lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2013, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_ioctl.h>
  49 #include <lustre_lite.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 struct kmem_cache *ll_file_data_slab;
  61
  62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  63 static DEFINE_SPINLOCK(ll_sb_lock);
  64
  65 #ifndef log2
  66 #define log2(n) ffz(~(n))
  67 #endif
  68
  69 static struct ll_sb_info *ll_init_sbi(void)
  70 {
  71         struct ll_sb_info *sbi = NULL;
  72         unsigned long pages;
  73         unsigned long lru_page_max;
  74         struct sysinfo si;
  75         class_uuid_t uuid;
  76         int i;
  77         ENTRY;
  78
  79         OBD_ALLOC(sbi, sizeof(*sbi));
  80         if (!sbi)
  81                 RETURN(NULL);
  82
  83         spin_lock_init(&sbi->ll_lock);
  84         mutex_init(&sbi->ll_lco.lco_lock);
  85         spin_lock_init(&sbi->ll_pp_extent_lock);
  86         spin_lock_init(&sbi->ll_process_lock);
  87         sbi->ll_rw_stats_on = 0;
  88
  89         si_meminfo(&si);
  90         pages = si.totalram - si.totalhigh;
  91         lru_page_max = pages / 2;
  92
  93         /* initialize ll_cache data */
  94         atomic_set(&sbi->ll_cache.ccc_users, 0);
  95         sbi->ll_cache.ccc_lru_max = lru_page_max;
  96         atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
  97         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
  98         INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
  99
 100         atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 101         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         spin_lock(&ll_sb_lock);
 116         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         spin_unlock(&ll_sb_lock);
 118
 119         sbi->ll_flags |= LL_SBI_VERBOSE;
 120 #ifdef ENABLE_CHECKSUM
 121         sbi->ll_flags |= LL_SBI_CHECKSUM;
 122 #endif
 123
 124 #ifdef HAVE_LRU_RESIZE_SUPPORT
 125         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 126 #endif
 127
 128         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 129                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 130                                pp_r_hist.oh_lock);
 131                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 132                                pp_w_hist.oh_lock);
 133         }
 134
 135         /* metadata statahead is enabled by default */
 136         sbi->ll_sa_max = LL_SA_RPC_DEF;
 137         atomic_set(&sbi->ll_sa_total, 0);
 138         atomic_set(&sbi->ll_sa_wrong, 0);
 139         atomic_set(&sbi->ll_agl_total, 0);
 140         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 141
 142         RETURN(sbi);
 143 }
 144
 145 static void ll_free_sbi(struct super_block *sb)
 146 {
 147         struct ll_sb_info *sbi = ll_s2sbi(sb);
 148         ENTRY;
 149
 150         if (sbi != NULL) {
 151                 spin_lock(&ll_sb_lock);
 152                 list_del(&sbi->ll_list);
 153                 spin_unlock(&ll_sb_lock);
 154                 OBD_FREE(sbi, sizeof(*sbi));
 155         }
 156         EXIT;
 157 }
 158
 159 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 160                                     struct vfsmount *mnt)
 161 {
 162         struct inode *root = 0;
 163         struct ll_sb_info *sbi = ll_s2sbi(sb);
 164         struct obd_device *obd;
 165         struct obd_capa *oc = NULL;
 166         struct obd_statfs *osfs = NULL;
 167         struct ptlrpc_request *request = NULL;
 168         struct obd_connect_data *data = NULL;
 169         struct obd_uuid *uuid;
 170         struct md_op_data *op_data;
 171         struct lustre_md lmd;
 172         obd_valid valid;
 173         int size, err, checksum;
 174         ENTRY;
 175
 176         obd = class_name2obd(md);
 177         if (!obd) {
 178                 CERROR("MD %s: not setup or attached\n", md);
 179                 RETURN(-EINVAL);
 180         }
 181
 182         OBD_ALLOC_PTR(data);
 183         if (data == NULL)
 184                 RETURN(-ENOMEM);
 185
 186         OBD_ALLOC_PTR(osfs);
 187         if (osfs == NULL) {
 188                 OBD_FREE_PTR(data);
 189                 RETURN(-ENOMEM);
 190         }
 191
 192         /* indicate the features supported by this client */
 193         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 194                                   OBD_CONNECT_ATTRFID  |
 195                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 196                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 197                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 198                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 199                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 200                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 201                                   OBD_CONNECT_EINPROGRESS |
 202                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 203                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 204                                   OBD_CONNECT_MAX_EASIZE |
 205                                   OBD_CONNECT_FLOCK_DEAD |
 206                                   OBD_CONNECT_DISP_STRIPE;
 207
 208         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 209                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 210
 211 #ifdef HAVE_LRU_RESIZE_SUPPORT
 212         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 213                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 214 #endif
 215 #ifdef CONFIG_FS_POSIX_ACL
 216         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 217 #endif
 218
 219         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 220                 /* flag mdc connection as lightweight, only used for test
 221                  * purpose, use with care */
 222                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 223
 224         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 225         data->ocd_version = LUSTRE_VERSION_CODE;
 226
 227         if (sb->s_flags & MS_RDONLY)
 228                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 229         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 230                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 231
 232 #ifdef HAVE_MS_FLOCK_LOCK
 233         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 234         sb->s_flags |= MS_FLOCK_LOCK;
 235 #endif
 236 #ifdef MS_HAS_NEW_AOPS
 237         sb->s_flags |= MS_HAS_NEW_AOPS;
 238 #endif
 239
 240         if (sbi->ll_flags & LL_SBI_FLOCK)
 241                 sbi->ll_fop = &ll_file_operations_flock;
 242         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 243                 sbi->ll_fop = &ll_file_operations;
 244         else
 245                 sbi->ll_fop = &ll_file_operations_noflock;
 246
 247         /* real client */
 248         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 249         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 250                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 251
 252         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 253
 254         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 255         if (err == -EBUSY) {
 256                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 257                                    "recovery, of which this client is not a "
 258                                    "part. Please wait for recovery to complete,"
 259                                    " abort, or time out.\n", md);
 260                 GOTO(out, err);
 261         } else if (err) {
 262                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 263                 GOTO(out, err);
 264         }
 265
 266         sbi->ll_md_exp->exp_connect_data = *data;
 267
 268         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 269                            LUSTRE_SEQ_METADATA);
 270         if (err) {
 271                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 272                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 273                 GOTO(out_md, err);
 274         }
 275
 276         /* For mount, we only need fs info from MDT0, and also in DNE, it
 277          * can make sure the client can be mounted as long as MDT0 is
 278          * avaible */
 279         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 280                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 281                         OBD_STATFS_FOR_MDT0);
 282         if (err)
 283                 GOTO(out_md_fid, err);
 284
 285         /* This needs to be after statfs to ensure connect has finished.
 286          * Note that "data" does NOT contain the valid connect reply.
 287          * If connecting to a 1.8 server there will be no LMV device, so
 288          * we can access the MDC export directly and exp_connect_flags will
 289          * be non-zero, but if accessing an upgraded 2.1 server it will
 290          * have the correct flags filled in.
 291          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 292         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 293         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 294             valid != CLIENT_CONNECT_MDT_REQD) {
 295                 char *buf;
 296
 297                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 298                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 299                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 300                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 301                                    "feature(s) needed for correct operation "
 302                                    "of this client (%s). Please upgrade "
 303                                    "server or downgrade client.\n",
 304                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 305                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 306                 GOTO(out_md_fid, err = -EPROTO);
 307         }
 308
 309         size = sizeof(*data);
 310         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 311                            KEY_CONN_DATA,  &size, data, NULL);
 312         if (err) {
 313                 CERROR("%s: Get connect data failed: rc = %d\n",
 314                        sbi->ll_md_exp->exp_obd->obd_name, err);
 315                 GOTO(out_md_fid, err);
 316         }
 317
 318         LASSERT(osfs->os_bsize);
 319         sb->s_blocksize = osfs->os_bsize;
 320         sb->s_blocksize_bits = log2(osfs->os_bsize);
 321         sb->s_magic = LL_SUPER_MAGIC;
 322         sb->s_maxbytes = MAX_LFS_FILESIZE;
 323         sbi->ll_namelen = osfs->os_namelen;
 324         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 325
 326         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 327             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 328                 LCONSOLE_INFO("Disabling user_xattr feature because "
 329                               "it is not supported on the server\n");
 330                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 331         }
 332
 333         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 334 #ifdef MS_POSIXACL
 335                 sb->s_flags |= MS_POSIXACL;
 336 #endif
 337                 sbi->ll_flags |= LL_SBI_ACL;
 338         } else {
 339                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 340 #ifdef MS_POSIXACL
 341                 sb->s_flags &= ~MS_POSIXACL;
 342 #endif
 343                 sbi->ll_flags &= ~LL_SBI_ACL;
 344         }
 345
 346         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 347                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 348                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 349                         LCONSOLE_INFO("client is set as remote by default.\n");
 350                 }
 351         } else {
 352                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 353                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 354                         LCONSOLE_INFO("client claims to be remote, but server "
 355                                       "rejected, forced to be local.\n");
 356                 }
 357         }
 358
 359         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 360                 LCONSOLE_INFO("client enabled MDS capability!\n");
 361                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 362         }
 363
 364         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 365                 LCONSOLE_INFO("client enabled OSS capability!\n");
 366                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 370                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 371
 372         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 373                 sbi->ll_md_brw_size = data->ocd_brw_size;
 374         else
 375                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 376
 377         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 378                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 379
 380         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 381                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 382                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 383                                       "unknown maximum xattr size.\n", dt);
 384                 } else {
 385                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 386                         sbi->ll_xattr_cache_enabled = 1;
 387                 }
 388         }
 389
 390         obd = class_name2obd(dt);
 391         if (!obd) {
 392                 CERROR("DT %s: not setup or attached\n", dt);
 393                 GOTO(out_md_fid, err = -ENODEV);
 394         }
 395
 396         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 397                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 398                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 399                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 400                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 401                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 402                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 403                                   OBD_CONNECT_MAXBYTES |
 404                                   OBD_CONNECT_EINPROGRESS |
 405                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 406                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 407
 408         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 409                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 410
 411         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 412                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 413                  * disabled by default, because it can still be enabled on the
 414                  * fly via /proc. As a consequence, we still need to come to an
 415                  * agreement on the supported algorithms at connect time */
 416                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 417
 418                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 419                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 420                 else
 421                         data->ocd_cksum_types = cksum_types_supported_client();
 422         }
 423
 424 #ifdef HAVE_LRU_RESIZE_SUPPORT
 425         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 426 #endif
 427         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 428                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 429
 430         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 431                "ocd_grant: %d\n", data->ocd_connect_flags,
 432                data->ocd_version, data->ocd_grant);
 433
 434         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 435         obd->obd_upcall.onu_upcall = cl_ocd_update;
 436
 437         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 438
 439         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 440                           NULL);
 441         if (err == -EBUSY) {
 442                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 443                                    "recovery, of which this client is not a "
 444                                    "part.  Please wait for recovery to "
 445                                    "complete, abort, or time out.\n", dt);
 446                 GOTO(out_md, err);
 447         } else if (err) {
 448                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 449                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 450                 GOTO(out_md, err);
 451         }
 452
 453         sbi->ll_dt_exp->exp_connect_data = *data;
 454
 455         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 456                            LUSTRE_SEQ_METADATA);
 457         if (err) {
 458                 CERROR("%s: Can't init data layer FID infrastructure, "
 459                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 460                 GOTO(out_dt, err);
 461         }
 462
 463         mutex_lock(&sbi->ll_lco.lco_lock);
 464         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 465         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 466         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 467         mutex_unlock(&sbi->ll_lco.lco_lock);
 468
 469         fid_zero(&sbi->ll_root_fid);
 470         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 471         if (err) {
 472                 CERROR("cannot mds_connect: rc = %d\n", err);
 473                 GOTO(out_lock_cn_cb, err);
 474         }
 475         if (!fid_is_sane(&sbi->ll_root_fid)) {
 476                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 477                        sbi->ll_md_exp->exp_obd->obd_name,
 478                        PFID(&sbi->ll_root_fid));
 479                 GOTO(out_lock_cn_cb, err = -EINVAL);
 480         }
 481         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 482
 483         sb->s_op = &lustre_super_operations;
 484 #if THREAD_SIZE >= 8192 /*b=17630*/
 485         sb->s_export_op = &lustre_export_operations;
 486 #endif
 487
 488         /* make root inode
 489          * XXX: move this to after cbd setup? */
 490         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 491                 OBD_MD_FLMODEASIZE;
 492         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 493                 valid |= OBD_MD_FLRMTPERM;
 494         else if (sbi->ll_flags & LL_SBI_ACL)
 495                 valid |= OBD_MD_FLACL;
 496
 497         OBD_ALLOC_PTR(op_data);
 498         if (op_data == NULL)
 499                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 500
 501         op_data->op_fid1 = sbi->ll_root_fid;
 502         op_data->op_mode = 0;
 503         op_data->op_capa1 = oc;
 504         op_data->op_valid = valid;
 505
 506         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 507         if (oc)
 508                 capa_put(oc);
 509         OBD_FREE_PTR(op_data);
 510         if (err) {
 511                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 512                        sbi->ll_md_exp->exp_obd->obd_name, err);
 513                 GOTO(out_lock_cn_cb, err);
 514         }
 515
 516         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 517                                sbi->ll_md_exp, &lmd);
 518         if (err) {
 519                 CERROR("failed to understand root inode md: rc = %d\n", err);
 520                 ptlrpc_req_finished(request);
 521                 GOTO(out_lock_cn_cb, err);
 522         }
 523
 524         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 525         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 526                                             sbi->ll_flags & LL_SBI_32BIT_API),
 527                        &lmd);
 528         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 529         ptlrpc_req_finished(request);
 530
 531         if (IS_ERR(root)) {
 532                 if (lmd.lsm)
 533                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 534 #ifdef CONFIG_FS_POSIX_ACL
 535                 if (lmd.posix_acl) {
 536                         posix_acl_release(lmd.posix_acl);
 537                         lmd.posix_acl = NULL;
 538                 }
 539 #endif
 540                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 541                 root = NULL;
 542                 CERROR("lustre_lite: bad iget4 for root\n");
 543                 GOTO(out_root, err);
 544         }
 545
 546         err = ll_close_thread_start(&sbi->ll_lcq);
 547         if (err) {
 548                 CERROR("cannot start close thread: rc %d\n", err);
 549                 GOTO(out_root, err);
 550         }
 551
 552 #ifdef CONFIG_FS_POSIX_ACL
 553         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 554                 rct_init(&sbi->ll_rct);
 555                 et_init(&sbi->ll_et);
 556         }
 557 #endif
 558
 559         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 560         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 561                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 562                                  NULL);
 563         cl_sb_init(sb);
 564
 565         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 566                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 567                                  &sbi->ll_cache, NULL);
 568
 569         sb->s_root = d_make_root(root);
 570         if (sb->s_root == NULL) {
 571                 CERROR("%s: can't make root dentry\n",
 572                         ll_get_fsname(sb, NULL, 0));
 573                 GOTO(out_root, err = -ENOMEM);
 574         }
 575 #ifdef HAVE_DCACHE_LOCK
 576         sb->s_root->d_op = &ll_d_ops;
 577 #endif
 578
 579         sbi->ll_sdev_orig = sb->s_dev;
 580
 581         /* We set sb->s_dev equal on all lustre clients in order to support
 582          * NFS export clustering.  NFSD requires that the FSID be the same
 583          * on all clients. */
 584         /* s_dev is also used in lt_compare() to compare two fs, but that is
 585          * only a node-local comparison. */
 586         uuid = obd_get_uuid(sbi->ll_md_exp);
 587         if (uuid != NULL)
 588                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 589
 590         if (data != NULL)
 591                 OBD_FREE_PTR(data);
 592         if (osfs != NULL)
 593                 OBD_FREE_PTR(osfs);
 594         if (proc_lustre_fs_root != NULL) {
 595                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 596                                                   dt, md);
 597                 if (err < 0) {
 598                         CERROR("%s: could not register mount in lprocfs: "
 599                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 600                         err = 0;
 601                 }
 602         }
 603
 604         RETURN(err);
 605 out_root:
 606         if (root)
 607                 iput(root);
 608 out_lock_cn_cb:
 609         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 610 out_dt:
 611         obd_disconnect(sbi->ll_dt_exp);
 612         sbi->ll_dt_exp = NULL;
 613         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 614         obd_zombie_barrier();
 615 out_md_fid:
 616         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 617 out_md:
 618         obd_disconnect(sbi->ll_md_exp);
 619         sbi->ll_md_exp = NULL;
 620 out:
 621         if (data != NULL)
 622                 OBD_FREE_PTR(data);
 623         if (osfs != NULL)
 624                 OBD_FREE_PTR(osfs);
 625         return err;
 626 }
 627
 628 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 629 {
 630         int size, rc;
 631
 632         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 633         size = sizeof(int);
 634         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 635                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 636         if (rc)
 637                 CERROR("Get max mdsize error rc %d\n", rc);
 638
 639         RETURN(rc);
 640 }
 641
 642 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 643 {
 644         int size, rc;
 645
 646         size = sizeof(int);
 647         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 648                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 649         if (rc)
 650                 CERROR("Get default mdsize error rc %d\n", rc);
 651
 652         RETURN(rc);
 653 }
 654
 655 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 656 {
 657         int size, rc;
 658
 659         size = sizeof(int);
 660         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 661                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 662         if (rc)
 663                 CERROR("Get max cookiesize error rc %d\n", rc);
 664
 665         RETURN(rc);
 666 }
 667
 668 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 669 {
 670         int size, rc;
 671
 672         size = sizeof(int);
 673         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 674                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 675         if (rc)
 676                 CERROR("Get default cookiesize error rc %d\n", rc);
 677
 678         RETURN(rc);
 679 }
 680
 681 static void ll_dump_inode(struct inode *inode)
 682 {
 683         struct ll_d_hlist_node *tmp;
 684         int dentry_count = 0;
 685
 686         LASSERT(inode != NULL);
 687
 688         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 689                 dentry_count++;
 690
 691         CERROR("%s: inode %p dump: dev=%s fid="DFID
 692                " mode=%o count=%u, %d dentries\n",
 693                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 694                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 695                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 696 }
 697
 698 void lustre_dump_dentry(struct dentry *dentry, int recur)
 699 {
 700         struct list_head *tmp;
 701         int subdirs = 0;
 702
 703         LASSERT(dentry != NULL);
 704
 705         list_for_each(tmp, &dentry->d_subdirs)
 706                 subdirs++;
 707
 708         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 709                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 710                dentry->d_name.len, dentry->d_name.name,
 711                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 712                dentry->d_parent, dentry->d_inode, d_count(dentry),
 713                dentry->d_flags, dentry->d_fsdata, subdirs);
 714         if (dentry->d_inode != NULL)
 715                 ll_dump_inode(dentry->d_inode);
 716
 717         if (recur == 0)
 718                 return;
 719
 720         list_for_each(tmp, &dentry->d_subdirs) {
 721                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 722                 lustre_dump_dentry(d, recur - 1);
 723         }
 724 }
 725
 726 static void client_common_put_super(struct super_block *sb)
 727 {
 728         struct ll_sb_info *sbi = ll_s2sbi(sb);
 729         ENTRY;
 730
 731 #ifdef CONFIG_FS_POSIX_ACL
 732         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 733                 et_fini(&sbi->ll_et);
 734                 rct_fini(&sbi->ll_rct);
 735         }
 736 #endif
 737
 738         ll_close_thread_shutdown(sbi->ll_lcq);
 739
 740         cl_sb_fini(sb);
 741
 742         list_del(&sbi->ll_conn_chain);
 743
 744         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 745         obd_disconnect(sbi->ll_dt_exp);
 746         sbi->ll_dt_exp = NULL;
 747         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 748          * see LU-2543. */
 749         obd_zombie_barrier();
 750
 751         lprocfs_unregister_mountpoint(sbi);
 752
 753         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 754         obd_disconnect(sbi->ll_md_exp);
 755         sbi->ll_md_exp = NULL;
 756
 757         EXIT;
 758 }
 759
 760 void ll_kill_super(struct super_block *sb)
 761 {
 762         struct ll_sb_info *sbi;
 763
 764         ENTRY;
 765
 766         /* not init sb ?*/
 767         if (!(sb->s_flags & MS_ACTIVE))
 768                 return;
 769
 770         sbi = ll_s2sbi(sb);
 771         /* we need restore s_dev from changed for clustred NFS before put_super
 772          * because new kernels have cached s_dev and change sb->s_dev in
 773          * put_super not affected real removing devices */
 774         if (sbi) {
 775                 sb->s_dev = sbi->ll_sdev_orig;
 776                 sbi->ll_umounting = 1;
 777         }
 778         EXIT;
 779 }
 780
 781 static inline int ll_set_opt(const char *opt, char *data, int fl)
 782 {
 783         if (strncmp(opt, data, strlen(opt)) != 0)
 784                 return(0);
 785         else
 786                 return(fl);
 787 }
 788
 789 /* non-client-specific mount options are parsed in lmd_parse */
 790 static int ll_options(char *options, int *flags)
 791 {
 792         int tmp;
 793         char *s1 = options, *s2;
 794         ENTRY;
 795
 796         if (!options)
 797                 RETURN(0);
 798
 799         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 800
 801         while (*s1) {
 802                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 803                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 804                 if (tmp) {
 805                         *flags |= tmp;
 806                         goto next;
 807                 }
 808                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 809                 if (tmp) {
 810                         *flags |= tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 819                 if (tmp) {
 820                         *flags &= ~tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 824                 if (tmp) {
 825                         *flags |= tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 829                 if (tmp) {
 830                         *flags &= ~tmp;
 831                         goto next;
 832                 }
 833 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
 834                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 835                 if (tmp) {
 836                         /* Ignore deprecated mount option.  The client will
 837                          * always try to mount with ACL support, whether this
 838                          * is used depends on whether server supports it. */
 839                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 840                                                   "mount option 'acl'.\n");
 841                         goto next;
 842                 }
 843                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 844                 if (tmp) {
 845                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 846                                                   "mount option 'noacl'.\n");
 847                         goto next;
 848                 }
 849 #else
 850 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 851 #endif
 852                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867
 868                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 869                 if (tmp) {
 870                         *flags |= tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 874                 if (tmp) {
 875                         *flags &= ~tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 879                 if (tmp) {
 880                         *flags |= tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 884                 if (tmp) {
 885                         *flags &= ~tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 889                 if (tmp) {
 890                         *flags |= tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 894                 if (tmp) {
 895                         *flags &= ~tmp;
 896                         goto next;
 897                 }
 898                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 899                 if (tmp) {
 900                         *flags |= tmp;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 904                 if (tmp) {
 905                         *flags |= tmp;
 906                         goto next;
 907                 }
 908                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 909                 if (tmp) {
 910                         *flags |= tmp;
 911                         goto next;
 912                 }
 913                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 914                 if (tmp) {
 915                         *flags &= ~tmp;
 916                         goto next;
 917                 }
 918                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 919                                    s1);
 920                 RETURN(-EINVAL);
 921
 922 next:
 923                 /* Find next opt */
 924                 s2 = strchr(s1, ',');
 925                 if (s2 == NULL)
 926                         break;
 927                 s1 = s2 + 1;
 928         }
 929         RETURN(0);
 930 }
 931
 932 void ll_lli_init(struct ll_inode_info *lli)
 933 {
 934         lli->lli_inode_magic = LLI_INODE_MAGIC;
 935         lli->lli_flags = 0;
 936         lli->lli_ioepoch = 0;
 937         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 938         spin_lock_init(&lli->lli_lock);
 939         lli->lli_posix_acl = NULL;
 940         lli->lli_remote_perms = NULL;
 941         mutex_init(&lli->lli_rmtperm_mutex);
 942         /* Do not set lli_fid, it has been initialized already. */
 943         fid_zero(&lli->lli_pfid);
 944         INIT_LIST_HEAD(&lli->lli_close_list);
 945         INIT_LIST_HEAD(&lli->lli_oss_capas);
 946         atomic_set(&lli->lli_open_count, 0);
 947         lli->lli_mds_capa = NULL;
 948         lli->lli_rmtperm_time = 0;
 949         lli->lli_pending_och = NULL;
 950         lli->lli_mds_read_och = NULL;
 951         lli->lli_mds_write_och = NULL;
 952         lli->lli_mds_exec_och = NULL;
 953         lli->lli_open_fd_read_count = 0;
 954         lli->lli_open_fd_write_count = 0;
 955         lli->lli_open_fd_exec_count = 0;
 956         mutex_init(&lli->lli_och_mutex);
 957         spin_lock_init(&lli->lli_agl_lock);
 958         lli->lli_has_smd = false;
 959         spin_lock_init(&lli->lli_layout_lock);
 960         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 961         lli->lli_clob = NULL;
 962
 963         init_rwsem(&lli->lli_xattrs_list_rwsem);
 964         mutex_init(&lli->lli_xattrs_enq_lock);
 965
 966         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 967         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 968                 mutex_init(&lli->lli_readdir_mutex);
 969                 lli->lli_opendir_key = NULL;
 970                 lli->lli_sai = NULL;
 971                 spin_lock_init(&lli->lli_sa_lock);
 972                 lli->lli_opendir_pid = 0;
 973         } else {
 974                 mutex_init(&lli->lli_size_mutex);
 975                 lli->lli_symlink_name = NULL;
 976                 init_rwsem(&lli->lli_trunc_sem);
 977                 mutex_init(&lli->lli_write_mutex);
 978                 init_rwsem(&lli->lli_glimpse_sem);
 979                 lli->lli_glimpse_time = 0;
 980                 INIT_LIST_HEAD(&lli->lli_agl_list);
 981                 lli->lli_agl_index = 0;
 982                 lli->lli_async_rc = 0;
 983         }
 984         mutex_init(&lli->lli_layout_mutex);
 985 }
 986
 987 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 988 {
 989         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 990
 991         bdi->name = "lustre";
 992         return bdi_register(bdi, NULL, "lustre-%d",
 993                             atomic_inc_return(&ll_bdi_num));
 994 }
 995
 996 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 997 {
 998         struct lustre_profile *lprof = NULL;
 999         struct lustre_sb_info *lsi = s2lsi(sb);
1000         struct ll_sb_info *sbi;
1001         char  *dt = NULL, *md = NULL;
1002         char  *profilenm = get_profile_name(sb);
1003         struct config_llog_instance *cfg;
1004         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1005         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1006         int    err;
1007         ENTRY;
1008
1009         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1010
1011         OBD_ALLOC_PTR(cfg);
1012         if (cfg == NULL)
1013                 RETURN(-ENOMEM);
1014
1015         try_module_get(THIS_MODULE);
1016
1017         /* client additional sb info */
1018         lsi->lsi_llsbi = sbi = ll_init_sbi();
1019         if (!sbi) {
1020                 module_put(THIS_MODULE);
1021                 OBD_FREE_PTR(cfg);
1022                 RETURN(-ENOMEM);
1023         }
1024
1025         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1026         if (err)
1027                 GOTO(out_free, err);
1028
1029         err = bdi_init(&lsi->lsi_bdi);
1030         if (err)
1031                 GOTO(out_free, err);
1032         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1033         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1034         err = ll_bdi_register(&lsi->lsi_bdi);
1035         if (err)
1036                 GOTO(out_free, err);
1037
1038         sb->s_bdi = &lsi->lsi_bdi;
1039 #ifndef HAVE_DCACHE_LOCK
1040         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1041         sb->s_d_op = &ll_d_ops;
1042 #endif
1043
1044         /* Generate a string unique to this super, in case some joker tries
1045            to mount the same fs at two mount points.
1046            Use the address of the super itself.*/
1047         cfg->cfg_instance = sb;
1048         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1049         cfg->cfg_callback = class_config_llog_handler;
1050         /* set up client obds */
1051         err = lustre_process_log(sb, profilenm, cfg);
1052         if (err < 0)
1053                 GOTO(out_free, err);
1054
1055         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1056         lprof = class_get_profile(profilenm);
1057         if (lprof == NULL) {
1058                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1059                                    " read from the MGS.  Does that filesystem "
1060                                    "exist?\n", profilenm);
1061                 GOTO(out_free, err = -EINVAL);
1062         }
1063         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1064                lprof->lp_md, lprof->lp_dt);
1065
1066         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1067         if (!dt)
1068                 GOTO(out_free, err = -ENOMEM);
1069         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1070
1071         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1072         if (!md)
1073                 GOTO(out_free, err = -ENOMEM);
1074         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1075
1076         /* connections, registrations, sb setup */
1077         err = client_common_fill_super(sb, md, dt, mnt);
1078
1079 out_free:
1080         if (md)
1081                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1082         if (dt)
1083                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1084         if (err)
1085                 ll_put_super(sb);
1086         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1087                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1088
1089         OBD_FREE_PTR(cfg);
1090         RETURN(err);
1091 } /* ll_fill_super */
1092
1093 void ll_put_super(struct super_block *sb)
1094 {
1095         struct config_llog_instance cfg, params_cfg;
1096         struct obd_device *obd;
1097         struct lustre_sb_info *lsi = s2lsi(sb);
1098         struct ll_sb_info *sbi = ll_s2sbi(sb);
1099         char *profilenm = get_profile_name(sb);
1100         int ccc_count, next, force = 1, rc = 0;
1101         ENTRY;
1102
1103         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1104
1105         ll_print_capa_stat(sbi);
1106
1107         cfg.cfg_instance = sb;
1108         lustre_end_log(sb, profilenm, &cfg);
1109
1110         params_cfg.cfg_instance = sb;
1111         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1112
1113         if (sbi->ll_md_exp) {
1114                 obd = class_exp2obd(sbi->ll_md_exp);
1115                 if (obd)
1116                         force = obd->obd_force;
1117         }
1118
1119         /* Wait for unstable pages to be committed to stable storage */
1120         if (force == 0) {
1121                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1122                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1123                         atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1124                         &lwi);
1125         }
1126
1127         ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1128         if (force == 0 && rc != -EINTR)
1129                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1130
1131
1132         /* We need to set force before the lov_disconnect in
1133            lustre_common_put_super, since l_d cleans up osc's as well. */
1134         if (force) {
1135                 next = 0;
1136                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1137                                                      &next)) != NULL) {
1138                         obd->obd_force = force;
1139                 }
1140         }
1141
1142         if (sbi->ll_lcq) {
1143                 /* Only if client_common_fill_super succeeded */
1144                 client_common_put_super(sb);
1145         }
1146
1147         next = 0;
1148         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1149                 class_manual_cleanup(obd);
1150         }
1151
1152         if (sbi->ll_flags & LL_SBI_VERBOSE)
1153                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1154
1155         if (profilenm)
1156                 class_del_profile(profilenm);
1157
1158         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1159                 bdi_destroy(&lsi->lsi_bdi);
1160                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1161         }
1162
1163         ll_free_sbi(sb);
1164         lsi->lsi_llsbi = NULL;
1165
1166         lustre_common_put_super(sb);
1167
1168         cl_env_cache_purge(~0);
1169
1170         module_put(THIS_MODULE);
1171
1172         EXIT;
1173 } /* client_put_super */
1174
1175 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1176 {
1177         struct inode *inode = NULL;
1178
1179         /* NOTE: we depend on atomic igrab() -bzzz */
1180         lock_res_and_lock(lock);
1181         if (lock->l_resource->lr_lvb_inode) {
1182                 struct ll_inode_info * lli;
1183                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1184                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1185                         inode = igrab(lock->l_resource->lr_lvb_inode);
1186                 } else {
1187                         inode = lock->l_resource->lr_lvb_inode;
1188                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1189                                          D_WARNING, lock, "lr_lvb_inode %p is "
1190                                          "bogus: magic %08x",
1191                                          lock->l_resource->lr_lvb_inode,
1192                                          lli->lli_inode_magic);
1193                         inode = NULL;
1194                 }
1195         }
1196         unlock_res_and_lock(lock);
1197         return inode;
1198 }
1199
1200 static void ll_dir_clear_lsm_md(struct inode *inode)
1201 {
1202         struct ll_inode_info *lli = ll_i2info(inode);
1203
1204         LASSERT(S_ISDIR(inode->i_mode));
1205
1206         if (lli->lli_lsm_md != NULL) {
1207                 lmv_free_memmd(lli->lli_lsm_md);
1208                 lli->lli_lsm_md = NULL;
1209         }
1210 }
1211
1212 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1213                                       const struct lu_fid *fid,
1214                                       struct lustre_md *md)
1215 {
1216         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1217         struct mdt_body         *body = md->body;
1218         struct inode            *inode;
1219         ino_t                   ino;
1220         ENTRY;
1221
1222         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1223         inode = iget_locked(sb, ino);
1224         if (inode == NULL) {
1225                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1226                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1227                 RETURN(ERR_PTR(-ENOENT));
1228         }
1229
1230         if (inode->i_state & I_NEW) {
1231                 struct ll_inode_info *lli = ll_i2info(inode);
1232                 struct lmv_stripe_md *lsm = md->lmv;
1233
1234                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1235                                 (body->mode & S_IFMT);
1236                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1237                          PFID(fid));
1238
1239                 LTIME_S(inode->i_mtime) = 0;
1240                 LTIME_S(inode->i_atime) = 0;
1241                 LTIME_S(inode->i_ctime) = 0;
1242                 inode->i_rdev = 0;
1243
1244                 /* initializing backing dev info. */
1245                 inode->i_mapping->backing_dev_info =
1246                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1247                 inode->i_op = &ll_dir_inode_operations;
1248                 inode->i_fop = &ll_dir_operations;
1249                 lli->lli_fid = *fid;
1250                 ll_lli_init(lli);
1251
1252                 LASSERT(lsm != NULL);
1253                 /* master stripe FID */
1254                 lli->lli_pfid = lsm->lsm_md_oinfo[0].lmo_fid;
1255                 CDEBUG(D_INODE, "lli %p master "DFID" slave "DFID"\n",
1256                        lli, PFID(fid), PFID(&lli->lli_pfid));
1257                 unlock_new_inode(inode);
1258         }
1259
1260         RETURN(inode);
1261 }
1262
1263 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1264 {
1265         struct lu_fid *fid;
1266         struct lmv_stripe_md *lsm = md->lmv;
1267         int i;
1268
1269         LASSERT(lsm != NULL);
1270         /* XXX sigh, this lsm_root initialization should be in
1271          * LMV layer, but it needs ll_iget right now, so we
1272          * put this here right now. */
1273         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1274                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1275                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1276                 if (i == 0) {
1277                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1278                 } else {
1279                         /* Unfortunately ll_iget will call ll_update_inode,
1280                          * where the initialization of slave inode is slightly
1281                          * different, so it reset lsm_md to NULL to avoid
1282                          * initializing lsm for slave inode. */
1283                         lsm->lsm_md_oinfo[i].lmo_root =
1284                                         ll_iget_anon_dir(inode->i_sb, fid, md);
1285                         if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1286                                 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1287
1288                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1289                                 return rc;
1290                         }
1291                 }
1292         }
1293
1294         /* Here is where the lsm is being initialized(fill lmo_info) after
1295          * client retrieve MD stripe information from MDT. */
1296         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1297                                 ll_md_blocking_ast);
1298 }
1299
1300 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1301                                 const struct lmv_stripe_md *lsm_md2)
1302 {
1303         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1304                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1305                lsm_md1->lsm_md_master_mdt_index ==
1306                                         lsm_md2->lsm_md_master_mdt_index &&
1307                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1308                lsm_md1->lsm_md_layout_version ==
1309                                         lsm_md2->lsm_md_layout_version &&
1310                strcmp(lsm_md1->lsm_md_pool_name,
1311                       lsm_md2->lsm_md_pool_name) == 0;
1312 }
1313
1314 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1315 {
1316         struct ll_inode_info *lli = ll_i2info(inode);
1317         struct lmv_stripe_md *lsm = md->lmv;
1318         int     idx;
1319         int     rc;
1320         ENTRY;
1321
1322         LASSERT(S_ISDIR(inode->i_mode));
1323         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1324                PFID(ll_inode2fid(inode)));
1325
1326         /* no striped information from request. */
1327         if (lsm == NULL) {
1328                 if (lli->lli_lsm_md == NULL) {
1329                         RETURN(0);
1330                 } else if (lli->lli_lsm_md->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1331                         /* migration is done, the temporay MIGRATE layout has
1332                          * been removed */
1333                         CDEBUG(D_INODE, DFID" finish migration.\n",
1334                                PFID(ll_inode2fid(inode)));
1335                         lmv_free_memmd(lli->lli_lsm_md);
1336                         lli->lli_lsm_md = NULL;
1337                         RETURN(0);
1338                 } else {
1339                         /* The lustre_md from req does not include stripeEA,
1340                          * see ll_md_setattr */
1341                         RETURN(0);
1342                 }
1343         }
1344
1345         /* set the directory layout */
1346         if (lli->lli_lsm_md == NULL) {
1347
1348                 rc = ll_init_lsm_md(inode, md);
1349                 if (rc != 0)
1350                         RETURN(rc);
1351
1352                 lli->lli_lsm_md = lsm;
1353                 /* set lsm_md to NULL, so the following free lustre_md
1354                  * will not free this lsm */
1355                 md->lmv = NULL;
1356                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1357                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1358                 RETURN(0);
1359         }
1360
1361         /* Compare the old and new stripe information */
1362         if (!lli_lsm_md_eq(lli->lli_lsm_md, lsm)) {
1363                 CERROR("inode %p %lu mismatch\n"
1364                        "    new(%p)     vs     lli_lsm_md(%p):\n"
1365                        "    magic:      %x                   %x\n"
1366                        "    count:      %x                   %x\n"
1367                        "    master:     %x                   %x\n"
1368                        "    hash_type:  %x                   %x\n"
1369                        "    layout:     %x                   %x\n"
1370                        "    pool:       %s                   %s\n",
1371                        inode, inode->i_ino, lsm, lli->lli_lsm_md,
1372                        lsm->lsm_md_magic, lli->lli_lsm_md->lsm_md_magic,
1373                        lsm->lsm_md_stripe_count,
1374                        lli->lli_lsm_md->lsm_md_stripe_count,
1375                        lsm->lsm_md_master_mdt_index,
1376                        lli->lli_lsm_md->lsm_md_master_mdt_index,
1377                        lsm->lsm_md_hash_type, lli->lli_lsm_md->lsm_md_hash_type,
1378                        lsm->lsm_md_layout_version,
1379                        lli->lli_lsm_md->lsm_md_layout_version,
1380                        lsm->lsm_md_pool_name,
1381                        lli->lli_lsm_md->lsm_md_pool_name);
1382                 RETURN(-EIO);
1383         }
1384
1385         for (idx = 0; idx < lli->lli_lsm_md->lsm_md_stripe_count; idx++) {
1386                 if (!lu_fid_eq(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid,
1387                                &lsm->lsm_md_oinfo[idx].lmo_fid)) {
1388                         CERROR("%s: FID in lsm mismatch idx %d, old: "DFID
1389                                "new:"DFID"\n",
1390                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1391                              PFID(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid),
1392                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1393                         RETURN(-EIO);
1394                 }
1395         }
1396
1397         rc = md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1398                               md->body, ll_md_blocking_ast);
1399
1400         RETURN(rc);
1401 }
1402
1403 void ll_clear_inode(struct inode *inode)
1404 {
1405         struct ll_inode_info *lli = ll_i2info(inode);
1406         struct ll_sb_info *sbi = ll_i2sbi(inode);
1407         ENTRY;
1408
1409         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1410                PFID(ll_inode2fid(inode)), inode);
1411
1412         if (S_ISDIR(inode->i_mode)) {
1413                 /* these should have been cleared in ll_file_release */
1414                 LASSERT(lli->lli_opendir_key == NULL);
1415                 LASSERT(lli->lli_sai == NULL);
1416                 LASSERT(lli->lli_opendir_pid == 0);
1417         }
1418
1419         spin_lock(&lli->lli_lock);
1420         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1421         spin_unlock(&lli->lli_lock);
1422         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1423
1424         LASSERT(!lli->lli_open_fd_write_count);
1425         LASSERT(!lli->lli_open_fd_read_count);
1426         LASSERT(!lli->lli_open_fd_exec_count);
1427
1428         if (lli->lli_mds_write_och)
1429                 ll_md_real_close(inode, FMODE_WRITE);
1430         if (lli->lli_mds_exec_och)
1431                 ll_md_real_close(inode, FMODE_EXEC);
1432         if (lli->lli_mds_read_och)
1433                 ll_md_real_close(inode, FMODE_READ);
1434
1435         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1436                 OBD_FREE(lli->lli_symlink_name,
1437                          strlen(lli->lli_symlink_name) + 1);
1438                 lli->lli_symlink_name = NULL;
1439         }
1440
1441         ll_xattr_cache_destroy(inode);
1442
1443         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1444                 LASSERT(lli->lli_posix_acl == NULL);
1445                 if (lli->lli_remote_perms) {
1446                         free_rmtperm_hash(lli->lli_remote_perms);
1447                         lli->lli_remote_perms = NULL;
1448                 }
1449         }
1450 #ifdef CONFIG_FS_POSIX_ACL
1451         else if (lli->lli_posix_acl) {
1452                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1453                 LASSERT(lli->lli_remote_perms == NULL);
1454                 posix_acl_release(lli->lli_posix_acl);
1455                 lli->lli_posix_acl = NULL;
1456         }
1457 #endif
1458         lli->lli_inode_magic = LLI_INODE_DEAD;
1459
1460         ll_clear_inode_capas(inode);
1461         if (S_ISDIR(inode->i_mode))
1462                 ll_dir_clear_lsm_md(inode);
1463         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1464                 LASSERT(list_empty(&lli->lli_agl_list));
1465
1466         /*
1467          * XXX This has to be done before lsm is freed below, because
1468          * cl_object still uses inode lsm.
1469          */
1470         cl_inode_fini(inode);
1471         lli->lli_has_smd = false;
1472
1473         EXIT;
1474 }
1475
1476 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1477                   struct md_open_data **mod)
1478 {
1479         struct lustre_md md;
1480         struct inode *inode = dentry->d_inode;
1481         struct ll_sb_info *sbi = ll_i2sbi(inode);
1482         struct ptlrpc_request *request = NULL;
1483         int rc, ia_valid;
1484         ENTRY;
1485
1486         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1487                                      LUSTRE_OPC_ANY, NULL);
1488         if (IS_ERR(op_data))
1489                 RETURN(PTR_ERR(op_data));
1490
1491         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1492                         &request, mod);
1493         if (rc) {
1494                 ptlrpc_req_finished(request);
1495                 if (rc == -ENOENT) {
1496                         clear_nlink(inode);
1497                         /* Unlinked special device node? Or just a race?
1498                          * Pretend we done everything. */
1499                         if (!S_ISREG(inode->i_mode) &&
1500                             !S_ISDIR(inode->i_mode)) {
1501                                 ia_valid = op_data->op_attr.ia_valid;
1502                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1503                                 rc = simple_setattr(dentry, &op_data->op_attr);
1504                                 op_data->op_attr.ia_valid = ia_valid;
1505                         }
1506                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1507                         CERROR("md_setattr fails: rc = %d\n", rc);
1508                 }
1509                 RETURN(rc);
1510         }
1511
1512         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1513                               sbi->ll_md_exp, &md);
1514         if (rc) {
1515                 ptlrpc_req_finished(request);
1516                 RETURN(rc);
1517         }
1518
1519         ia_valid = op_data->op_attr.ia_valid;
1520         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1521          * cache is not cleared yet. */
1522         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1523         rc = simple_setattr(dentry, &op_data->op_attr);
1524         op_data->op_attr.ia_valid = ia_valid;
1525
1526         /* Extract epoch data if obtained. */
1527         op_data->op_handle = md.body->handle;
1528         op_data->op_ioepoch = md.body->ioepoch;
1529
1530         rc = ll_update_inode(inode, &md);
1531         ptlrpc_req_finished(request);
1532
1533         RETURN(rc);
1534 }
1535
1536 /* Close IO epoch and send Size-on-MDS attribute update. */
1537 static int ll_setattr_done_writing(struct inode *inode,
1538                                    struct md_op_data *op_data,
1539                                    struct md_open_data *mod)
1540 {
1541         struct ll_inode_info *lli = ll_i2info(inode);
1542         int rc = 0;
1543         ENTRY;
1544
1545         LASSERT(op_data != NULL);
1546         if (!S_ISREG(inode->i_mode))
1547                 RETURN(0);
1548
1549         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1550                op_data->op_ioepoch, PFID(&lli->lli_fid));
1551
1552         op_data->op_flags = MF_EPOCH_CLOSE;
1553         ll_done_writing_attr(inode, op_data);
1554         ll_pack_inode2opdata(inode, op_data, NULL);
1555
1556         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1557         if (rc == -EAGAIN) {
1558                 /* MDS has instructed us to obtain Size-on-MDS attribute
1559                  * from OSTs and send setattr to back to MDS. */
1560                 rc = ll_som_update(inode, op_data);
1561         } else if (rc) {
1562                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1563                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1564                        PFID(ll_inode2fid(inode)), rc);
1565         }
1566         RETURN(rc);
1567 }
1568
1569 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1570 {
1571         struct obd_capa *capa;
1572         int rc;
1573
1574         if (attr->ia_valid & ATTR_SIZE)
1575                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1576         else
1577                 capa = ll_mdscapa_get(inode);
1578
1579         rc = cl_setattr_ost(inode, attr, capa);
1580
1581         if (attr->ia_valid & ATTR_SIZE)
1582                 ll_truncate_free_capa(capa);
1583         else
1584                 capa_put(capa);
1585
1586         return rc;
1587 }
1588
1589 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1590  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1591  * keep these values until such a time that objects are allocated for it.
1592  * We do the MDS operations first, as it is checking permissions for us.
1593  * We don't to the MDS RPC if there is nothing that we want to store there,
1594  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1595  * going to do an RPC anyways.
1596  *
1597  * If we are doing a truncate, we will send the mtime and ctime updates
1598  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1599  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1600  * at the same time.
1601  *
1602  * In case of HSMimport, we only set attr on MDS.
1603  */
1604 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1605 {
1606         struct inode *inode = dentry->d_inode;
1607         struct ll_inode_info *lli = ll_i2info(inode);
1608         struct md_op_data *op_data = NULL;
1609         struct md_open_data *mod = NULL;
1610         bool file_is_released = false;
1611         int rc = 0, rc1 = 0;
1612         ENTRY;
1613
1614         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1615                "valid %x, hsm_import %d\n",
1616                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1617                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1618                hsm_import);
1619
1620         if (attr->ia_valid & ATTR_SIZE) {
1621                 /* Check new size against VFS/VM file size limit and rlimit */
1622                 rc = inode_newsize_ok(inode, attr->ia_size);
1623                 if (rc)
1624                         RETURN(rc);
1625
1626                 /* The maximum Lustre file size is variable, based on the
1627                  * OST maximum object size and number of stripes.  This
1628                  * needs another check in addition to the VFS check above. */
1629                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1630                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1631                                PFID(&lli->lli_fid), attr->ia_size,
1632                                ll_file_maxbytes(inode));
1633                         RETURN(-EFBIG);
1634                 }
1635
1636                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1637         }
1638
1639         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1640         if (attr->ia_valid & TIMES_SET_FLAGS) {
1641                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1642                     !cfs_capable(CFS_CAP_FOWNER))
1643                         RETURN(-EPERM);
1644         }
1645
1646         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1647         if (attr->ia_valid & ATTR_CTIME) {
1648                 attr->ia_ctime = CFS_CURRENT_TIME;
1649                 attr->ia_valid |= ATTR_CTIME_SET;
1650         }
1651         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1652             (attr->ia_valid & ATTR_ATIME)) {
1653                 attr->ia_atime = CFS_CURRENT_TIME;
1654                 attr->ia_valid |= ATTR_ATIME_SET;
1655         }
1656         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1657             (attr->ia_valid & ATTR_MTIME)) {
1658                 attr->ia_mtime = CFS_CURRENT_TIME;
1659                 attr->ia_valid |= ATTR_MTIME_SET;
1660         }
1661
1662         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1663                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1664                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1665                        cfs_time_current_sec());
1666
1667         /* We always do an MDS RPC, even if we're only changing the size;
1668          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1669
1670         OBD_ALLOC_PTR(op_data);
1671         if (op_data == NULL)
1672                 RETURN(-ENOMEM);
1673
1674         if (!S_ISDIR(inode->i_mode)) {
1675                 if (attr->ia_valid & ATTR_SIZE)
1676                         inode_dio_write_done(inode);
1677                 mutex_unlock(&inode->i_mutex);
1678         }
1679
1680         /* truncate on a released file must failed with -ENODATA,
1681          * so size must not be set on MDS for released file
1682          * but other attributes must be set
1683          */
1684         if (S_ISREG(inode->i_mode)) {
1685                 struct lov_stripe_md *lsm;
1686                 __u32 gen;
1687
1688                 ll_layout_refresh(inode, &gen);
1689                 lsm = ccc_inode_lsm_get(inode);
1690                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1691                         file_is_released = true;
1692                 ccc_inode_lsm_put(inode, lsm);
1693
1694                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1695                         if (file_is_released) {
1696                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1697                                 if (rc < 0)
1698                                         GOTO(out, rc);
1699
1700                                 file_is_released = false;
1701                                 ll_layout_refresh(inode, &gen);
1702                         }
1703
1704                         /* If we are changing file size, file content is
1705                          * modified, flag it. */
1706                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1707                         spin_lock(&lli->lli_lock);
1708                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1709                         spin_unlock(&lli->lli_lock);
1710                         op_data->op_bias |= MDS_DATA_MODIFIED;
1711                 }
1712         }
1713
1714         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1715
1716         /* Open epoch for truncate. */
1717         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1718             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1719                 op_data->op_flags = MF_EPOCH_OPEN;
1720
1721         rc = ll_md_setattr(dentry, op_data, &mod);
1722         if (rc)
1723                 GOTO(out, rc);
1724
1725         /* RPC to MDT is sent, cancel data modification flag */
1726         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1727                 spin_lock(&lli->lli_lock);
1728                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1729                 spin_unlock(&lli->lli_lock);
1730         }
1731
1732         ll_ioepoch_open(lli, op_data->op_ioepoch);
1733         if (!S_ISREG(inode->i_mode) || file_is_released)
1734                 GOTO(out, rc = 0);
1735
1736         if (attr->ia_valid & (ATTR_SIZE |
1737                               ATTR_ATIME | ATTR_ATIME_SET |
1738                               ATTR_MTIME | ATTR_MTIME_SET)) {
1739                 /* For truncate and utimes sending attributes to OSTs, setting
1740                  * mtime/atime to the past will be performed under PW [0:EOF]
1741                  * extent lock (new_size:EOF for truncate).  It may seem
1742                  * excessive to send mtime/atime updates to OSTs when not
1743                  * setting times to past, but it is necessary due to possible
1744                  * time de-synchronization between MDT inode and OST objects */
1745                 if (attr->ia_valid & ATTR_SIZE)
1746                         down_write(&lli->lli_trunc_sem);
1747                 rc = ll_setattr_ost(inode, attr);
1748                 if (attr->ia_valid & ATTR_SIZE)
1749                         up_write(&lli->lli_trunc_sem);
1750         }
1751         EXIT;
1752 out:
1753         if (op_data) {
1754                 if (op_data->op_ioepoch) {
1755                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1756                         if (!rc)
1757                                 rc = rc1;
1758                 }
1759                 ll_finish_md_op_data(op_data);
1760         }
1761         if (!S_ISDIR(inode->i_mode)) {
1762                 mutex_lock(&inode->i_mutex);
1763                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1764                         inode_dio_wait(inode);
1765         }
1766
1767         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1768                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1769
1770         return rc;
1771 }
1772
1773 int ll_setattr(struct dentry *de, struct iattr *attr)
1774 {
1775         int mode = de->d_inode->i_mode;
1776
1777         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1778                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1779                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1780
1781         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1782                                (ATTR_SIZE|ATTR_MODE)) &&
1783             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1784              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1785               !(attr->ia_mode & S_ISGID))))
1786                 attr->ia_valid |= ATTR_FORCE;
1787
1788         if ((mode & S_ISUID) &&
1789             !(attr->ia_mode & S_ISUID) &&
1790             !(attr->ia_valid & ATTR_KILL_SUID))
1791                 attr->ia_valid |= ATTR_KILL_SUID;
1792
1793         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1794             !(attr->ia_mode & S_ISGID) &&
1795             !(attr->ia_valid & ATTR_KILL_SGID))
1796                 attr->ia_valid |= ATTR_KILL_SGID;
1797
1798         return ll_setattr_raw(de, attr, false);
1799 }
1800
1801 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1802                        __u64 max_age, __u32 flags)
1803 {
1804         struct ll_sb_info *sbi = ll_s2sbi(sb);
1805         struct obd_statfs obd_osfs;
1806         int rc;
1807         ENTRY;
1808
1809         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1810         if (rc) {
1811                 CERROR("md_statfs fails: rc = %d\n", rc);
1812                 RETURN(rc);
1813         }
1814
1815         osfs->os_type = sb->s_magic;
1816
1817         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1818                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1819
1820         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1821                 flags |= OBD_STATFS_NODELAY;
1822
1823         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1824         if (rc) {
1825                 CERROR("obd_statfs fails: rc = %d\n", rc);
1826                 RETURN(rc);
1827         }
1828
1829         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1830                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1831                obd_osfs.os_files);
1832
1833         osfs->os_bsize = obd_osfs.os_bsize;
1834         osfs->os_blocks = obd_osfs.os_blocks;
1835         osfs->os_bfree = obd_osfs.os_bfree;
1836         osfs->os_bavail = obd_osfs.os_bavail;
1837
1838         /* If we don't have as many objects free on the OST as inodes
1839          * on the MDS, we reduce the total number of inodes to
1840          * compensate, so that the "inodes in use" number is correct.
1841          */
1842         if (obd_osfs.os_ffree < osfs->os_ffree) {
1843                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1844                         obd_osfs.os_ffree;
1845                 osfs->os_ffree = obd_osfs.os_ffree;
1846         }
1847
1848         RETURN(rc);
1849 }
1850 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1851 {
1852         struct super_block *sb = de->d_sb;
1853         struct obd_statfs osfs;
1854         __u64 fsid = huge_encode_dev(sb->s_dev);
1855         int rc;
1856
1857         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1858         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1859
1860         /* Some amount of caching on the client is allowed */
1861         rc = ll_statfs_internal(sb, &osfs,
1862                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1863                                 0);
1864         if (rc)
1865                 return rc;
1866
1867         statfs_unpack(sfs, &osfs);
1868
1869         /* We need to downshift for all 32-bit kernels, because we can't
1870          * tell if the kernel is being called via sys_statfs64() or not.
1871          * Stop before overflowing f_bsize - in which case it is better
1872          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1873         if (sizeof(long) < 8) {
1874                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1875                         sfs->f_bsize <<= 1;
1876
1877                         osfs.os_blocks >>= 1;
1878                         osfs.os_bfree >>= 1;
1879                         osfs.os_bavail >>= 1;
1880                 }
1881         }
1882
1883         sfs->f_blocks = osfs.os_blocks;
1884         sfs->f_bfree = osfs.os_bfree;
1885         sfs->f_bavail = osfs.os_bavail;
1886         sfs->f_fsid.val[0] = (__u32)fsid;
1887         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1888         return 0;
1889 }
1890
1891 void ll_inode_size_lock(struct inode *inode)
1892 {
1893         struct ll_inode_info *lli;
1894
1895         LASSERT(!S_ISDIR(inode->i_mode));
1896
1897         lli = ll_i2info(inode);
1898         mutex_lock(&lli->lli_size_mutex);
1899 }
1900
1901 void ll_inode_size_unlock(struct inode *inode)
1902 {
1903         struct ll_inode_info *lli;
1904
1905         lli = ll_i2info(inode);
1906         mutex_unlock(&lli->lli_size_mutex);
1907 }
1908
1909 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1910 {
1911         struct ll_inode_info *lli = ll_i2info(inode);
1912         struct mdt_body *body = md->body;
1913         struct lov_stripe_md *lsm = md->lsm;
1914         struct ll_sb_info *sbi = ll_i2sbi(inode);
1915
1916         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1917         if (lsm != NULL) {
1918                 if (!lli->lli_has_smd &&
1919                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1920                         cl_file_inode_init(inode, md);
1921
1922                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1923                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1924                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1925         }
1926
1927         if (S_ISDIR(inode->i_mode)) {
1928                 int     rc;
1929
1930                 rc = ll_update_lsm_md(inode, md);
1931                 if (rc != 0)
1932                         return rc;
1933         }
1934
1935         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1936                 if (body->valid & OBD_MD_FLRMTPERM)
1937                         ll_update_remote_perm(inode, md->remote_perm);
1938         }
1939 #ifdef CONFIG_FS_POSIX_ACL
1940         else if (body->valid & OBD_MD_FLACL) {
1941                 spin_lock(&lli->lli_lock);
1942                 if (lli->lli_posix_acl)
1943                         posix_acl_release(lli->lli_posix_acl);
1944                 lli->lli_posix_acl = md->posix_acl;
1945                 spin_unlock(&lli->lli_lock);
1946         }
1947 #endif
1948         inode->i_ino = cl_fid_build_ino(&body->fid1,
1949                                         sbi->ll_flags & LL_SBI_32BIT_API);
1950         inode->i_generation = cl_fid_build_gen(&body->fid1);
1951
1952         if (body->valid & OBD_MD_FLATIME) {
1953                 if (body->atime > LTIME_S(inode->i_atime))
1954                         LTIME_S(inode->i_atime) = body->atime;
1955                 lli->lli_lvb.lvb_atime = body->atime;
1956         }
1957         if (body->valid & OBD_MD_FLMTIME) {
1958                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1959                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1960                                "to "LPU64"\n", inode->i_ino,
1961                                LTIME_S(inode->i_mtime), body->mtime);
1962                         LTIME_S(inode->i_mtime) = body->mtime;
1963                 }
1964                 lli->lli_lvb.lvb_mtime = body->mtime;
1965         }
1966         if (body->valid & OBD_MD_FLCTIME) {
1967                 if (body->ctime > LTIME_S(inode->i_ctime))
1968                         LTIME_S(inode->i_ctime) = body->ctime;
1969                 lli->lli_lvb.lvb_ctime = body->ctime;
1970         }
1971         if (body->valid & OBD_MD_FLMODE)
1972                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1973         if (body->valid & OBD_MD_FLTYPE)
1974                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1975         LASSERT(inode->i_mode != 0);
1976         if (S_ISREG(inode->i_mode)) {
1977                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1978         } else {
1979                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1980         }
1981         if (body->valid & OBD_MD_FLUID)
1982                 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1983         if (body->valid & OBD_MD_FLGID)
1984                 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1985         if (body->valid & OBD_MD_FLFLAGS)
1986                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1987         if (body->valid & OBD_MD_FLNLINK)
1988                 set_nlink(inode, body->nlink);
1989         if (body->valid & OBD_MD_FLRDEV)
1990                 inode->i_rdev = old_decode_dev(body->rdev);
1991
1992         if (body->valid & OBD_MD_FLID) {
1993                 /* FID shouldn't be changed! */
1994                 if (fid_is_sane(&lli->lli_fid)) {
1995                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1996                                  "Trying to change FID "DFID
1997                                  " to the "DFID", inode "DFID"(%p)\n",
1998                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1999                                  PFID(ll_inode2fid(inode)), inode);
2000                 } else {
2001                         lli->lli_fid = body->fid1;
2002                 }
2003         }
2004
2005         LASSERT(fid_seq(&lli->lli_fid) != 0);
2006
2007         if (body->valid & OBD_MD_FLSIZE) {
2008                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2009                     S_ISREG(inode->i_mode)) {
2010                         struct lustre_handle lockh;
2011                         ldlm_mode_t mode;
2012
2013                         /* As it is possible a blocking ast has been processed
2014                          * by this time, we need to check there is an UPDATE
2015                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2016                          * it. */
2017                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2018                                                &lockh, LDLM_FL_CBPENDING,
2019                                                LCK_CR | LCK_CW |
2020                                                LCK_PR | LCK_PW);
2021                         if (mode) {
2022                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2023                                                       LLIF_EPOCH_PENDING |
2024                                                       LLIF_SOM_DIRTY)) {
2025                                         CERROR("%s: inode "DFID" flags %u still"
2026                                                " has size authority! do not "
2027                                                "trust the size from MDS\n",
2028                                                sbi->ll_md_exp->exp_obd->obd_name,
2029                                                PFID(ll_inode2fid(inode)),
2030                                                lli->lli_flags);
2031                                 } else {
2032                                         /* Use old size assignment to avoid
2033                                          * deadlock bz14138 & bz14326 */
2034                                         i_size_write(inode, body->size);
2035                                         spin_lock(&lli->lli_lock);
2036                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2037                                         spin_unlock(&lli->lli_lock);
2038                                 }
2039                                 ldlm_lock_decref(&lockh, mode);
2040                         }
2041                 } else {
2042                         /* Use old size assignment to avoid
2043                          * deadlock bz14138 & bz14326 */
2044                         i_size_write(inode, body->size);
2045
2046                         CDEBUG(D_VFSTRACE,
2047                                "inode="DFID", updating i_size %llu\n",
2048                                PFID(ll_inode2fid(inode)),
2049                                (unsigned long long)body->size);
2050                 }
2051
2052                 if (body->valid & OBD_MD_FLBLOCKS)
2053                         inode->i_blocks = body->blocks;
2054         }
2055
2056         if (body->valid & OBD_MD_FLMDSCAPA) {
2057                 LASSERT(md->mds_capa);
2058                 ll_add_capa(inode, md->mds_capa);
2059         }
2060         if (body->valid & OBD_MD_FLOSSCAPA) {
2061                 LASSERT(md->oss_capa);
2062                 ll_add_capa(inode, md->oss_capa);
2063         }
2064
2065         if (body->valid & OBD_MD_TSTATE) {
2066                 if (body->t_state & MS_RESTORE)
2067                         lli->lli_flags |= LLIF_FILE_RESTORING;
2068         }
2069
2070         return 0;
2071 }
2072
2073 int ll_read_inode2(struct inode *inode, void *opaque)
2074 {
2075         struct lustre_md *md = opaque;
2076         struct ll_inode_info *lli = ll_i2info(inode);
2077         int     rc;
2078         ENTRY;
2079
2080         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2081                PFID(&lli->lli_fid), inode);
2082
2083         LASSERT(!lli->lli_has_smd);
2084
2085         /* Core attributes from the MDS first.  This is a new inode, and
2086          * the VFS doesn't zero times in the core inode so we have to do
2087          * it ourselves.  They will be overwritten by either MDS or OST
2088          * attributes - we just need to make sure they aren't newer. */
2089         LTIME_S(inode->i_mtime) = 0;
2090         LTIME_S(inode->i_atime) = 0;
2091         LTIME_S(inode->i_ctime) = 0;
2092         inode->i_rdev = 0;
2093         rc = ll_update_inode(inode, md);
2094         if (rc != 0)
2095                 RETURN(rc);
2096
2097         /* OIDEBUG(inode); */
2098
2099         /* initializing backing dev info. */
2100         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2101
2102
2103         if (S_ISREG(inode->i_mode)) {
2104                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2105                 inode->i_op = &ll_file_inode_operations;
2106                 inode->i_fop = sbi->ll_fop;
2107                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2108                 EXIT;
2109         } else if (S_ISDIR(inode->i_mode)) {
2110                 inode->i_op = &ll_dir_inode_operations;
2111                 inode->i_fop = &ll_dir_operations;
2112                 EXIT;
2113         } else if (S_ISLNK(inode->i_mode)) {
2114                 inode->i_op = &ll_fast_symlink_inode_operations;
2115                 EXIT;
2116         } else {
2117                 inode->i_op = &ll_special_inode_operations;
2118
2119                 init_special_inode(inode, inode->i_mode,
2120                                    inode->i_rdev);
2121
2122                 EXIT;
2123         }
2124
2125         return 0;
2126 }
2127
2128 void ll_delete_inode(struct inode *inode)
2129 {
2130         struct cl_inode_info *lli = cl_i2info(inode);
2131         ENTRY;
2132
2133         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2134                 /* discard all dirty pages before truncating them, required by
2135                  * osc_extent implementation at LU-1030. */
2136                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2137                                    CL_FSYNC_DISCARD, 1);
2138
2139         truncate_inode_pages(&inode->i_data, 0);
2140
2141         /* Workaround for LU-118 */
2142         if (inode->i_data.nrpages) {
2143                 spin_lock_irq(&inode->i_data.tree_lock);
2144                 spin_unlock_irq(&inode->i_data.tree_lock);
2145                 LASSERTF(inode->i_data.nrpages == 0,
2146                          "inode="DFID"(%p) nrpages=%lu, see "
2147                          "http://jira.whamcloud.com/browse/LU-118\n",
2148                          PFID(ll_inode2fid(inode)), inode,
2149                          inode->i_data.nrpages);
2150         }
2151         /* Workaround end */
2152
2153 #ifdef HAVE_SBOPS_EVICT_INODE
2154         ll_clear_inode(inode);
2155 #endif
2156         clear_inode(inode);
2157
2158         EXIT;
2159 }
2160
2161 int ll_iocontrol(struct inode *inode, struct file *file,
2162                  unsigned int cmd, unsigned long arg)
2163 {
2164         struct ll_sb_info *sbi = ll_i2sbi(inode);
2165         struct ptlrpc_request *req = NULL;
2166         int rc, flags = 0;
2167         ENTRY;
2168
2169         switch(cmd) {
2170         case FSFILT_IOC_GETFLAGS: {
2171                 struct mdt_body *body;
2172                 struct md_op_data *op_data;
2173
2174                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2175                                              0, 0, LUSTRE_OPC_ANY,
2176                                              NULL);
2177                 if (IS_ERR(op_data))
2178                         RETURN(PTR_ERR(op_data));
2179
2180                 op_data->op_valid = OBD_MD_FLFLAGS;
2181                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2182                 ll_finish_md_op_data(op_data);
2183                 if (rc) {
2184                         CERROR("%s: failure inode "DFID": rc = %d\n",
2185                                sbi->ll_md_exp->exp_obd->obd_name,
2186                                PFID(ll_inode2fid(inode)), rc);
2187                         RETURN(-abs(rc));
2188                 }
2189
2190                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2191
2192                 flags = body->flags;
2193
2194                 ptlrpc_req_finished(req);
2195
2196                 RETURN(put_user(flags, (int *)arg));
2197         }
2198         case FSFILT_IOC_SETFLAGS: {
2199                 struct lov_stripe_md *lsm;
2200                 struct obd_info oinfo = { { { 0 } } };
2201                 struct md_op_data *op_data;
2202
2203                 if (get_user(flags, (int *)arg))
2204                         RETURN(-EFAULT);
2205
2206                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2207                                              LUSTRE_OPC_ANY, NULL);
2208                 if (IS_ERR(op_data))
2209                         RETURN(PTR_ERR(op_data));
2210
2211                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2212                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2213                 rc = md_setattr(sbi->ll_md_exp, op_data,
2214                                 NULL, 0, NULL, 0, &req, NULL);
2215                 ll_finish_md_op_data(op_data);
2216                 ptlrpc_req_finished(req);
2217                 if (rc)
2218                         RETURN(rc);
2219
2220                 inode->i_flags = ll_ext_to_inode_flags(flags);
2221
2222                 lsm = ccc_inode_lsm_get(inode);
2223                 if (!lsm_has_objects(lsm)) {
2224                         ccc_inode_lsm_put(inode, lsm);
2225                         RETURN(0);
2226                 }
2227
2228                 OBDO_ALLOC(oinfo.oi_oa);
2229                 if (!oinfo.oi_oa) {
2230                         ccc_inode_lsm_put(inode, lsm);
2231                         RETURN(-ENOMEM);
2232                 }
2233                 oinfo.oi_md = lsm;
2234                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2235                 oinfo.oi_oa->o_flags = flags;
2236                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2237                                        OBD_MD_FLGROUP;
2238                 oinfo.oi_capa = ll_mdscapa_get(inode);
2239                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2240                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2241                 capa_put(oinfo.oi_capa);
2242                 OBDO_FREE(oinfo.oi_oa);
2243                 ccc_inode_lsm_put(inode, lsm);
2244
2245                 if (rc && rc != -EPERM && rc != -EACCES)
2246                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2247
2248                 RETURN(rc);
2249         }
2250         default:
2251                 RETURN(-ENOSYS);
2252         }
2253
2254         RETURN(0);
2255 }
2256
2257 int ll_flush_ctx(struct inode *inode)
2258 {
2259         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2260
2261         CDEBUG(D_SEC, "flush context for user %d\n",
2262                from_kuid(&init_user_ns, current_uid()));
2263
2264         obd_set_info_async(NULL, sbi->ll_md_exp,
2265                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2266                            0, NULL, NULL);
2267         obd_set_info_async(NULL, sbi->ll_dt_exp,
2268                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2269                            0, NULL, NULL);
2270         return 0;
2271 }
2272
2273 /* umount -f client means force down, don't save state */
2274 void ll_umount_begin(struct super_block *sb)
2275 {
2276         struct ll_sb_info *sbi = ll_s2sbi(sb);
2277         struct obd_device *obd;
2278         struct obd_ioctl_data *ioc_data;
2279         ENTRY;
2280
2281         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2282                sb->s_count, atomic_read(&sb->s_active));
2283
2284         obd = class_exp2obd(sbi->ll_md_exp);
2285         if (obd == NULL) {
2286                 CERROR("Invalid MDC connection handle "LPX64"\n",
2287                        sbi->ll_md_exp->exp_handle.h_cookie);
2288                 EXIT;
2289                 return;
2290         }
2291         obd->obd_force = 1;
2292
2293         obd = class_exp2obd(sbi->ll_dt_exp);
2294         if (obd == NULL) {
2295                 CERROR("Invalid LOV connection handle "LPX64"\n",
2296                        sbi->ll_dt_exp->exp_handle.h_cookie);
2297                 EXIT;
2298                 return;
2299         }
2300         obd->obd_force = 1;
2301
2302         OBD_ALLOC_PTR(ioc_data);
2303         if (ioc_data) {
2304                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2305                               sizeof *ioc_data, ioc_data, NULL);
2306
2307                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2308                               sizeof *ioc_data, ioc_data, NULL);
2309
2310                 OBD_FREE_PTR(ioc_data);
2311         }
2312
2313         /* Really, we'd like to wait until there are no requests outstanding,
2314          * and then continue.  For now, we just invalidate the requests,
2315          * schedule() and sleep one second if needed, and hope.
2316          */
2317         schedule();
2318         EXIT;
2319 }
2320
2321 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2322 {
2323         struct ll_sb_info *sbi = ll_s2sbi(sb);
2324         char *profilenm = get_profile_name(sb);
2325         int err;
2326         __u32 read_only;
2327
2328         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2329                 read_only = *flags & MS_RDONLY;
2330                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2331                                          sizeof(KEY_READ_ONLY),
2332                                          KEY_READ_ONLY, sizeof(read_only),
2333                                          &read_only, NULL);
2334                 if (err) {
2335                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2336                                       profilenm, read_only ?
2337                                       "read-only" : "read-write", err);
2338                         return err;
2339                 }
2340
2341                 if (read_only)
2342                         sb->s_flags |= MS_RDONLY;
2343                 else
2344                         sb->s_flags &= ~MS_RDONLY;
2345
2346                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2347                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2348                                       read_only ?  "read-only" : "read-write");
2349         }
2350         return 0;
2351 }
2352
2353 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2354                   struct super_block *sb, struct lookup_intent *it)
2355 {
2356         struct ll_sb_info *sbi = NULL;
2357         struct lustre_md md = { 0 };
2358         int rc;
2359         ENTRY;
2360
2361         LASSERT(*inode || sb);
2362         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2363         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2364                               sbi->ll_md_exp, &md);
2365         if (rc)
2366                 RETURN(rc);
2367
2368         if (*inode) {
2369                 rc = ll_update_inode(*inode, &md);
2370                 if (rc != 0)
2371                         GOTO(out, rc);
2372         } else {
2373                 LASSERT(sb != NULL);
2374
2375                 /*
2376                  * At this point server returns to client's same fid as client
2377                  * generated for creating. So using ->fid1 is okay here.
2378                  */
2379                 LASSERT(fid_is_sane(&md.body->fid1));
2380
2381                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2382                                              sbi->ll_flags & LL_SBI_32BIT_API),
2383                                  &md);
2384                 if (IS_ERR(*inode)) {
2385 #ifdef CONFIG_FS_POSIX_ACL
2386                         if (md.posix_acl) {
2387                                 posix_acl_release(md.posix_acl);
2388                                 md.posix_acl = NULL;
2389                         }
2390 #endif
2391                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2392                         *inode = NULL;
2393                         CERROR("new_inode -fatal: rc %d\n", rc);
2394                         GOTO(out, rc);
2395                 }
2396         }
2397
2398         /* Handling piggyback layout lock.
2399          * Layout lock can be piggybacked by getattr and open request.
2400          * The lsm can be applied to inode only if it comes with a layout lock
2401          * otherwise correct layout may be overwritten, for example:
2402          * 1. proc1: mdt returns a lsm but not granting layout
2403          * 2. layout was changed by another client
2404          * 3. proc2: refresh layout and layout lock granted
2405          * 4. proc1: to apply a stale layout */
2406         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2407                 struct lustre_handle lockh;
2408                 struct ldlm_lock *lock;
2409
2410                 lockh.cookie = it->d.lustre.it_lock_handle;
2411                 lock = ldlm_handle2lock(&lockh);
2412                 LASSERT(lock != NULL);
2413                 if (ldlm_has_layout(lock)) {
2414                         struct cl_object_conf conf;
2415
2416                         memset(&conf, 0, sizeof(conf));
2417                         conf.coc_opc = OBJECT_CONF_SET;
2418                         conf.coc_inode = *inode;
2419                         conf.coc_lock = lock;
2420                         conf.u.coc_md = &md;
2421                         (void)ll_layout_conf(*inode, &conf);
2422                 }
2423                 LDLM_LOCK_PUT(lock);
2424         }
2425
2426 out:
2427         if (md.lsm != NULL)
2428                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2429         md_free_lustre_md(sbi->ll_md_exp, &md);
2430         RETURN(rc);
2431 }
2432
2433 int ll_obd_statfs(struct inode *inode, void *arg)
2434 {
2435         struct ll_sb_info *sbi = NULL;
2436         struct obd_export *exp;
2437         char *buf = NULL;
2438         struct obd_ioctl_data *data = NULL;
2439         __u32 type;
2440         __u32 flags;
2441         int len = 0, rc;
2442
2443         if (!inode || !(sbi = ll_i2sbi(inode)))
2444                 GOTO(out_statfs, rc = -EINVAL);
2445
2446         rc = obd_ioctl_getdata(&buf, &len, arg);
2447         if (rc)
2448                 GOTO(out_statfs, rc);
2449
2450         data = (void*)buf;
2451         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2452             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2453                 GOTO(out_statfs, rc = -EINVAL);
2454
2455         if (data->ioc_inllen1 != sizeof(__u32) ||
2456             data->ioc_inllen2 != sizeof(__u32) ||
2457             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2458             data->ioc_plen2 != sizeof(struct obd_uuid))
2459                 GOTO(out_statfs, rc = -EINVAL);
2460
2461         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2462         if (type & LL_STATFS_LMV)
2463                 exp = sbi->ll_md_exp;
2464         else if (type & LL_STATFS_LOV)
2465                 exp = sbi->ll_dt_exp;
2466         else
2467                 GOTO(out_statfs, rc = -ENODEV);
2468
2469         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2470         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2471         if (rc)
2472                 GOTO(out_statfs, rc);
2473 out_statfs:
2474         if (buf)
2475                 obd_ioctl_freedata(buf, len);
2476         return rc;
2477 }
2478
2479 int ll_process_config(struct lustre_cfg *lcfg)
2480 {
2481         struct super_block *sb;
2482         unsigned long x;
2483         int rc = 0;
2484         char *ptr;
2485
2486         /* The instance name contains the sb: lustre-client-aacfe000 */
2487         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2488         if (!ptr || !*(++ptr))
2489                 return -EINVAL;
2490         if (sscanf(ptr, "%lx", &x) != 1)
2491                 return -EINVAL;
2492         sb = (struct super_block *)x;
2493         /* This better be a real Lustre superblock! */
2494         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2495
2496         /* Note we have not called client_common_fill_super yet, so
2497            proc fns must be able to handle that! */
2498         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2499                                           lcfg, sb);
2500         if (rc > 0)
2501                 rc = 0;
2502         return rc;
2503 }
2504
2505 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2506 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2507                                        struct inode *i1, struct inode *i2,
2508                                        const char *name, int namelen,
2509                                        int mode, __u32 opc, void *data)
2510 {
2511         LASSERT(i1 != NULL);
2512
2513         if (namelen > ll_i2sbi(i1)->ll_namelen)
2514                 return ERR_PTR(-ENAMETOOLONG);
2515
2516         if (op_data == NULL)
2517                 OBD_ALLOC_PTR(op_data);
2518
2519         if (op_data == NULL)
2520                 return ERR_PTR(-ENOMEM);
2521
2522         ll_i2gids(op_data->op_suppgids, i1, i2);
2523         op_data->op_fid1 = *ll_inode2fid(i1);
2524         op_data->op_capa1 = ll_mdscapa_get(i1);
2525         if (S_ISDIR(i1->i_mode))
2526                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2527
2528         if (i2) {
2529                 op_data->op_fid2 = *ll_inode2fid(i2);
2530                 op_data->op_capa2 = ll_mdscapa_get(i2);
2531                 if (S_ISDIR(i2->i_mode))
2532                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2533         } else {
2534                 fid_zero(&op_data->op_fid2);
2535                 op_data->op_capa2 = NULL;
2536         }
2537
2538         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2539                 op_data->op_cli_flags |= CLI_HASH64;
2540
2541         if (ll_need_32bit_api(ll_i2sbi(i1)))
2542                 op_data->op_cli_flags |= CLI_API32;
2543
2544         op_data->op_name = name;
2545         op_data->op_namelen = namelen;
2546         op_data->op_mode = mode;
2547         op_data->op_mod_time = cfs_time_current_sec();
2548         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2549         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2550         op_data->op_cap = cfs_curproc_cap_pack();
2551         op_data->op_bias = 0;
2552         op_data->op_cli_flags = 0;
2553         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2554              filename_is_volatile(name, namelen, NULL))
2555                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2556         op_data->op_mds = 0;
2557         op_data->op_data = data;
2558
2559         /* If the file is being opened after mknod() (normally due to NFS)
2560          * try to use the default stripe data from parent directory for
2561          * allocating OST objects.  Try to pass the parent FID to MDS. */
2562         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2563             !ll_i2info(i2)->lli_has_smd) {
2564                 struct ll_inode_info *lli = ll_i2info(i2);
2565
2566                 spin_lock(&lli->lli_lock);
2567                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2568                         op_data->op_fid1 = lli->lli_pfid;
2569                 spin_unlock(&lli->lli_lock);
2570                 /** We ignore parent's capability temporary. */
2571         }
2572
2573         /* When called by ll_setattr_raw, file is i1. */
2574         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2575                 op_data->op_bias |= MDS_DATA_MODIFIED;
2576
2577         return op_data;
2578 }
2579
2580 void ll_finish_md_op_data(struct md_op_data *op_data)
2581 {
2582         capa_put(op_data->op_capa1);
2583         capa_put(op_data->op_capa2);
2584         OBD_FREE_PTR(op_data);
2585 }
2586
2587 #ifdef HAVE_SUPEROPS_USE_DENTRY
2588 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2589 #else
2590 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2591 #endif
2592 {
2593         struct ll_sb_info *sbi;
2594
2595 #ifdef HAVE_SUPEROPS_USE_DENTRY
2596         LASSERT((seq != NULL) && (dentry != NULL));
2597         sbi = ll_s2sbi(dentry->d_sb);
2598 #else
2599         LASSERT((seq != NULL) && (vfs != NULL));
2600         sbi = ll_s2sbi(vfs->mnt_sb);
2601 #endif
2602
2603         if (sbi->ll_flags & LL_SBI_NOLCK)
2604                 seq_puts(seq, ",nolock");
2605
2606         if (sbi->ll_flags & LL_SBI_FLOCK)
2607                 seq_puts(seq, ",flock");
2608
2609         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2610                 seq_puts(seq, ",localflock");
2611
2612         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2613                 seq_puts(seq, ",user_xattr");
2614
2615         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2616                 seq_puts(seq, ",lazystatfs");
2617
2618         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2619                 seq_puts(seq, ",user_fid2path");
2620
2621         RETURN(0);
2622 }
2623
2624 /**
2625  * Get obd name by cmd, and copy out to user space
2626  */
2627 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2628 {
2629         struct ll_sb_info *sbi = ll_i2sbi(inode);
2630         struct obd_device *obd;
2631         ENTRY;
2632
2633         if (cmd == OBD_IOC_GETDTNAME)
2634                 obd = class_exp2obd(sbi->ll_dt_exp);
2635         else if (cmd == OBD_IOC_GETMDNAME)
2636                 obd = class_exp2obd(sbi->ll_md_exp);
2637         else
2638                 RETURN(-EINVAL);
2639
2640         if (!obd)
2641                 RETURN(-ENOENT);
2642
2643         if (copy_to_user((void *)arg, obd->obd_name,
2644                          strlen(obd->obd_name) + 1))
2645                 RETURN(-EFAULT);
2646
2647         RETURN(0);
2648 }
2649
2650 /**
2651  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2652  * fsname will be returned in this buffer; otherwise, a static buffer will be
2653  * used to store the fsname and returned to caller.
2654  */
2655 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2656 {
2657         static char fsname_static[MTI_NAME_MAXLEN];
2658         struct lustre_sb_info *lsi = s2lsi(sb);
2659         char *ptr;
2660         int len;
2661
2662         if (buf == NULL) {
2663                 /* this means the caller wants to use static buffer
2664                  * and it doesn't care about race. Usually this is
2665                  * in error reporting path */
2666                 buf = fsname_static;
2667                 buflen = sizeof(fsname_static);
2668         }
2669
2670         len = strlen(lsi->lsi_lmd->lmd_profile);
2671         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2672         if (ptr && (strcmp(ptr, "-client") == 0))
2673                 len -= 7;
2674
2675         if (unlikely(len >= buflen))
2676                 len = buflen - 1;
2677         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2678         buf[len] = '\0';
2679
2680         return buf;
2681 }
2682
2683 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2684 {
2685         char *path = NULL;
2686
2687         struct path p;
2688
2689         p.dentry = dentry;
2690         p.mnt = current->fs->root.mnt;
2691         path_get(&p);
2692         path = d_path(&p, buf, bufsize);
2693         path_put(&p);
2694         return path;
2695 }
2696
2697 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2698 {
2699         char *buf, *path = NULL;
2700         struct dentry *dentry = NULL;
2701         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2702
2703         /* this can be called inside spin lock so use GFP_ATOMIC. */
2704         buf = (char *)__get_free_page(GFP_ATOMIC);
2705         if (buf != NULL) {
2706                 dentry = d_find_alias(page->mapping->host);
2707                 if (dentry != NULL)
2708                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2709         }
2710
2711         CDEBUG(D_WARNING,
2712                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2713                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2714                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2715                PFID(&obj->cob_header.coh_lu.loh_fid),
2716                (path && !IS_ERR(path)) ? path : "", ioret);
2717
2718         if (dentry != NULL)
2719                 dput(dentry);
2720
2721         if (buf != NULL)
2722                 free_page((unsigned long)buf);
2723 }