(&sbi->ll_agl_total, 0);
 143         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 144
 145         /* root squash */
 146         sbi->ll_squash.rsi_uid = 0;
 147         sbi->ll_squash.rsi_gid = 0;
 148         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 149         init_rwsem(&sbi->ll_squash.rsi_sem);
 150
 151         RETURN(sbi);
 152 }
 153
 154 static void ll_free_sbi(struct super_block *sb)
 155 {
 156         struct ll_sb_info *sbi = ll_s2sbi(sb);
 157         ENTRY;
 158
 159         if (sbi != NULL) {
 160                 spin_lock(&ll_sb_lock);
 161                 list_del(&sbi->ll_list);
 162                 spin_unlock(&ll_sb_lock);
 163                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 164                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 165                 OBD_FREE(sbi, sizeof(*sbi));
 166         }
 167         EXIT;
 168 }
 169
 170 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 171                                     struct vfsmount *mnt)
 172 {
 173         struct inode *root = 0;
 174         struct ll_sb_info *sbi = ll_s2sbi(sb);
 175         struct obd_device *obd;
 176         struct obd_capa *oc = NULL;
 177         struct obd_statfs *osfs = NULL;
 178         struct ptlrpc_request *request = NULL;
 179         struct obd_connect_data *data = NULL;
 180         struct obd_uuid *uuid;
 181         struct md_op_data *op_data;
 182         struct lustre_md lmd;
 183         obd_valid valid;
 184         int size, err, checksum;
 185         ENTRY;
 186
 187         obd = class_name2obd(md);
 188         if (!obd) {
 189                 CERROR("MD %s: not setup or attached\n", md);
 190                 RETURN(-EINVAL);
 191         }
 192
 193         OBD_ALLOC_PTR(data);
 194         if (data == NULL)
 195                 RETURN(-ENOMEM);
 196
 197         OBD_ALLOC_PTR(osfs);
 198         if (osfs == NULL) {
 199                 OBD_FREE_PTR(data);
 200                 RETURN(-ENOMEM);
 201         }
 202
 203         /* indicate the features supported by this client */
 204         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 205                                   OBD_CONNECT_ATTRFID  |
 206                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 207                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 208                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 209                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 210                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 211                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 212                                   OBD_CONNECT_EINPROGRESS |
 213                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 214                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 215                                   OBD_CONNECT_MAX_EASIZE |
 216                                   OBD_CONNECT_FLOCK_DEAD |
 217                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 218                                   OBD_CONNECT_OPEN_BY_FID |
 219                                   OBD_CONNECT_DIR_STRIPE;
 220
 221         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 222                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 223
 224 #ifdef HAVE_LRU_RESIZE_SUPPORT
 225         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 226                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 227 #endif
 228 #ifdef CONFIG_FS_POSIX_ACL
 229         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 230 #endif
 231
 232         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 233                 /* flag mdc connection as lightweight, only used for test
 234                  * purpose, use with care */
 235                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 236
 237         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 238         data->ocd_version = LUSTRE_VERSION_CODE;
 239
 240         if (sb->s_flags & MS_RDONLY)
 241                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 242         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 243                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 244
 245 #ifdef HAVE_MS_FLOCK_LOCK
 246         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 247         sb->s_flags |= MS_FLOCK_LOCK;
 248 #endif
 249 #ifdef MS_HAS_NEW_AOPS
 250         sb->s_flags |= MS_HAS_NEW_AOPS;
 251 #endif
 252
 253         if (sbi->ll_flags & LL_SBI_FLOCK)
 254                 sbi->ll_fop = &ll_file_operations_flock;
 255         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 256                 sbi->ll_fop = &ll_file_operations;
 257         else
 258                 sbi->ll_fop = &ll_file_operations_noflock;
 259
 260         /* real client */
 261         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 262         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 263                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 264
 265         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 266
 267         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 268         if (err == -EBUSY) {
 269                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 270                                    "recovery, of which this client is not a "
 271                                    "part. Please wait for recovery to complete,"
 272                                    " abort, or time out.\n", md);
 273                 GOTO(out, err);
 274         } else if (err) {
 275                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 276                 GOTO(out, err);
 277         }
 278
 279         sbi->ll_md_exp->exp_connect_data = *data;
 280
 281         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 282                            LUSTRE_SEQ_METADATA);
 283         if (err) {
 284                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 285                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 286                 GOTO(out_md, err);
 287         }
 288
 289         /* For mount, we only need fs info from MDT0, and also in DNE, it
 290          * can make sure the client can be mounted as long as MDT0 is
 291          * avaible */
 292         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 293                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 294                         OBD_STATFS_FOR_MDT0);
 295         if (err)
 296                 GOTO(out_md_fid, err);
 297
 298         /* This needs to be after statfs to ensure connect has finished.
 299          * Note that "data" does NOT contain the valid connect reply.
 300          * If connecting to a 1.8 server there will be no LMV device, so
 301          * we can access the MDC export directly and exp_connect_flags will
 302          * be non-zero, but if accessing an upgraded 2.1 server it will
 303          * have the correct flags filled in.
 304          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 305         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 306         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 307             valid != CLIENT_CONNECT_MDT_REQD) {
 308                 char *buf;
 309
 310                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 311                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 312                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 313                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 314                                    "feature(s) needed for correct operation "
 315                                    "of this client (%s). Please upgrade "
 316                                    "server or downgrade client.\n",
 317                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 318                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 319                 GOTO(out_md_fid, err = -EPROTO);
 320         }
 321
 322         size = sizeof(*data);
 323         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 324                            KEY_CONN_DATA,  &size, data, NULL);
 325         if (err) {
 326                 CERROR("%s: Get connect data failed: rc = %d\n",
 327                        sbi->ll_md_exp->exp_obd->obd_name, err);
 328                 GOTO(out_md_fid, err);
 329         }
 330
 331         LASSERT(osfs->os_bsize);
 332         sb->s_blocksize = osfs->os_bsize;
 333         sb->s_blocksize_bits = log2(osfs->os_bsize);
 334         sb->s_magic = LL_SUPER_MAGIC;
 335         sb->s_maxbytes = MAX_LFS_FILESIZE;
 336         sbi->ll_namelen = osfs->os_namelen;
 337         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 338
 339         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 340             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 341                 LCONSOLE_INFO("Disabling user_xattr feature because "
 342                               "it is not supported on the server\n");
 343                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 344         }
 345
 346         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 347 #ifdef MS_POSIXACL
 348                 sb->s_flags |= MS_POSIXACL;
 349 #endif
 350                 sbi->ll_flags |= LL_SBI_ACL;
 351         } else {
 352                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 353 #ifdef MS_POSIXACL
 354                 sb->s_flags &= ~MS_POSIXACL;
 355 #endif
 356                 sbi->ll_flags &= ~LL_SBI_ACL;
 357         }
 358
 359         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 360                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 361                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 362                         LCONSOLE_INFO("client is set as remote by default.\n");
 363                 }
 364         } else {
 365                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 366                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 367                         LCONSOLE_INFO("client claims to be remote, but server "
 368                                       "rejected, forced to be local.\n");
 369                 }
 370         }
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 373                 LCONSOLE_INFO("client enabled MDS capability!\n");
 374                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 375         }
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 378                 LCONSOLE_INFO("client enabled OSS capability!\n");
 379                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 380         }
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 383                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 384
 385         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 386                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 387         else
 388                 sbi->ll_md_brw_pages = 1;
 389
 390         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 391                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 392
 393         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 394                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 395                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 396                                       "unknown maximum xattr size.\n", dt);
 397                 } else {
 398                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 399                         sbi->ll_xattr_cache_enabled = 1;
 400                 }
 401         }
 402
 403         obd = class_name2obd(dt);
 404         if (!obd) {
 405                 CERROR("DT %s: not setup or attached\n", dt);
 406                 GOTO(out_md_fid, err = -ENODEV);
 407         }
 408
 409         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 410                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 411                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 412                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 413                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 414                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 415                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 416                                   OBD_CONNECT_MAXBYTES |
 417                                   OBD_CONNECT_EINPROGRESS |
 418                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 419                                   OBD_CONNECT_LAYOUTLOCK |
 420                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 421
 422         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 423                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 424
 425         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 426                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 427                  * disabled by default, because it can still be enabled on the
 428                  * fly via /proc. As a consequence, we still need to come to an
 429                  * agreement on the supported algorithms at connect time */
 430                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 431
 432                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 433                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 434                 else
 435                         data->ocd_cksum_types = cksum_types_supported_client();
 436         }
 437
 438 #ifdef HAVE_LRU_RESIZE_SUPPORT
 439         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 440 #endif
 441         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 442                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 443
 444         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 445                "ocd_grant: %d\n", data->ocd_connect_flags,
 446                data->ocd_version, data->ocd_grant);
 447
 448         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 449         obd->obd_upcall.onu_upcall = cl_ocd_update;
 450
 451         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 452
 453         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 454                           NULL);
 455         if (err == -EBUSY) {
 456                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 457                                    "recovery, of which this client is not a "
 458                                    "part.  Please wait for recovery to "
 459                                    "complete, abort, or time out.\n", dt);
 460                 GOTO(out_md, err);
 461         } else if (err) {
 462                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 463                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 464                 GOTO(out_md, err);
 465         }
 466
 467         sbi->ll_dt_exp->exp_connect_data = *data;
 468
 469         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 470                            LUSTRE_SEQ_METADATA);
 471         if (err) {
 472                 CERROR("%s: Can't init data layer FID infrastructure, "
 473                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 474                 GOTO(out_dt, err);
 475         }
 476
 477         mutex_lock(&sbi->ll_lco.lco_lock);
 478         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 479         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 480         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 481         mutex_unlock(&sbi->ll_lco.lco_lock);
 482
 483         fid_zero(&sbi->ll_root_fid);
 484         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 485         if (err) {
 486                 CERROR("cannot mds_connect: rc = %d\n", err);
 487                 GOTO(out_lock_cn_cb, err);
 488         }
 489         if (!fid_is_sane(&sbi->ll_root_fid)) {
 490                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 491                        sbi->ll_md_exp->exp_obd->obd_name,
 492                        PFID(&sbi->ll_root_fid));
 493                 GOTO(out_lock_cn_cb, err = -EINVAL);
 494         }
 495         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 496
 497         sb->s_op = &lustre_super_operations;
 498 #if THREAD_SIZE >= 8192 /*b=17630*/
 499         sb->s_export_op = &lustre_export_operations;
 500 #endif
 501
 502         /* make root inode
 503          * XXX: move this to after cbd setup? */
 504         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 505                 OBD_MD_FLMODEASIZE;
 506         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 507                 valid |= OBD_MD_FLRMTPERM;
 508         else if (sbi->ll_flags & LL_SBI_ACL)
 509                 valid |= OBD_MD_FLACL;
 510
 511         OBD_ALLOC_PTR(op_data);
 512         if (op_data == NULL)
 513                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 514
 515         op_data->op_fid1 = sbi->ll_root_fid;
 516         op_data->op_mode = 0;
 517         op_data->op_capa1 = oc;
 518         op_data->op_valid = valid;
 519
 520         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 521         if (oc)
 522                 capa_put(oc);
 523         OBD_FREE_PTR(op_data);
 524         if (err) {
 525                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 526                        sbi->ll_md_exp->exp_obd->obd_name, err);
 527                 GOTO(out_lock_cn_cb, err);
 528         }
 529
 530         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 531                                sbi->ll_md_exp, &lmd);
 532         if (err) {
 533                 CERROR("failed to understand root inode md: rc = %d\n", err);
 534                 ptlrpc_req_finished(request);
 535                 GOTO(out_lock_cn_cb, err);
 536         }
 537
 538         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 539         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 540                                             sbi->ll_flags & LL_SBI_32BIT_API),
 541                        &lmd);
 542         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 543         ptlrpc_req_finished(request);
 544
 545         if (IS_ERR(root)) {
 546                 if (lmd.lsm)
 547                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 548 #ifdef CONFIG_FS_POSIX_ACL
 549                 if (lmd.posix_acl) {
 550                         posix_acl_release(lmd.posix_acl);
 551                         lmd.posix_acl = NULL;
 552                 }
 553 #endif
 554                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 555                 root = NULL;
 556                 CERROR("lustre_lite: bad iget4 for root\n");
 557                 GOTO(out_root, err);
 558         }
 559
 560         err = ll_close_thread_start(&sbi->ll_lcq);
 561         if (err) {
 562                 CERROR("cannot start close thread: rc %d\n", err);
 563                 GOTO(out_root, err);
 564         }
 565
 566 #ifdef CONFIG_FS_POSIX_ACL
 567         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 568                 rct_init(&sbi->ll_rct);
 569                 et_init(&sbi->ll_et);
 570         }
 571 #endif
 572
 573         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 574         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 575                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 576                                  NULL);
 577         cl_sb_init(sb);
 578
 579         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 580                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 581                                  &sbi->ll_cache, NULL);
 582
 583         sb->s_root = d_make_root(root);
 584         if (sb->s_root == NULL) {
 585                 CERROR("%s: can't make root dentry\n",
 586                         ll_get_fsname(sb, NULL, 0));
 587                 GOTO(out_root, err = -ENOMEM);
 588         }
 589 #ifdef HAVE_DCACHE_LOCK
 590         sb->s_root->d_op = &ll_d_ops;
 591 #endif
 592
 593         sbi->ll_sdev_orig = sb->s_dev;
 594
 595         /* We set sb->s_dev equal on all lustre clients in order to support
 596          * NFS export clustering.  NFSD requires that the FSID be the same
 597          * on all clients. */
 598         /* s_dev is also used in lt_compare() to compare two fs, but that is
 599          * only a node-local comparison. */
 600         uuid = obd_get_uuid(sbi->ll_md_exp);
 601         if (uuid != NULL)
 602                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 603
 604         if (data != NULL)
 605                 OBD_FREE_PTR(data);
 606         if (osfs != NULL)
 607                 OBD_FREE_PTR(osfs);
 608         if (proc_lustre_fs_root != NULL) {
 609                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 610                                                   dt, md);
 611                 if (err < 0) {
 612                         CERROR("%s: could not register mount in lprocfs: "
 613                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 614                         err = 0;
 615                 }
 616         }
 617
 618         RETURN(err);
 619 out_root:
 620         if (root)
 621                 iput(root);
 622 out_lock_cn_cb:
 623         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 624 out_dt:
 625         obd_disconnect(sbi->ll_dt_exp);
 626         sbi->ll_dt_exp = NULL;
 627         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 628         obd_zombie_barrier();
 629 out_md_fid:
 630         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 631 out_md:
 632         obd_disconnect(sbi->ll_md_exp);
 633         sbi->ll_md_exp = NULL;
 634 out:
 635         if (data != NULL)
 636                 OBD_FREE_PTR(data);
 637         if (osfs != NULL)
 638                 OBD_FREE_PTR(osfs);
 639         return err;
 640 }
 641
 642 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 643 {
 644         int size, rc;
 645
 646         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 647         size = sizeof(int);
 648         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 649                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 650         if (rc)
 651                 CERROR("Get max mdsize error rc %d\n", rc);
 652
 653         RETURN(rc);
 654 }
 655
 656 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 657 {
 658         int size, rc;
 659
 660         size = sizeof(int);
 661         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 662                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 663         if (rc)
 664                 CERROR("Get default mdsize error rc %d\n", rc);
 665
 666         RETURN(rc);
 667 }
 668
 669 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 670 {
 671         int size, rc;
 672
 673         size = sizeof(int);
 674         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 675                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 676         if (rc)
 677                 CERROR("Get max cookiesize error rc %d\n", rc);
 678
 679         RETURN(rc);
 680 }
 681
 682 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 683 {
 684         int size, rc;
 685
 686         size = sizeof(int);
 687         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 688                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 689         if (rc)
 690                 CERROR("Get default cookiesize error rc %d\n", rc);
 691
 692         RETURN(rc);
 693 }
 694
 695 static void ll_dump_inode(struct inode *inode)
 696 {
 697         struct ll_d_hlist_node *tmp;
 698         int dentry_count = 0;
 699
 700         LASSERT(inode != NULL);
 701
 702         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 703                 dentry_count++;
 704
 705         CERROR("%s: inode %p dump: dev=%s fid="DFID
 706                " mode=%o count=%u, %d dentries\n",
 707                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 708                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 709                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 710 }
 711
 712 void lustre_dump_dentry(struct dentry *dentry, int recur)
 713 {
 714         struct list_head *tmp;
 715         int subdirs = 0;
 716
 717         LASSERT(dentry != NULL);
 718
 719         list_for_each(tmp, &dentry->d_subdirs)
 720                 subdirs++;
 721
 722         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 723                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 724                dentry->d_name.len, dentry->d_name.name,
 725                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 726                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 727                dentry->d_flags, dentry->d_fsdata, subdirs);
 728         if (dentry->d_inode != NULL)
 729                 ll_dump_inode(dentry->d_inode);
 730
 731         if (recur == 0)
 732                 return;
 733
 734         list_for_each(tmp, &dentry->d_subdirs) {
 735                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 736                 lustre_dump_dentry(d, recur - 1);
 737         }
 738 }
 739
 740 static void client_common_put_super(struct super_block *sb)
 741 {
 742         struct ll_sb_info *sbi = ll_s2sbi(sb);
 743         ENTRY;
 744
 745 #ifdef CONFIG_FS_POSIX_ACL
 746         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 747                 et_fini(&sbi->ll_et);
 748                 rct_fini(&sbi->ll_rct);
 749         }
 750 #endif
 751
 752         ll_close_thread_shutdown(sbi->ll_lcq);
 753
 754         cl_sb_fini(sb);
 755
 756         list_del(&sbi->ll_conn_chain);
 757
 758         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 759         obd_disconnect(sbi->ll_dt_exp);
 760         sbi->ll_dt_exp = NULL;
 761         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 762          * see LU-2543. */
 763         obd_zombie_barrier();
 764
 765         lprocfs_unregister_mountpoint(sbi);
 766
 767         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 768         obd_disconnect(sbi->ll_md_exp);
 769         sbi->ll_md_exp = NULL;
 770
 771         EXIT;
 772 }
 773
 774 void ll_kill_super(struct super_block *sb)
 775 {
 776         struct ll_sb_info *sbi;
 777
 778         ENTRY;
 779
 780         /* not init sb ?*/
 781         if (!(sb->s_flags & MS_ACTIVE))
 782                 return;
 783
 784         sbi = ll_s2sbi(sb);
 785         /* we need restore s_dev from changed for clustred NFS before put_super
 786          * because new kernels have cached s_dev and change sb->s_dev in
 787          * put_super not affected real removing devices */
 788         if (sbi) {
 789                 sb->s_dev = sbi->ll_sdev_orig;
 790                 sbi->ll_umounting = 1;
 791         }
 792         EXIT;
 793 }
 794
 795 static inline int ll_set_opt(const char *opt, char *data, int fl)
 796 {
 797         if (strncmp(opt, data, strlen(opt)) != 0)
 798                 return(0);
 799         else
 800                 return(fl);
 801 }
 802
 803 /* non-client-specific mount options are parsed in lmd_parse */
 804 static int ll_options(char *options, int *flags)
 805 {
 806         int tmp;
 807         char *s1 = options, *s2;
 808         ENTRY;
 809
 810         if (!options)
 811                 RETURN(0);
 812
 813         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 814
 815         while (*s1) {
 816                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 817                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 818                 if (tmp) {
 819                         *flags |= tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 828                 if (tmp) {
 829                         *flags |= tmp;
 830                         goto next;
 831                 }
 832                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 833                 if (tmp) {
 834                         *flags &= ~tmp;
 835                         goto next;
 836                 }
 837                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 838                 if (tmp) {
 839                         *flags |= tmp;
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 843                 if (tmp) {
 844                         *flags &= ~tmp;
 845                         goto next;
 846                 }
 847 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 52, 0)
 848                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 849                 if (tmp) {
 850                         /* Ignore deprecated mount option.  The client will
 851                          * always try to mount with ACL support, whether this
 852                          * is used depends on whether server supports it. */
 853                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 854                                                   "mount option 'acl'.\n");
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 858                 if (tmp) {
 859                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 860                                                   "mount option 'noacl'.\n");
 861                         goto next;
 862                 }
 863 #else
 864 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 865 #endif
 866                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 872                 if (tmp) {
 873                         *flags |= tmp;
 874                         goto next;
 875                 }
 876                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 877                 if (tmp) {
 878                         *flags &= ~tmp;
 879                         goto next;
 880                 }
 881
 882                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 883                 if (tmp) {
 884                         *flags |= tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 888                 if (tmp) {
 889                         *flags &= ~tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 893                 if (tmp) {
 894                         *flags |= tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 898                 if (tmp) {
 899                         *flags &= ~tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 903                 if (tmp) {
 904                         *flags |= tmp;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 908                 if (tmp) {
 909                         *flags &= ~tmp;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 913                 if (tmp) {
 914                         *flags |= tmp;
 915                         goto next;
 916                 }
 917                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 918                 if (tmp) {
 919                         *flags |= tmp;
 920                         goto next;
 921                 }
 922                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 923                 if (tmp) {
 924                         *flags |= tmp;
 925                         goto next;
 926                 }
 927                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 928                 if (tmp) {
 929                         *flags &= ~tmp;
 930                         goto next;
 931                 }
 932                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 933                                    s1);
 934                 RETURN(-EINVAL);
 935
 936 next:
 937                 /* Find next opt */
 938                 s2 = strchr(s1, ',');
 939                 if (s2 == NULL)
 940                         break;
 941                 s1 = s2 + 1;
 942         }
 943         RETURN(0);
 944 }
 945
 946 void ll_lli_init(struct ll_inode_info *lli)
 947 {
 948         lli->lli_inode_magic = LLI_INODE_MAGIC;
 949         lli->lli_flags = 0;
 950         lli->lli_ioepoch = 0;
 951         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 952         spin_lock_init(&lli->lli_lock);
 953         lli->lli_posix_acl = NULL;
 954         lli->lli_remote_perms = NULL;
 955         mutex_init(&lli->lli_rmtperm_mutex);
 956         /* Do not set lli_fid, it has been initialized already. */
 957         fid_zero(&lli->lli_pfid);
 958         INIT_LIST_HEAD(&lli->lli_close_list);
 959         INIT_LIST_HEAD(&lli->lli_oss_capas);
 960         atomic_set(&lli->lli_open_count, 0);
 961         lli->lli_mds_capa = NULL;
 962         lli->lli_rmtperm_time = 0;
 963         lli->lli_pending_och = NULL;
 964         lli->lli_mds_read_och = NULL;
 965         lli->lli_mds_write_och = NULL;
 966         lli->lli_mds_exec_och = NULL;
 967         lli->lli_open_fd_read_count = 0;
 968         lli->lli_open_fd_write_count = 0;
 969         lli->lli_open_fd_exec_count = 0;
 970         mutex_init(&lli->lli_och_mutex);
 971         spin_lock_init(&lli->lli_agl_lock);
 972         lli->lli_has_smd = false;
 973         spin_lock_init(&lli->lli_layout_lock);
 974         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 975         lli->lli_clob = NULL;
 976
 977         init_rwsem(&lli->lli_xattrs_list_rwsem);
 978         mutex_init(&lli->lli_xattrs_enq_lock);
 979
 980         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 981         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 982                 mutex_init(&lli->lli_readdir_mutex);
 983                 lli->lli_opendir_key = NULL;
 984                 lli->lli_sai = NULL;
 985                 spin_lock_init(&lli->lli_sa_lock);
 986                 lli->lli_opendir_pid = 0;
 987         } else {
 988                 mutex_init(&lli->lli_size_mutex);
 989                 lli->lli_symlink_name = NULL;
 990                 init_rwsem(&lli->lli_trunc_sem);
 991                 mutex_init(&lli->lli_write_mutex);
 992                 init_rwsem(&lli->lli_glimpse_sem);
 993                 lli->lli_glimpse_time = 0;
 994                 INIT_LIST_HEAD(&lli->lli_agl_list);
 995                 lli->lli_agl_index = 0;
 996                 lli->lli_async_rc = 0;
 997         }
 998         mutex_init(&lli->lli_layout_mutex);
 999 }
1000
1001 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1002 {
1003         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1004
1005         bdi->name = "lustre";
1006         return bdi_register(bdi, NULL, "lustre-%d",
1007                             atomic_inc_return(&ll_bdi_num));
1008 }
1009
1010 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1011 {
1012         struct lustre_profile *lprof = NULL;
1013         struct lustre_sb_info *lsi = s2lsi(sb);
1014         struct ll_sb_info *sbi;
1015         char  *dt = NULL, *md = NULL;
1016         char  *profilenm = get_profile_name(sb);
1017         struct config_llog_instance *cfg;
1018         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1019         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1020         int    err;
1021         ENTRY;
1022
1023         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1024
1025         OBD_ALLOC_PTR(cfg);
1026         if (cfg == NULL)
1027                 RETURN(-ENOMEM);
1028
1029         try_module_get(THIS_MODULE);
1030
1031         /* client additional sb info */
1032         lsi->lsi_llsbi = sbi = ll_init_sbi();
1033         if (!sbi) {
1034                 module_put(THIS_MODULE);
1035                 OBD_FREE_PTR(cfg);
1036                 RETURN(-ENOMEM);
1037         }
1038
1039         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1040         if (err)
1041                 GOTO(out_free, err);
1042
1043         err = bdi_init(&lsi->lsi_bdi);
1044         if (err)
1045                 GOTO(out_free, err);
1046         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1047         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1048         err = ll_bdi_register(&lsi->lsi_bdi);
1049         if (err)
1050                 GOTO(out_free, err);
1051
1052         sb->s_bdi = &lsi->lsi_bdi;
1053 #ifndef HAVE_DCACHE_LOCK
1054         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1055         sb->s_d_op = &ll_d_ops;
1056 #endif
1057
1058         /* Generate a string unique to this super, in case some joker tries
1059            to mount the same fs at two mount points.
1060            Use the address of the super itself.*/
1061         cfg->cfg_instance = sb;
1062         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1063         cfg->cfg_callback = class_config_llog_handler;
1064         /* set up client obds */
1065         err = lustre_process_log(sb, profilenm, cfg);
1066         if (err < 0)
1067                 GOTO(out_free, err);
1068
1069         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1070         lprof = class_get_profile(profilenm);
1071         if (lprof == NULL) {
1072                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1073                                    " read from the MGS.  Does that filesystem "
1074                                    "exist?\n", profilenm);
1075                 GOTO(out_free, err = -EINVAL);
1076         }
1077         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1078                lprof->lp_md, lprof->lp_dt);
1079
1080         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1081         if (!dt)
1082                 GOTO(out_free, err = -ENOMEM);
1083         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1084
1085         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1086         if (!md)
1087                 GOTO(out_free, err = -ENOMEM);
1088         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1089
1090         /* connections, registrations, sb setup */
1091         err = client_common_fill_super(sb, md, dt, mnt);
1092
1093 out_free:
1094         if (md)
1095                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1096         if (dt)
1097                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1098         if (err)
1099                 ll_put_super(sb);
1100         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1101                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1102
1103         OBD_FREE_PTR(cfg);
1104         RETURN(err);
1105 } /* ll_fill_super */
1106
1107 void ll_put_super(struct super_block *sb)
1108 {
1109         struct config_llog_instance cfg, params_cfg;
1110         struct obd_device *obd;
1111         struct lustre_sb_info *lsi = s2lsi(sb);
1112         struct ll_sb_info *sbi = ll_s2sbi(sb);
1113         char *profilenm = get_profile_name(sb);
1114         int ccc_count, next, force = 1, rc = 0;
1115         ENTRY;
1116
1117         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1118
1119         ll_print_capa_stat(sbi);
1120
1121         cfg.cfg_instance = sb;
1122         lustre_end_log(sb, profilenm, &cfg);
1123
1124         params_cfg.cfg_instance = sb;
1125         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1126
1127         if (sbi->ll_md_exp) {
1128                 obd = class_exp2obd(sbi->ll_md_exp);
1129                 if (obd)
1130                         force = obd->obd_force;
1131         }
1132
1133         /* Wait for unstable pages to be committed to stable storage */
1134         if (force == 0) {
1135                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1136                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1137                         atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1138                         &lwi);
1139         }
1140
1141         ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1142         if (force == 0 && rc != -EINTR)
1143                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1144
1145
1146         /* We need to set force before the lov_disconnect in
1147            lustre_common_put_super, since l_d cleans up osc's as well. */
1148         if (force) {
1149                 next = 0;
1150                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1151                                                      &next)) != NULL) {
1152                         obd->obd_force = force;
1153                 }
1154         }
1155
1156         if (sbi->ll_lcq) {
1157                 /* Only if client_common_fill_super succeeded */
1158                 client_common_put_super(sb);
1159         }
1160
1161         next = 0;
1162         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1163                 class_manual_cleanup(obd);
1164         }
1165
1166         if (sbi->ll_flags & LL_SBI_VERBOSE)
1167                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1168
1169         if (profilenm)
1170                 class_del_profile(profilenm);
1171
1172         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1173                 bdi_destroy(&lsi->lsi_bdi);
1174                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1175         }
1176
1177         ll_free_sbi(sb);
1178         lsi->lsi_llsbi = NULL;
1179
1180         lustre_common_put_super(sb);
1181
1182         cl_env_cache_purge(~0);
1183
1184         module_put(THIS_MODULE);
1185
1186         EXIT;
1187 } /* client_put_super */
1188
1189 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1190 {
1191         struct inode *inode = NULL;
1192
1193         /* NOTE: we depend on atomic igrab() -bzzz */
1194         lock_res_and_lock(lock);
1195         if (lock->l_resource->lr_lvb_inode) {
1196                 struct ll_inode_info * lli;
1197                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1198                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1199                         inode = igrab(lock->l_resource->lr_lvb_inode);
1200                 } else {
1201                         inode = lock->l_resource->lr_lvb_inode;
1202                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1203                                          D_WARNING, lock, "lr_lvb_inode %p is "
1204                                          "bogus: magic %08x",
1205                                          lock->l_resource->lr_lvb_inode,
1206                                          lli->lli_inode_magic);
1207                         inode = NULL;
1208                 }
1209         }
1210         unlock_res_and_lock(lock);
1211         return inode;
1212 }
1213
1214 static void ll_dir_clear_lsm_md(struct inode *inode)
1215 {
1216         struct ll_inode_info *lli = ll_i2info(inode);
1217
1218         LASSERT(S_ISDIR(inode->i_mode));
1219
1220         if (lli->lli_lsm_md != NULL) {
1221                 lmv_free_memmd(lli->lli_lsm_md);
1222                 lli->lli_lsm_md = NULL;
1223         }
1224 }
1225
1226 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1227                                       const struct lu_fid *fid,
1228                                       struct lustre_md *md)
1229 {
1230         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1231         struct mdt_body         *body = md->body;
1232         struct inode            *inode;
1233         ino_t                   ino;
1234         ENTRY;
1235
1236         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1237         inode = iget_locked(sb, ino);
1238         if (inode == NULL) {
1239                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1240                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1241                 RETURN(ERR_PTR(-ENOENT));
1242         }
1243
1244         if (inode->i_state & I_NEW) {
1245                 struct ll_inode_info *lli = ll_i2info(inode);
1246                 struct lmv_stripe_md *lsm = md->lmv;
1247
1248                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1249                                 (body->mbo_mode & S_IFMT);
1250                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1251                          PFID(fid));
1252
1253                 LTIME_S(inode->i_mtime) = 0;
1254                 LTIME_S(inode->i_atime) = 0;
1255                 LTIME_S(inode->i_ctime) = 0;
1256                 inode->i_rdev = 0;
1257
1258                 /* initializing backing dev info. */
1259                 inode->i_mapping->backing_dev_info =
1260                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1261                 inode->i_op = &ll_dir_inode_operations;
1262                 inode->i_fop = &ll_dir_operations;
1263                 lli->lli_fid = *fid;
1264                 ll_lli_init(lli);
1265
1266                 LASSERT(lsm != NULL);
1267                 /* master object FID */
1268                 lli->lli_pfid = body->mbo_fid1;
1269                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1270                        lli, PFID(fid), PFID(&lli->lli_pfid));
1271                 unlock_new_inode(inode);
1272         }
1273
1274         RETURN(inode);
1275 }
1276
1277 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1278 {
1279         struct lu_fid *fid;
1280         struct lmv_stripe_md *lsm = md->lmv;
1281         int i;
1282
1283         LASSERT(lsm != NULL);
1284         /* XXX sigh, this lsm_root initialization should be in
1285          * LMV layer, but it needs ll_iget right now, so we
1286          * put this here right now. */
1287         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1288                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1289                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1290                 /* Unfortunately ll_iget will call ll_update_inode,
1291                  * where the initialization of slave inode is slightly
1292                  * different, so it reset lsm_md to NULL to avoid
1293                  * initializing lsm for slave inode. */
1294                 /* For migrating inode, master stripe and master object will
1295                  * be same, so we only need assign this inode */
1296                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1297                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1298                 else
1299                         lsm->lsm_md_oinfo[i].lmo_root =
1300                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1301
1302                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1303                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1304
1305                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1306                         return rc;
1307                 }
1308         }
1309
1310         /* Here is where the lsm is being initialized(fill lmo_info) after
1311          * client retrieve MD stripe information from MDT. */
1312         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1313                                 ll_md_blocking_ast);
1314 }
1315
1316 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1317                                 const struct lmv_stripe_md *lsm_md2)
1318 {
1319         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1320                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1321                lsm_md1->lsm_md_master_mdt_index ==
1322                                         lsm_md2->lsm_md_master_mdt_index &&
1323                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1324                lsm_md1->lsm_md_layout_version ==
1325                                         lsm_md2->lsm_md_layout_version &&
1326                strcmp(lsm_md1->lsm_md_pool_name,
1327                       lsm_md2->lsm_md_pool_name) == 0;
1328 }
1329
1330 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1331 {
1332         struct ll_inode_info *lli = ll_i2info(inode);
1333         struct lmv_stripe_md *lsm = md->lmv;
1334         int     rc;
1335         ENTRY;
1336
1337         LASSERT(S_ISDIR(inode->i_mode));
1338         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1339                PFID(ll_inode2fid(inode)));
1340
1341         /* no striped information from request. */
1342         if (lsm == NULL) {
1343                 if (lli->lli_lsm_md == NULL) {
1344                         RETURN(0);
1345                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1346                                                 LMV_HASH_FLAG_MIGRATION) {
1347                         /* migration is done, the temporay MIGRATE layout has
1348                          * been removed */
1349                         CDEBUG(D_INODE, DFID" finish migration.\n",
1350                                PFID(ll_inode2fid(inode)));
1351                         lmv_free_memmd(lli->lli_lsm_md);
1352                         lli->lli_lsm_md = NULL;
1353                         RETURN(0);
1354                 } else {
1355                         /* The lustre_md from req does not include stripeEA,
1356                          * see ll_md_setattr */
1357                         RETURN(0);
1358                 }
1359         }
1360
1361         /* set the directory layout */
1362         if (lli->lli_lsm_md == NULL) {
1363
1364                 rc = ll_init_lsm_md(inode, md);
1365                 if (rc != 0)
1366                         RETURN(rc);
1367
1368                 lli->lli_lsm_md = lsm;
1369                 /* set lsm_md to NULL, so the following free lustre_md
1370                  * will not free this lsm */
1371                 md->lmv = NULL;
1372                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1373                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1374                 RETURN(0);
1375         }
1376
1377         /* Compare the old and new stripe information */
1378         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1379                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1380                 int                     idx;
1381
1382                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1383                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1384                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1385                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1386                        inode, lsm, old_lsm,
1387                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1388                        lsm->lsm_md_stripe_count,
1389                        old_lsm->lsm_md_stripe_count,
1390                        lsm->lsm_md_master_mdt_index,
1391                        old_lsm->lsm_md_master_mdt_index,
1392                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1393                        lsm->lsm_md_layout_version,
1394                        old_lsm->lsm_md_layout_version,
1395                        lsm->lsm_md_pool_name,
1396                        old_lsm->lsm_md_pool_name);
1397
1398                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1399                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1400                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1401                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1402                 }
1403
1404                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1405                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1406                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1407                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1408                 }
1409
1410                 RETURN(-EIO);
1411         }
1412
1413         rc = md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1414                               md->body, ll_md_blocking_ast);
1415
1416         RETURN(rc);
1417 }
1418
1419 void ll_clear_inode(struct inode *inode)
1420 {
1421         struct ll_inode_info *lli = ll_i2info(inode);
1422         struct ll_sb_info *sbi = ll_i2sbi(inode);
1423         ENTRY;
1424
1425         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1426                PFID(ll_inode2fid(inode)), inode);
1427
1428         if (S_ISDIR(inode->i_mode)) {
1429                 /* these should have been cleared in ll_file_release */
1430                 LASSERT(lli->lli_opendir_key == NULL);
1431                 LASSERT(lli->lli_sai == NULL);
1432                 LASSERT(lli->lli_opendir_pid == 0);
1433         }
1434
1435         spin_lock(&lli->lli_lock);
1436         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1437         spin_unlock(&lli->lli_lock);
1438         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1439
1440         LASSERT(!lli->lli_open_fd_write_count);
1441         LASSERT(!lli->lli_open_fd_read_count);
1442         LASSERT(!lli->lli_open_fd_exec_count);
1443
1444         if (lli->lli_mds_write_och)
1445                 ll_md_real_close(inode, FMODE_WRITE);
1446         if (lli->lli_mds_exec_och)
1447                 ll_md_real_close(inode, FMODE_EXEC);
1448         if (lli->lli_mds_read_och)
1449                 ll_md_real_close(inode, FMODE_READ);
1450
1451         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1452                 OBD_FREE(lli->lli_symlink_name,
1453                          strlen(lli->lli_symlink_name) + 1);
1454                 lli->lli_symlink_name = NULL;
1455         }
1456
1457         ll_xattr_cache_destroy(inode);
1458
1459         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1460                 LASSERT(lli->lli_posix_acl == NULL);
1461                 if (lli->lli_remote_perms) {
1462                         free_rmtperm_hash(lli->lli_remote_perms);
1463                         lli->lli_remote_perms = NULL;
1464                 }
1465         }
1466 #ifdef CONFIG_FS_POSIX_ACL
1467         else if (lli->lli_posix_acl) {
1468                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1469                 LASSERT(lli->lli_remote_perms == NULL);
1470                 posix_acl_release(lli->lli_posix_acl);
1471                 lli->lli_posix_acl = NULL;
1472         }
1473 #endif
1474         lli->lli_inode_magic = LLI_INODE_DEAD;
1475
1476         ll_clear_inode_capas(inode);
1477         if (S_ISDIR(inode->i_mode))
1478                 ll_dir_clear_lsm_md(inode);
1479         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1480                 LASSERT(list_empty(&lli->lli_agl_list));
1481
1482         /*
1483          * XXX This has to be done before lsm is freed below, because
1484          * cl_object still uses inode lsm.
1485          */
1486         cl_inode_fini(inode);
1487         lli->lli_has_smd = false;
1488
1489         EXIT;
1490 }
1491
1492 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1493                   struct md_open_data **mod)
1494 {
1495         struct lustre_md md;
1496         struct inode *inode = dentry->d_inode;
1497         struct ll_sb_info *sbi = ll_i2sbi(inode);
1498         struct ptlrpc_request *request = NULL;
1499         int rc, ia_valid;
1500         ENTRY;
1501
1502         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1503                                      LUSTRE_OPC_ANY, NULL);
1504         if (IS_ERR(op_data))
1505                 RETURN(PTR_ERR(op_data));
1506
1507         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1508                         &request, mod);
1509         if (rc) {
1510                 ptlrpc_req_finished(request);
1511                 if (rc == -ENOENT) {
1512                         clear_nlink(inode);
1513                         /* Unlinked special device node? Or just a race?
1514                          * Pretend we done everything. */
1515                         if (!S_ISREG(inode->i_mode) &&
1516                             !S_ISDIR(inode->i_mode)) {
1517                                 ia_valid = op_data->op_attr.ia_valid;
1518                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1519                                 rc = simple_setattr(dentry, &op_data->op_attr);
1520                                 op_data->op_attr.ia_valid = ia_valid;
1521                         }
1522                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1523                         CERROR("md_setattr fails: rc = %d\n", rc);
1524                 }
1525                 RETURN(rc);
1526         }
1527
1528         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1529                               sbi->ll_md_exp, &md);
1530         if (rc) {
1531                 ptlrpc_req_finished(request);
1532                 RETURN(rc);
1533         }
1534
1535         ia_valid = op_data->op_attr.ia_valid;
1536         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1537          * cache is not cleared yet. */
1538         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1539         rc = simple_setattr(dentry, &op_data->op_attr);
1540         op_data->op_attr.ia_valid = ia_valid;
1541
1542         /* Extract epoch data if obtained. */
1543         op_data->op_handle = md.body->mbo_handle;
1544         op_data->op_ioepoch = md.body->mbo_ioepoch;
1545
1546         rc = ll_update_inode(inode, &md);
1547         ptlrpc_req_finished(request);
1548
1549         RETURN(rc);
1550 }
1551
1552 /* Close IO epoch and send Size-on-MDS attribute update. */
1553 static int ll_setattr_done_writing(struct inode *inode,
1554                                    struct md_op_data *op_data,
1555                                    struct md_open_data *mod)
1556 {
1557         struct ll_inode_info *lli = ll_i2info(inode);
1558         int rc = 0;
1559         ENTRY;
1560
1561         LASSERT(op_data != NULL);
1562         if (!S_ISREG(inode->i_mode))
1563                 RETURN(0);
1564
1565         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1566                op_data->op_ioepoch, PFID(&lli->lli_fid));
1567
1568         op_data->op_flags = MF_EPOCH_CLOSE;
1569         ll_done_writing_attr(inode, op_data);
1570         ll_pack_inode2opdata(inode, op_data, NULL);
1571
1572         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1573         if (rc == -EAGAIN) {
1574                 /* MDS has instructed us to obtain Size-on-MDS attribute
1575                  * from OSTs and send setattr to back to MDS. */
1576                 rc = ll_som_update(inode, op_data);
1577         } else if (rc) {
1578                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1579                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1580                        PFID(ll_inode2fid(inode)), rc);
1581         }
1582         RETURN(rc);
1583 }
1584
1585 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1586 {
1587         struct obd_capa *capa;
1588         int rc;
1589
1590         if (attr->ia_valid & ATTR_SIZE)
1591                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1592         else
1593                 capa = ll_mdscapa_get(inode);
1594
1595         rc = cl_setattr_ost(inode, attr, capa);
1596
1597         if (attr->ia_valid & ATTR_SIZE)
1598                 ll_truncate_free_capa(capa);
1599         else
1600                 capa_put(capa);
1601
1602         return rc;
1603 }
1604
1605 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1606  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1607  * keep these values until such a time that objects are allocated for it.
1608  * We do the MDS operations first, as it is checking permissions for us.
1609  * We don't to the MDS RPC if there is nothing that we want to store there,
1610  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1611  * going to do an RPC anyways.
1612  *
1613  * If we are doing a truncate, we will send the mtime and ctime updates
1614  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1615  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1616  * at the same time.
1617  *
1618  * In case of HSMimport, we only set attr on MDS.
1619  */
1620 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1621 {
1622         struct inode *inode = dentry->d_inode;
1623         struct ll_inode_info *lli = ll_i2info(inode);
1624         struct md_op_data *op_data = NULL;
1625         struct md_open_data *mod = NULL;
1626         bool file_is_released = false;
1627         int rc = 0, rc1 = 0;
1628         ENTRY;
1629
1630         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1631                "valid %x, hsm_import %d\n",
1632                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1633                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1634                hsm_import);
1635
1636         if (attr->ia_valid & ATTR_SIZE) {
1637                 /* Check new size against VFS/VM file size limit and rlimit */
1638                 rc = inode_newsize_ok(inode, attr->ia_size);
1639                 if (rc)
1640                         RETURN(rc);
1641
1642                 /* The maximum Lustre file size is variable, based on the
1643                  * OST maximum object size and number of stripes.  This
1644                  * needs another check in addition to the VFS check above. */
1645                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1646                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1647                                PFID(&lli->lli_fid), attr->ia_size,
1648                                ll_file_maxbytes(inode));
1649                         RETURN(-EFBIG);
1650                 }
1651
1652                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1653         }
1654
1655         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1656         if (attr->ia_valid & TIMES_SET_FLAGS) {
1657                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1658                     !cfs_capable(CFS_CAP_FOWNER))
1659                         RETURN(-EPERM);
1660         }
1661
1662         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1663         if (attr->ia_valid & ATTR_CTIME) {
1664                 attr->ia_ctime = CFS_CURRENT_TIME;
1665                 attr->ia_valid |= ATTR_CTIME_SET;
1666         }
1667         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1668             (attr->ia_valid & ATTR_ATIME)) {
1669                 attr->ia_atime = CFS_CURRENT_TIME;
1670                 attr->ia_valid |= ATTR_ATIME_SET;
1671         }
1672         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1673             (attr->ia_valid & ATTR_MTIME)) {
1674                 attr->ia_mtime = CFS_CURRENT_TIME;
1675                 attr->ia_valid |= ATTR_MTIME_SET;
1676         }
1677
1678         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1679                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1680                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1681                        cfs_time_current_sec());
1682
1683         /* We always do an MDS RPC, even if we're only changing the size;
1684          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1685
1686         OBD_ALLOC_PTR(op_data);
1687         if (op_data == NULL)
1688                 RETURN(-ENOMEM);
1689
1690         if (!S_ISDIR(inode->i_mode)) {
1691                 if (attr->ia_valid & ATTR_SIZE)
1692                         inode_dio_write_done(inode);
1693                 mutex_unlock(&inode->i_mutex);
1694         }
1695
1696         /* truncate on a released file must failed with -ENODATA,
1697          * so size must not be set on MDS for released file
1698          * but other attributes must be set
1699          */
1700         if (S_ISREG(inode->i_mode)) {
1701                 struct lov_stripe_md *lsm;
1702                 __u32 gen;
1703
1704                 ll_layout_refresh(inode, &gen);
1705                 lsm = ccc_inode_lsm_get(inode);
1706                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1707                         file_is_released = true;
1708                 ccc_inode_lsm_put(inode, lsm);
1709
1710                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1711                         if (file_is_released) {
1712                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1713                                 if (rc < 0)
1714                                         GOTO(out, rc);
1715
1716                                 file_is_released = false;
1717                                 ll_layout_refresh(inode, &gen);
1718                         }
1719
1720                         /* If we are changing file size, file content is
1721                          * modified, flag it. */
1722                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1723                         spin_lock(&lli->lli_lock);
1724                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1725                         spin_unlock(&lli->lli_lock);
1726                         op_data->op_bias |= MDS_DATA_MODIFIED;
1727                 }
1728         }
1729
1730         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1731
1732         /* Open epoch for truncate. */
1733         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1734             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1735                 op_data->op_flags = MF_EPOCH_OPEN;
1736
1737         rc = ll_md_setattr(dentry, op_data, &mod);
1738         if (rc)
1739                 GOTO(out, rc);
1740
1741         /* RPC to MDT is sent, cancel data modification flag */
1742         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1743                 spin_lock(&lli->lli_lock);
1744                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1745                 spin_unlock(&lli->lli_lock);
1746         }
1747
1748         ll_ioepoch_open(lli, op_data->op_ioepoch);
1749         if (!S_ISREG(inode->i_mode) || file_is_released)
1750                 GOTO(out, rc = 0);
1751
1752         if (attr->ia_valid & (ATTR_SIZE |
1753                               ATTR_ATIME | ATTR_ATIME_SET |
1754                               ATTR_MTIME | ATTR_MTIME_SET)) {
1755                 /* For truncate and utimes sending attributes to OSTs, setting
1756                  * mtime/atime to the past will be performed under PW [0:EOF]
1757                  * extent lock (new_size:EOF for truncate).  It may seem
1758                  * excessive to send mtime/atime updates to OSTs when not
1759                  * setting times to past, but it is necessary due to possible
1760                  * time de-synchronization between MDT inode and OST objects */
1761                 if (attr->ia_valid & ATTR_SIZE)
1762                         down_write(&lli->lli_trunc_sem);
1763                 rc = ll_setattr_ost(inode, attr);
1764                 if (attr->ia_valid & ATTR_SIZE)
1765                         up_write(&lli->lli_trunc_sem);
1766         }
1767         EXIT;
1768 out:
1769         if (op_data) {
1770                 if (op_data->op_ioepoch) {
1771                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1772                         if (!rc)
1773                                 rc = rc1;
1774                 }
1775                 ll_finish_md_op_data(op_data);
1776         }
1777         if (!S_ISDIR(inode->i_mode)) {
1778                 mutex_lock(&inode->i_mutex);
1779                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1780                         inode_dio_wait(inode);
1781         }
1782
1783         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1784                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1785
1786         return rc;
1787 }
1788
1789 int ll_setattr(struct dentry *de, struct iattr *attr)
1790 {
1791         int mode = de->d_inode->i_mode;
1792
1793         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1794                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1795                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1796
1797         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1798                                (ATTR_SIZE|ATTR_MODE)) &&
1799             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1800              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1801               !(attr->ia_mode & S_ISGID))))
1802                 attr->ia_valid |= ATTR_FORCE;
1803
1804         if ((attr->ia_valid & ATTR_MODE) &&
1805             (mode & S_ISUID) &&
1806             !(attr->ia_mode & S_ISUID) &&
1807             !(attr->ia_valid & ATTR_KILL_SUID))
1808                 attr->ia_valid |= ATTR_KILL_SUID;
1809
1810         if ((attr->ia_valid & ATTR_MODE) &&
1811             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1812             !(attr->ia_mode & S_ISGID) &&
1813             !(attr->ia_valid & ATTR_KILL_SGID))
1814                 attr->ia_valid |= ATTR_KILL_SGID;
1815
1816         return ll_setattr_raw(de, attr, false);
1817 }
1818
1819 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1820                        __u64 max_age, __u32 flags)
1821 {
1822         struct ll_sb_info *sbi = ll_s2sbi(sb);
1823         struct obd_statfs obd_osfs;
1824         int rc;
1825         ENTRY;
1826
1827         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1828         if (rc) {
1829                 CERROR("md_statfs fails: rc = %d\n", rc);
1830                 RETURN(rc);
1831         }
1832
1833         osfs->os_type = sb->s_magic;
1834
1835         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1836                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1837
1838         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1839                 flags |= OBD_STATFS_NODELAY;
1840
1841         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1842         if (rc) {
1843                 CERROR("obd_statfs fails: rc = %d\n", rc);
1844                 RETURN(rc);
1845         }
1846
1847         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1848                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1849                obd_osfs.os_files);
1850
1851         osfs->os_bsize = obd_osfs.os_bsize;
1852         osfs->os_blocks = obd_osfs.os_blocks;
1853         osfs->os_bfree = obd_osfs.os_bfree;
1854         osfs->os_bavail = obd_osfs.os_bavail;
1855
1856         /* If we don't have as many objects free on the OST as inodes
1857          * on the MDS, we reduce the total number of inodes to
1858          * compensate, so that the "inodes in use" number is correct.
1859          */
1860         if (obd_osfs.os_ffree < osfs->os_ffree) {
1861                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1862                         obd_osfs.os_ffree;
1863                 osfs->os_ffree = obd_osfs.os_ffree;
1864         }
1865
1866         RETURN(rc);
1867 }
1868 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1869 {
1870         struct super_block *sb = de->d_sb;
1871         struct obd_statfs osfs;
1872         __u64 fsid = huge_encode_dev(sb->s_dev);
1873         int rc;
1874
1875         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1876         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1877
1878         /* Some amount of caching on the client is allowed */
1879         rc = ll_statfs_internal(sb, &osfs,
1880                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1881                                 0);
1882         if (rc)
1883                 return rc;
1884
1885         statfs_unpack(sfs, &osfs);
1886
1887         /* We need to downshift for all 32-bit kernels, because we can't
1888          * tell if the kernel is being called via sys_statfs64() or not.
1889          * Stop before overflowing f_bsize - in which case it is better
1890          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1891         if (sizeof(long) < 8) {
1892                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1893                         sfs->f_bsize <<= 1;
1894
1895                         osfs.os_blocks >>= 1;
1896                         osfs.os_bfree >>= 1;
1897                         osfs.os_bavail >>= 1;
1898                 }
1899         }
1900
1901         sfs->f_blocks = osfs.os_blocks;
1902         sfs->f_bfree = osfs.os_bfree;
1903         sfs->f_bavail = osfs.os_bavail;
1904         sfs->f_fsid.val[0] = (__u32)fsid;
1905         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1906         return 0;
1907 }
1908
1909 void ll_inode_size_lock(struct inode *inode)
1910 {
1911         struct ll_inode_info *lli;
1912
1913         LASSERT(!S_ISDIR(inode->i_mode));
1914
1915         lli = ll_i2info(inode);
1916         mutex_lock(&lli->lli_size_mutex);
1917 }
1918
1919 void ll_inode_size_unlock(struct inode *inode)
1920 {
1921         struct ll_inode_info *lli;
1922
1923         lli = ll_i2info(inode);
1924         mutex_unlock(&lli->lli_size_mutex);
1925 }
1926
1927 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1928 {
1929         struct ll_inode_info *lli = ll_i2info(inode);
1930         struct mdt_body *body = md->body;
1931         struct lov_stripe_md *lsm = md->lsm;
1932         struct ll_sb_info *sbi = ll_i2sbi(inode);
1933
1934         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1935         if (lsm != NULL) {
1936                 if (!lli->lli_has_smd &&
1937                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1938                         cl_file_inode_init(inode, md);
1939
1940                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1941                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1942                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1943         }
1944
1945         if (S_ISDIR(inode->i_mode)) {
1946                 int     rc;
1947
1948                 rc = ll_update_lsm_md(inode, md);
1949                 if (rc != 0)
1950                         return rc;
1951         }
1952
1953         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1954                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1955                         ll_update_remote_perm(inode, md->remote_perm);
1956         }
1957 #ifdef CONFIG_FS_POSIX_ACL
1958         else if (body->mbo_valid & OBD_MD_FLACL) {
1959                 spin_lock(&lli->lli_lock);
1960                 if (lli->lli_posix_acl)
1961                         posix_acl_release(lli->lli_posix_acl);
1962                 lli->lli_posix_acl = md->posix_acl;
1963                 spin_unlock(&lli->lli_lock);
1964         }
1965 #endif
1966         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1967                                         sbi->ll_flags & LL_SBI_32BIT_API);
1968         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1969
1970         if (body->mbo_valid & OBD_MD_FLATIME) {
1971                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1972                         LTIME_S(inode->i_atime) = body->mbo_atime;
1973                 lli->lli_lvb.lvb_atime = body->mbo_atime;
1974         }
1975
1976         if (body->mbo_valid & OBD_MD_FLMTIME) {
1977                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1978                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1979                                "to "LPU64"\n", inode->i_ino,
1980                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1981                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1982                 }
1983                 lli->lli_lvb.lvb_mtime = body->mbo_mtime;
1984         }
1985
1986         if (body->mbo_valid & OBD_MD_FLCTIME) {
1987                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1988                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1989                 lli->lli_lvb.lvb_ctime = body->mbo_ctime;
1990         }
1991
1992         if (body->mbo_valid & OBD_MD_FLMODE)
1993                 inode->i_mode = (inode->i_mode & S_IFMT) |
1994                                 (body->mbo_mode & ~S_IFMT);
1995
1996         if (body->mbo_valid & OBD_MD_FLTYPE)
1997                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1998                                 (body->mbo_mode & S_IFMT);
1999
2000         LASSERT(inode->i_mode != 0);
2001         if (S_ISREG(inode->i_mode))
2002                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2003                                        LL_MAX_BLKSIZE_BITS);
2004         else
2005                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2006
2007         if (body->mbo_valid & OBD_MD_FLUID)
2008                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2009         if (body->mbo_valid & OBD_MD_FLGID)
2010                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2011         if (body->mbo_valid & OBD_MD_FLFLAGS)
2012                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
2013         if (body->mbo_valid & OBD_MD_FLNLINK)
2014                 set_nlink(inode, body->mbo_nlink);
2015         if (body->mbo_valid & OBD_MD_FLRDEV)
2016                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2017
2018         if (body->mbo_valid & OBD_MD_FLID) {
2019                 /* FID shouldn't be changed! */
2020                 if (fid_is_sane(&lli->lli_fid)) {
2021                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2022                                  "Trying to change FID "DFID
2023                                  " to the "DFID", inode "DFID"(%p)\n",
2024                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2025                                  PFID(ll_inode2fid(inode)), inode);
2026                 } else {
2027                         lli->lli_fid = body->mbo_fid1;
2028                 }
2029         }
2030
2031         LASSERT(fid_seq(&lli->lli_fid) != 0);
2032
2033         if (body->mbo_valid & OBD_MD_FLSIZE) {
2034                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2035                     S_ISREG(inode->i_mode)) {
2036                         struct lustre_handle lockh;
2037                         ldlm_mode_t mode;
2038
2039                         /* As it is possible a blocking ast has been processed
2040                          * by this time, we need to check there is an UPDATE
2041                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2042                          * it. */
2043                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2044                                                &lockh, LDLM_FL_CBPENDING,
2045                                                LCK_CR | LCK_CW |
2046                                                LCK_PR | LCK_PW);
2047                         if (mode) {
2048                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2049                                                       LLIF_EPOCH_PENDING |
2050                                                       LLIF_SOM_DIRTY)) {
2051                                         CERROR("%s: inode "DFID" flags %u still"
2052                                                " has size authority! do not "
2053                                                "trust the size from MDS\n",
2054                                                sbi->ll_md_exp->exp_obd->obd_name,
2055                                                PFID(ll_inode2fid(inode)),
2056                                                lli->lli_flags);
2057                                 } else {
2058                                         /* Use old size assignment to avoid
2059                                          * deadlock bz14138 & bz14326 */
2060                                         i_size_write(inode, body->mbo_size);
2061                                         spin_lock(&lli->lli_lock);
2062                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2063                                         spin_unlock(&lli->lli_lock);
2064                                 }
2065                                 ldlm_lock_decref(&lockh, mode);
2066                         }
2067                 } else {
2068                         /* Use old size assignment to avoid
2069                          * deadlock bz14138 & bz14326 */
2070                         i_size_write(inode, body->mbo_size);
2071
2072                         CDEBUG(D_VFSTRACE,
2073                                "inode="DFID", updating i_size %llu\n",
2074                                PFID(ll_inode2fid(inode)),
2075                                (unsigned long long)body->mbo_size);
2076                 }
2077
2078                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2079                         inode->i_blocks = body->mbo_blocks;
2080         }
2081
2082         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2083                 LASSERT(md->mds_capa);
2084                 ll_add_capa(inode, md->mds_capa);
2085         }
2086
2087         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2088                 LASSERT(md->oss_capa);
2089                 ll_add_capa(inode, md->oss_capa);
2090         }
2091
2092         if (body->mbo_valid & OBD_MD_TSTATE) {
2093                 if (body->mbo_t_state & MS_RESTORE)
2094                         lli->lli_flags |= LLIF_FILE_RESTORING;
2095         }
2096
2097         return 0;
2098 }
2099
2100 int ll_read_inode2(struct inode *inode, void *opaque)
2101 {
2102         struct lustre_md *md = opaque;
2103         struct ll_inode_info *lli = ll_i2info(inode);
2104         int     rc;
2105         ENTRY;
2106
2107         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2108                PFID(&lli->lli_fid), inode);
2109
2110         LASSERT(!lli->lli_has_smd);
2111
2112         /* Core attributes from the MDS first.  This is a new inode, and
2113          * the VFS doesn't zero times in the core inode so we have to do
2114          * it ourselves.  They will be overwritten by either MDS or OST
2115          * attributes - we just need to make sure they aren't newer. */
2116         LTIME_S(inode->i_mtime) = 0;
2117         LTIME_S(inode->i_atime) = 0;
2118         LTIME_S(inode->i_ctime) = 0;
2119         inode->i_rdev = 0;
2120         rc = ll_update_inode(inode, md);
2121         if (rc != 0)
2122                 RETURN(rc);
2123
2124         /* OIDEBUG(inode); */
2125
2126         /* initializing backing dev info. */
2127         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2128
2129
2130         if (S_ISREG(inode->i_mode)) {
2131                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2132                 inode->i_op = &ll_file_inode_operations;
2133                 inode->i_fop = sbi->ll_fop;
2134                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2135                 EXIT;
2136         } else if (S_ISDIR(inode->i_mode)) {
2137                 inode->i_op = &ll_dir_inode_operations;
2138                 inode->i_fop = &ll_dir_operations;
2139                 EXIT;
2140         } else if (S_ISLNK(inode->i_mode)) {
2141                 inode->i_op = &ll_fast_symlink_inode_operations;
2142                 EXIT;
2143         } else {
2144                 inode->i_op = &ll_special_inode_operations;
2145
2146                 init_special_inode(inode, inode->i_mode,
2147                                    inode->i_rdev);
2148
2149                 EXIT;
2150         }
2151
2152         return 0;
2153 }
2154
2155 void ll_delete_inode(struct inode *inode)
2156 {
2157         struct cl_inode_info *lli = cl_i2info(inode);
2158         ENTRY;
2159
2160         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2161                 /* discard all dirty pages before truncating them, required by
2162                  * osc_extent implementation at LU-1030. */
2163                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2164                                    CL_FSYNC_DISCARD, 1);
2165
2166         truncate_inode_pages(&inode->i_data, 0);
2167
2168         /* Workaround for LU-118 */
2169         if (inode->i_data.nrpages) {
2170                 spin_lock_irq(&inode->i_data.tree_lock);
2171                 spin_unlock_irq(&inode->i_data.tree_lock);
2172                 LASSERTF(inode->i_data.nrpages == 0,
2173                          "inode="DFID"(%p) nrpages=%lu, see "
2174                          "https://jira.hpdd.intel.com/browse/LU-118\n",
2175                          PFID(ll_inode2fid(inode)), inode,
2176                          inode->i_data.nrpages);
2177         }
2178         /* Workaround end */
2179
2180 #ifdef HAVE_SBOPS_EVICT_INODE
2181         ll_clear_inode(inode);
2182 #endif
2183         clear_inode(inode);
2184
2185         EXIT;
2186 }
2187
2188 int ll_iocontrol(struct inode *inode, struct file *file,
2189                  unsigned int cmd, unsigned long arg)
2190 {
2191         struct ll_sb_info *sbi = ll_i2sbi(inode);
2192         struct ptlrpc_request *req = NULL;
2193         int rc, flags = 0;
2194         ENTRY;
2195
2196         switch(cmd) {
2197         case FSFILT_IOC_GETFLAGS: {
2198                 struct mdt_body *body;
2199                 struct md_op_data *op_data;
2200
2201                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2202                                              0, 0, LUSTRE_OPC_ANY,
2203                                              NULL);
2204                 if (IS_ERR(op_data))
2205                         RETURN(PTR_ERR(op_data));
2206
2207                 op_data->op_valid = OBD_MD_FLFLAGS;
2208                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2209                 ll_finish_md_op_data(op_data);
2210                 if (rc) {
2211                         CERROR("%s: failure inode "DFID": rc = %d\n",
2212                                sbi->ll_md_exp->exp_obd->obd_name,
2213                                PFID(ll_inode2fid(inode)), rc);
2214                         RETURN(-abs(rc));
2215                 }
2216
2217                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2218
2219                 flags = body->mbo_flags;
2220
2221                 ptlrpc_req_finished(req);
2222
2223                 RETURN(put_user(flags, (int *)arg));
2224         }
2225         case FSFILT_IOC_SETFLAGS: {
2226                 struct lov_stripe_md *lsm;
2227                 struct obd_info oinfo = { { { 0 } } };
2228                 struct md_op_data *op_data;
2229
2230                 if (get_user(flags, (int *)arg))
2231                         RETURN(-EFAULT);
2232
2233                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2234                                              LUSTRE_OPC_ANY, NULL);
2235                 if (IS_ERR(op_data))
2236                         RETURN(PTR_ERR(op_data));
2237
2238                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2239                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2240                 rc = md_setattr(sbi->ll_md_exp, op_data,
2241                                 NULL, 0, NULL, 0, &req, NULL);
2242                 ll_finish_md_op_data(op_data);
2243                 ptlrpc_req_finished(req);
2244                 if (rc)
2245                         RETURN(rc);
2246
2247                 inode->i_flags = ll_ext_to_inode_flags(flags);
2248
2249                 lsm = ccc_inode_lsm_get(inode);
2250                 if (!lsm_has_objects(lsm)) {
2251                         ccc_inode_lsm_put(inode, lsm);
2252                         RETURN(0);
2253                 }
2254
2255                 OBDO_ALLOC(oinfo.oi_oa);
2256                 if (!oinfo.oi_oa) {
2257                         ccc_inode_lsm_put(inode, lsm);
2258                         RETURN(-ENOMEM);
2259                 }
2260                 oinfo.oi_md = lsm;
2261                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2262                 oinfo.oi_oa->o_flags = flags;
2263                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2264                                        OBD_MD_FLGROUP;
2265                 oinfo.oi_capa = ll_mdscapa_get(inode);
2266                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2267                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2268                 capa_put(oinfo.oi_capa);
2269                 OBDO_FREE(oinfo.oi_oa);
2270                 ccc_inode_lsm_put(inode, lsm);
2271
2272                 if (rc && rc != -EPERM && rc != -EACCES)
2273                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2274
2275                 RETURN(rc);
2276         }
2277         default:
2278                 RETURN(-ENOSYS);
2279         }
2280
2281         RETURN(0);
2282 }
2283
2284 int ll_flush_ctx(struct inode *inode)
2285 {
2286         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2287
2288         CDEBUG(D_SEC, "flush context for user %d\n",
2289                from_kuid(&init_user_ns, current_uid()));
2290
2291         obd_set_info_async(NULL, sbi->ll_md_exp,
2292                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2293                            0, NULL, NULL);
2294         obd_set_info_async(NULL, sbi->ll_dt_exp,
2295                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2296                            0, NULL, NULL);
2297         return 0;
2298 }
2299
2300 /* umount -f client means force down, don't save state */
2301 void ll_umount_begin(struct super_block *sb)
2302 {
2303         struct ll_sb_info *sbi = ll_s2sbi(sb);
2304         struct obd_device *obd;
2305         struct obd_ioctl_data *ioc_data;
2306         ENTRY;
2307
2308         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2309                sb->s_count, atomic_read(&sb->s_active));
2310
2311         obd = class_exp2obd(sbi->ll_md_exp);
2312         if (obd == NULL) {
2313                 CERROR("Invalid MDC connection handle "LPX64"\n",
2314                        sbi->ll_md_exp->exp_handle.h_cookie);
2315                 EXIT;
2316                 return;
2317         }
2318         obd->obd_force = 1;
2319
2320         obd = class_exp2obd(sbi->ll_dt_exp);
2321         if (obd == NULL) {
2322                 CERROR("Invalid LOV connection handle "LPX64"\n",
2323                        sbi->ll_dt_exp->exp_handle.h_cookie);
2324                 EXIT;
2325                 return;
2326         }
2327         obd->obd_force = 1;
2328
2329         OBD_ALLOC_PTR(ioc_data);
2330         if (ioc_data) {
2331                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2332                               sizeof *ioc_data, ioc_data, NULL);
2333
2334                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2335                               sizeof *ioc_data, ioc_data, NULL);
2336
2337                 OBD_FREE_PTR(ioc_data);
2338         }
2339
2340         /* Really, we'd like to wait until there are no requests outstanding,
2341          * and then continue.  For now, we just invalidate the requests,
2342          * schedule() and sleep one second if needed, and hope.
2343          */
2344         schedule();
2345         EXIT;
2346 }
2347
2348 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2349 {
2350         struct ll_sb_info *sbi = ll_s2sbi(sb);
2351         char *profilenm = get_profile_name(sb);
2352         int err;
2353         __u32 read_only;
2354
2355         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2356                 read_only = *flags & MS_RDONLY;
2357                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2358                                          sizeof(KEY_READ_ONLY),
2359                                          KEY_READ_ONLY, sizeof(read_only),
2360                                          &read_only, NULL);
2361                 if (err) {
2362                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2363                                       profilenm, read_only ?
2364                                       "read-only" : "read-write", err);
2365                         return err;
2366                 }
2367
2368                 if (read_only)
2369                         sb->s_flags |= MS_RDONLY;
2370                 else
2371                         sb->s_flags &= ~MS_RDONLY;
2372
2373                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2374                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2375                                       read_only ?  "read-only" : "read-write");
2376         }
2377         return 0;
2378 }
2379
2380 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2381                   struct super_block *sb, struct lookup_intent *it)
2382 {
2383         struct ll_sb_info *sbi = NULL;
2384         struct lustre_md md = { 0 };
2385         int rc;
2386         ENTRY;
2387
2388         LASSERT(*inode || sb);
2389         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2390         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2391                               sbi->ll_md_exp, &md);
2392         if (rc)
2393                 RETURN(rc);
2394
2395         if (*inode) {
2396                 rc = ll_update_inode(*inode, &md);
2397                 if (rc != 0)
2398                         GOTO(out, rc);
2399         } else {
2400                 LASSERT(sb != NULL);
2401
2402                 /*
2403                  * At this point server returns to client's same fid as client
2404                  * generated for creating. So using ->fid1 is okay here.
2405                  */
2406                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2407
2408                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2409                                              sbi->ll_flags & LL_SBI_32BIT_API),
2410                                  &md);
2411                 if (IS_ERR(*inode)) {
2412 #ifdef CONFIG_FS_POSIX_ACL
2413                         if (md.posix_acl) {
2414                                 posix_acl_release(md.posix_acl);
2415                                 md.posix_acl = NULL;
2416                         }
2417 #endif
2418                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2419                         *inode = NULL;
2420                         CERROR("new_inode -fatal: rc %d\n", rc);
2421                         GOTO(out, rc);
2422                 }
2423         }
2424
2425         /* Handling piggyback layout lock.
2426          * Layout lock can be piggybacked by getattr and open request.
2427          * The lsm can be applied to inode only if it comes with a layout lock
2428          * otherwise correct layout may be overwritten, for example:
2429          * 1. proc1: mdt returns a lsm but not granting layout
2430          * 2. layout was changed by another client
2431          * 3. proc2: refresh layout and layout lock granted
2432          * 4. proc1: to apply a stale layout */
2433         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2434                 struct lustre_handle lockh;
2435                 struct ldlm_lock *lock;
2436
2437                 lockh.cookie = it->d.lustre.it_lock_handle;
2438                 lock = ldlm_handle2lock(&lockh);
2439                 LASSERT(lock != NULL);
2440                 if (ldlm_has_layout(lock)) {
2441                         struct cl_object_conf conf;
2442
2443                         memset(&conf, 0, sizeof(conf));
2444                         conf.coc_opc = OBJECT_CONF_SET;
2445                         conf.coc_inode = *inode;
2446                         conf.coc_lock = lock;
2447                         conf.u.coc_md = &md;
2448                         (void)ll_layout_conf(*inode, &conf);
2449                 }
2450                 LDLM_LOCK_PUT(lock);
2451         }
2452
2453 out:
2454         if (md.lsm != NULL)
2455                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2456         md_free_lustre_md(sbi->ll_md_exp, &md);
2457         RETURN(rc);
2458 }
2459
2460 int ll_obd_statfs(struct inode *inode, void *arg)
2461 {
2462         struct ll_sb_info *sbi = NULL;
2463         struct obd_export *exp;
2464         char *buf = NULL;
2465         struct obd_ioctl_data *data = NULL;
2466         __u32 type;
2467         __u32 flags;
2468         int len = 0, rc;
2469
2470         if (!inode || !(sbi = ll_i2sbi(inode)))
2471                 GOTO(out_statfs, rc = -EINVAL);
2472
2473         rc = obd_ioctl_getdata(&buf, &len, arg);
2474         if (rc)
2475                 GOTO(out_statfs, rc);
2476
2477         data = (void*)buf;
2478         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2479             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2480                 GOTO(out_statfs, rc = -EINVAL);
2481
2482         if (data->ioc_inllen1 != sizeof(__u32) ||
2483             data->ioc_inllen2 != sizeof(__u32) ||
2484             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2485             data->ioc_plen2 != sizeof(struct obd_uuid))
2486                 GOTO(out_statfs, rc = -EINVAL);
2487
2488         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2489         if (type & LL_STATFS_LMV)
2490                 exp = sbi->ll_md_exp;
2491         else if (type & LL_STATFS_LOV)
2492                 exp = sbi->ll_dt_exp;
2493         else
2494                 GOTO(out_statfs, rc = -ENODEV);
2495
2496         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2497         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2498         if (rc)
2499                 GOTO(out_statfs, rc);
2500 out_statfs:
2501         if (buf)
2502                 obd_ioctl_freedata(buf, len);
2503         return rc;
2504 }
2505
2506 int ll_process_config(struct lustre_cfg *lcfg)
2507 {
2508         struct super_block *sb;
2509         unsigned long x;
2510         int rc = 0;
2511         char *ptr;
2512
2513         /* The instance name contains the sb: lustre-client-aacfe000 */
2514         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2515         if (!ptr || !*(++ptr))
2516                 return -EINVAL;
2517         if (sscanf(ptr, "%lx", &x) != 1)
2518                 return -EINVAL;
2519         sb = (struct super_block *)x;
2520         /* This better be a real Lustre superblock! */
2521         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2522
2523         /* Note we have not called client_common_fill_super yet, so
2524            proc fns must be able to handle that! */
2525         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2526                                           lcfg, sb);
2527         if (rc > 0)
2528                 rc = 0;
2529         return rc;
2530 }
2531
2532 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2533 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2534                                        struct inode *i1, struct inode *i2,
2535                                        const char *name, int namelen,
2536                                        int mode, __u32 opc, void *data)
2537 {
2538         LASSERT(i1 != NULL);
2539
2540         if (name == NULL) {
2541                 /* Do not reuse namelen for something else. */
2542                 if (namelen != 0)
2543                         return ERR_PTR(-EINVAL);
2544         } else {
2545                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2546                         return ERR_PTR(-ENAMETOOLONG);
2547
2548                 if (!lu_name_is_valid_2(name, namelen))
2549                         return ERR_PTR(-EINVAL);
2550         }
2551
2552         if (op_data == NULL)
2553                 OBD_ALLOC_PTR(op_data);
2554
2555         if (op_data == NULL)
2556                 return ERR_PTR(-ENOMEM);
2557
2558         ll_i2gids(op_data->op_suppgids, i1, i2);
2559         op_data->op_fid1 = *ll_inode2fid(i1);
2560         op_data->op_capa1 = ll_mdscapa_get(i1);
2561         if (S_ISDIR(i1->i_mode))
2562                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2563
2564         if (i2) {
2565                 op_data->op_fid2 = *ll_inode2fid(i2);
2566                 op_data->op_capa2 = ll_mdscapa_get(i2);
2567                 if (S_ISDIR(i2->i_mode))
2568                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2569         } else {
2570                 fid_zero(&op_data->op_fid2);
2571                 op_data->op_capa2 = NULL;
2572         }
2573
2574         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2575                 op_data->op_cli_flags |= CLI_HASH64;
2576
2577         if (ll_need_32bit_api(ll_i2sbi(i1)))
2578                 op_data->op_cli_flags |= CLI_API32;
2579
2580         op_data->op_name = name;
2581         op_data->op_namelen = namelen;
2582         op_data->op_mode = mode;
2583         op_data->op_mod_time = cfs_time_current_sec();
2584         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2585         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2586         op_data->op_cap = cfs_curproc_cap_pack();
2587         op_data->op_bias = 0;
2588         op_data->op_cli_flags = 0;
2589         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2590              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2591                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2592         } else {
2593                 op_data->op_mds = 0;
2594         }
2595         op_data->op_data = data;
2596
2597         /* When called by ll_setattr_raw, file is i1. */
2598         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2599                 op_data->op_bias |= MDS_DATA_MODIFIED;
2600
2601         return op_data;
2602 }
2603
2604 void ll_finish_md_op_data(struct md_op_data *op_data)
2605 {
2606         capa_put(op_data->op_capa1);
2607         capa_put(op_data->op_capa2);
2608         OBD_FREE_PTR(op_data);
2609 }
2610
2611 #ifdef HAVE_SUPEROPS_USE_DENTRY
2612 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2613 #else
2614 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2615 #endif
2616 {
2617         struct ll_sb_info *sbi;
2618
2619 #ifdef HAVE_SUPEROPS_USE_DENTRY
2620         LASSERT((seq != NULL) && (dentry != NULL));
2621         sbi = ll_s2sbi(dentry->d_sb);
2622 #else
2623         LASSERT((seq != NULL) && (vfs != NULL));
2624         sbi = ll_s2sbi(vfs->mnt_sb);
2625 #endif
2626
2627         if (sbi->ll_flags & LL_SBI_NOLCK)
2628                 seq_puts(seq, ",nolock");
2629
2630         if (sbi->ll_flags & LL_SBI_FLOCK)
2631                 seq_puts(seq, ",flock");
2632
2633         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2634                 seq_puts(seq, ",localflock");
2635
2636         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2637                 seq_puts(seq, ",user_xattr");
2638
2639         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2640                 seq_puts(seq, ",lazystatfs");
2641
2642         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2643                 seq_puts(seq, ",user_fid2path");
2644
2645         RETURN(0);
2646 }
2647
2648 /**
2649  * Get obd name by cmd, and copy out to user space
2650  */
2651 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2652 {
2653         struct ll_sb_info *sbi = ll_i2sbi(inode);
2654         struct obd_device *obd;
2655         ENTRY;
2656
2657         if (cmd == OBD_IOC_GETDTNAME)
2658                 obd = class_exp2obd(sbi->ll_dt_exp);
2659         else if (cmd == OBD_IOC_GETMDNAME)
2660                 obd = class_exp2obd(sbi->ll_md_exp);
2661         else
2662                 RETURN(-EINVAL);
2663
2664         if (!obd)
2665                 RETURN(-ENOENT);
2666
2667         if (copy_to_user((void *)arg, obd->obd_name,
2668                          strlen(obd->obd_name) + 1))
2669                 RETURN(-EFAULT);
2670
2671         RETURN(0);
2672 }
2673
2674 /**
2675  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2676  * fsname will be returned in this buffer; otherwise, a static buffer will be
2677  * used to store the fsname and returned to caller.
2678  */
2679 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2680 {
2681         static char fsname_static[MTI_NAME_MAXLEN];
2682         struct lustre_sb_info *lsi = s2lsi(sb);
2683         char *ptr;
2684         int len;
2685
2686         if (buf == NULL) {
2687                 /* this means the caller wants to use static buffer
2688                  * and it doesn't care about race. Usually this is
2689                  * in error reporting path */
2690                 buf = fsname_static;
2691                 buflen = sizeof(fsname_static);
2692         }
2693
2694         len = strlen(lsi->lsi_lmd->lmd_profile);
2695         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2696         if (ptr && (strcmp(ptr, "-client") == 0))
2697                 len -= 7;
2698
2699         if (unlikely(len >= buflen))
2700                 len = buflen - 1;
2701         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2702         buf[len] = '\0';
2703
2704         return buf;
2705 }
2706
2707 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2708 {
2709         char *path = NULL;
2710
2711         struct path p;
2712
2713         p.dentry = dentry;
2714         p.mnt = current->fs->root.mnt;
2715         path_get(&p);
2716         path = d_path(&p, buf, bufsize);
2717         path_put(&p);
2718         return path;
2719 }
2720
2721 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2722 {
2723         char *buf, *path = NULL;
2724         struct dentry *dentry = NULL;
2725         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2726
2727         /* this can be called inside spin lock so use GFP_ATOMIC. */
2728         buf = (char *)__get_free_page(GFP_ATOMIC);
2729         if (buf != NULL) {
2730                 dentry = d_find_alias(page->mapping->host);
2731                 if (dentry != NULL)
2732                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2733         }
2734
2735         CDEBUG(D_WARNING,
2736                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2737                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2738                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2739                PFID(&obj->cob_header.coh_lu.loh_fid),
2740                (path && !IS_ERR(path)) ? path : "", ioret);
2741
2742         if (dentry != NULL)
2743                 dput(dentry);
2744
2745         if (buf != NULL)
2746                 free_page((unsigned long)buf);
2747 }
2748
2749 /*
2750  * Compute llite root squash state after a change of root squash
2751  * configuration setting or add/remove of a lnet nid
2752  */
2753 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2754 {
2755         struct root_squash_info *squash = &sbi->ll_squash;
2756         int i;
2757         bool matched;
2758         lnet_process_id_t id;
2759
2760         /* Update norootsquash flag */
2761         down_write(&squash->rsi_sem);
2762         if (list_empty(&squash->rsi_nosquash_nids))
2763                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2764         else {
2765                 /* Do not apply root squash as soon as one of our NIDs is
2766                  * in the nosquash_nids list */
2767                 matched = false;
2768                 i = 0;
2769                 while (LNetGetId(i++, &id) != -ENOENT) {
2770                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2771                                 continue;
2772                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2773                                 matched = true;
2774                                 break;
2775                         }
2776                 }
2777                 if (matched)
2778                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2779                 else
2780                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2781         }
2782         up_write(&squash->rsi_sem);
2783 }
2784
2785