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, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/statfs.h>
  45 #include <linux/time.h>
  46 #include <linux/types.h>
  47 #include <linux/version.h>
  48 #include <linux/mm.h>
  49 #include <linux/user_namespace.h>
  50 #ifdef HAVE_UIDGID_HEADER
  51 # include <linux/uidgid.h>
  52 #endif
  53
  54 #include <lustre_ioctl.h>
  55 #include <lustre_ha.h>
  56 #include <lustre_dlm.h>
  57 #include <lprocfs_status.h>
  58 #include <lustre_disk.h>
  59 #include <lustre_param.h>
  60 #include <lustre_log.h>
  61 #include <cl_object.h>
  62 #include <obd_cksum.h>
  63 #include "llite_internal.h"
  64
  65 struct kmem_cache *ll_file_data_slab;
  66
  67 #ifndef log2
  68 #define log2(n) ffz(~(n))
  69 #endif
  70
  71 static struct ll_sb_info *ll_init_sbi(void)
  72 {
  73         struct ll_sb_info *sbi = NULL;
  74         unsigned long pages;
  75         unsigned long lru_page_max;
  76         struct sysinfo si;
  77         class_uuid_t uuid;
  78         int i;
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(NULL);
  84
  85         spin_lock_init(&sbi->ll_lock);
  86         mutex_init(&sbi->ll_lco.lco_lock);
  87         spin_lock_init(&sbi->ll_pp_extent_lock);
  88         spin_lock_init(&sbi->ll_process_lock);
  89         sbi->ll_rw_stats_on = 0;
  90
  91         si_meminfo(&si);
  92         pages = si.totalram - si.totalhigh;
  93         lru_page_max = pages / 2;
  94
  95         /* initialize ll_cache data */
  96         sbi->ll_cache = cl_cache_init(lru_page_max);
  97         if (sbi->ll_cache == NULL) {
  98                 OBD_FREE(sbi, sizeof(*sbi));
  99                 RETURN(NULL);
 100         }
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107
 108         ll_generate_random_uuid(uuid);
 109         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 110         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 111
 112         sbi->ll_flags |= LL_SBI_VERBOSE;
 113 #ifdef ENABLE_CHECKSUM
 114         sbi->ll_flags |= LL_SBI_CHECKSUM;
 115 #endif
 116
 117 #ifdef HAVE_LRU_RESIZE_SUPPORT
 118         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 119 #endif
 120         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 121
 122         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 123                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 124                                pp_r_hist.oh_lock);
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_w_hist.oh_lock);
 127         }
 128
 129         /* metadata statahead is enabled by default */
 130         sbi->ll_sa_max = LL_SA_RPC_DEF;
 131         atomic_set(&sbi->ll_sa_total, 0);
 132         atomic_set(&sbi->ll_sa_wrong, 0);
 133         atomic_set(&sbi->ll_sa_running, 0);
 134         atomic_set(&sbi->ll_agl_total, 0);
 135         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 136         sbi->ll_flags |= LL_SBI_FAST_READ;
 137
 138         /* root squash */
 139         sbi->ll_squash.rsi_uid = 0;
 140         sbi->ll_squash.rsi_gid = 0;
 141         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 142         init_rwsem(&sbi->ll_squash.rsi_sem);
 143
 144         RETURN(sbi);
 145 }
 146
 147 static void ll_free_sbi(struct super_block *sb)
 148 {
 149         struct ll_sb_info *sbi = ll_s2sbi(sb);
 150         ENTRY;
 151
 152         if (sbi != NULL) {
 153                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 154                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 155                 if (sbi->ll_cache != NULL) {
 156                         cl_cache_decref(sbi->ll_cache);
 157                         sbi->ll_cache = NULL;
 158                 }
 159                 OBD_FREE(sbi, sizeof(*sbi));
 160         }
 161         EXIT;
 162 }
 163
 164 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 165                                     struct vfsmount *mnt)
 166 {
 167         struct inode *root = NULL;
 168         struct ll_sb_info *sbi = ll_s2sbi(sb);
 169         struct obd_device *obd;
 170         struct obd_statfs *osfs = NULL;
 171         struct ptlrpc_request *request = NULL;
 172         struct obd_connect_data *data = NULL;
 173         struct obd_uuid *uuid;
 174         struct md_op_data *op_data;
 175         struct lustre_md lmd;
 176         u64 valid;
 177         int size, err, checksum;
 178         ENTRY;
 179
 180         obd = class_name2obd(md);
 181         if (!obd) {
 182                 CERROR("MD %s: not setup or attached\n", md);
 183                 RETURN(-EINVAL);
 184         }
 185
 186         OBD_ALLOC_PTR(data);
 187         if (data == NULL)
 188                 RETURN(-ENOMEM);
 189
 190         OBD_ALLOC_PTR(osfs);
 191         if (osfs == NULL) {
 192                 OBD_FREE_PTR(data);
 193                 RETURN(-ENOMEM);
 194         }
 195
 196         /* indicate the features supported by this client */
 197         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 198                                   OBD_CONNECT_ATTRFID  |
 199                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 200                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 201                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 202                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 203                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 204                                   OBD_CONNECT_64BITHASH |
 205                                   OBD_CONNECT_EINPROGRESS |
 206                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 207                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 208                                   OBD_CONNECT_MAX_EASIZE |
 209                                   OBD_CONNECT_FLOCK_DEAD |
 210                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 211                                   OBD_CONNECT_OPEN_BY_FID |
 212                                   OBD_CONNECT_DIR_STRIPE |
 213                                   OBD_CONNECT_BULK_MBITS |
 214                                   OBD_CONNECT_SUBTREE |
 215                                   OBD_CONNECT_FLAGS2;
 216
 217         data->ocd_connect_flags2 = 0;
 218
 219 #ifdef HAVE_LRU_RESIZE_SUPPORT
 220         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 221                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 222 #endif
 223 #ifdef CONFIG_FS_POSIX_ACL
 224         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 225 #endif
 226
 227         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 228                 /* flag mdc connection as lightweight, only used for test
 229                  * purpose, use with care */
 230                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 231
 232         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 233         data->ocd_version = LUSTRE_VERSION_CODE;
 234
 235         if (sb->s_flags & MS_RDONLY)
 236                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 237         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 238                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 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         /* always ping even if server suppress_pings */
 248         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 249                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 250
 251         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 252
 253         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 254         if (err == -EBUSY) {
 255                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 256                                    "recovery, of which this client is not a "
 257                                    "part. Please wait for recovery to complete,"
 258                                    " abort, or time out.\n", md);
 259                 GOTO(out, err);
 260         } else if (err) {
 261                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 262                 GOTO(out, err);
 263         }
 264
 265         sbi->ll_md_exp->exp_connect_data = *data;
 266
 267         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 268                            LUSTRE_SEQ_METADATA);
 269         if (err) {
 270                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 271                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 272                 GOTO(out_md, err);
 273         }
 274
 275         /* For mount, we only need fs info from MDT0, and also in DNE, it
 276          * can make sure the client can be mounted as long as MDT0 is
 277          * avaible */
 278         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 279                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 280                         OBD_STATFS_FOR_MDT0);
 281         if (err)
 282                 GOTO(out_md_fid, err);
 283
 284         /* This needs to be after statfs to ensure connect has finished.
 285          * Note that "data" does NOT contain the valid connect reply.
 286          * If connecting to a 1.8 server there will be no LMV device, so
 287          * we can access the MDC export directly and exp_connect_flags will
 288          * be non-zero, but if accessing an upgraded 2.1 server it will
 289          * have the correct flags filled in.
 290          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 291         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 292         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 293             valid != CLIENT_CONNECT_MDT_REQD) {
 294                 char *buf;
 295
 296                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 297                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 298                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 299                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 300                                    "feature(s) needed for correct operation "
 301                                    "of this client (%s). Please upgrade "
 302                                    "server or downgrade client.\n",
 303                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 304                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 305                 GOTO(out_md_fid, err = -EPROTO);
 306         }
 307
 308         size = sizeof(*data);
 309         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 310                            KEY_CONN_DATA,  &size, data);
 311         if (err) {
 312                 CERROR("%s: Get connect data failed: rc = %d\n",
 313                        sbi->ll_md_exp->exp_obd->obd_name, err);
 314                 GOTO(out_md_fid, err);
 315         }
 316
 317         LASSERT(osfs->os_bsize);
 318         sb->s_blocksize = osfs->os_bsize;
 319         sb->s_blocksize_bits = log2(osfs->os_bsize);
 320         sb->s_magic = LL_SUPER_MAGIC;
 321         sb->s_maxbytes = MAX_LFS_FILESIZE;
 322         sbi->ll_namelen = osfs->os_namelen;
 323
 324         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 325             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 326                 LCONSOLE_INFO("Disabling user_xattr feature because "
 327                               "it is not supported on the server\n");
 328                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 329         }
 330
 331         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 332 #ifdef MS_POSIXACL
 333                 sb->s_flags |= MS_POSIXACL;
 334 #endif
 335                 sbi->ll_flags |= LL_SBI_ACL;
 336         } else {
 337                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 338 #ifdef MS_POSIXACL
 339                 sb->s_flags &= ~MS_POSIXACL;
 340 #endif
 341                 sbi->ll_flags &= ~LL_SBI_ACL;
 342         }
 343
 344         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 345                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 346
 347         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 348                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 349         else
 350                 sbi->ll_md_brw_pages = 1;
 351
 352         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 353                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 354
 355         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 356                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 357                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 358                                       "unknown maximum xattr size.\n", dt);
 359                 } else {
 360                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 361                         sbi->ll_xattr_cache_enabled = 1;
 362                 }
 363         }
 364
 365         obd = class_name2obd(dt);
 366         if (!obd) {
 367                 CERROR("DT %s: not setup or attached\n", dt);
 368                 GOTO(out_md_fid, err = -ENODEV);
 369         }
 370
 371         /* pass client page size via ocd_grant_blkbits, the server should report
 372          * back its backend blocksize for grant calculation purpose */
 373         data->ocd_grant_blkbits = PAGE_SHIFT;
 374
 375         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 376                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 377                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 378                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 379                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 380                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 381                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 382                                   OBD_CONNECT_EINPROGRESS |
 383                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 384                                   OBD_CONNECT_LAYOUTLOCK |
 385                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 386                                   OBD_CONNECT_BULK_MBITS;
 387
 388         data->ocd_connect_flags2 = 0;
 389
 390         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 391                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 392
 393         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 394                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 395                  * disabled by default, because it can still be enabled on the
 396                  * fly via /proc. As a consequence, we still need to come to an
 397                  * agreement on the supported algorithms at connect time */
 398                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 399
 400                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 401                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 402                 else
 403                         data->ocd_cksum_types = cksum_types_supported_client();
 404         }
 405
 406 #ifdef HAVE_LRU_RESIZE_SUPPORT
 407         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 408 #endif
 409         /* always ping even if server suppress_pings */
 410         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 411                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 412
 413         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 414                "ocd_grant: %d\n", data->ocd_connect_flags,
 415                data->ocd_version, data->ocd_grant);
 416
 417         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 418         obd->obd_upcall.onu_upcall = cl_ocd_update;
 419
 420         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 421
 422         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 423                           NULL);
 424         if (err == -EBUSY) {
 425                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 426                                    "recovery, of which this client is not a "
 427                                    "part.  Please wait for recovery to "
 428                                    "complete, abort, or time out.\n", dt);
 429                 GOTO(out_md, err);
 430         } else if (err) {
 431                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 432                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 433                 GOTO(out_md, err);
 434         }
 435
 436         sbi->ll_dt_exp->exp_connect_data = *data;
 437
 438         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 439                            LUSTRE_SEQ_METADATA);
 440         if (err) {
 441                 CERROR("%s: Can't init data layer FID infrastructure, "
 442                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 443                 GOTO(out_dt, err);
 444         }
 445
 446         mutex_lock(&sbi->ll_lco.lco_lock);
 447         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 448         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 449         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 450         mutex_unlock(&sbi->ll_lco.lco_lock);
 451
 452         fid_zero(&sbi->ll_root_fid);
 453         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 454                            &sbi->ll_root_fid);
 455         if (err) {
 456                 CERROR("cannot mds_connect: rc = %d\n", err);
 457                 GOTO(out_lock_cn_cb, err);
 458         }
 459         if (!fid_is_sane(&sbi->ll_root_fid)) {
 460                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 461                        sbi->ll_md_exp->exp_obd->obd_name,
 462                        PFID(&sbi->ll_root_fid));
 463                 GOTO(out_lock_cn_cb, err = -EINVAL);
 464         }
 465         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 466
 467         sb->s_op = &lustre_super_operations;
 468 #if THREAD_SIZE >= 8192 /*b=17630*/
 469         sb->s_export_op = &lustre_export_operations;
 470 #endif
 471
 472         /* make root inode
 473          * XXX: move this to after cbd setup? */
 474         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 475         if (sbi->ll_flags & LL_SBI_ACL)
 476                 valid |= OBD_MD_FLACL;
 477
 478         OBD_ALLOC_PTR(op_data);
 479         if (op_data == NULL)
 480                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 481
 482         op_data->op_fid1 = sbi->ll_root_fid;
 483         op_data->op_mode = 0;
 484         op_data->op_valid = valid;
 485
 486         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 487
 488         OBD_FREE_PTR(op_data);
 489         if (err) {
 490                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 491                        sbi->ll_md_exp->exp_obd->obd_name, err);
 492                 GOTO(out_lock_cn_cb, err);
 493         }
 494
 495         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 496                                sbi->ll_md_exp, &lmd);
 497         if (err) {
 498                 CERROR("failed to understand root inode md: rc = %d\n", err);
 499                 ptlrpc_req_finished(request);
 500                 GOTO(out_lock_cn_cb, err);
 501         }
 502
 503         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 504         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 505                                             sbi->ll_flags & LL_SBI_32BIT_API),
 506                        &lmd);
 507         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 508         ptlrpc_req_finished(request);
 509
 510         if (IS_ERR(root)) {
 511 #ifdef CONFIG_FS_POSIX_ACL
 512                 if (lmd.posix_acl) {
 513                         posix_acl_release(lmd.posix_acl);
 514                         lmd.posix_acl = NULL;
 515                 }
 516 #endif
 517                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 518                 root = NULL;
 519                 CERROR("lustre_lite: bad iget4 for root\n");
 520                 GOTO(out_root, err);
 521         }
 522
 523         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 524         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 525                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 526                                  NULL);
 527         if (err) {
 528                 CERROR("%s: Set checksum failed: rc = %d\n",
 529                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 530                 GOTO(out_root, err);
 531         }
 532         cl_sb_init(sb);
 533
 534         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 535                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 536                                  sbi->ll_cache, NULL);
 537         if (err) {
 538                 CERROR("%s: Set cache_set failed: rc = %d\n",
 539                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 540                 GOTO(out_root, err);
 541         }
 542
 543         sb->s_root = d_make_root(root);
 544         if (sb->s_root == NULL) {
 545                 CERROR("%s: can't make root dentry\n",
 546                         ll_get_fsname(sb, NULL, 0));
 547                 GOTO(out_root, err = -ENOMEM);
 548         }
 549 #ifdef HAVE_DCACHE_LOCK
 550         sb->s_root->d_op = &ll_d_ops;
 551 #endif
 552
 553         sbi->ll_sdev_orig = sb->s_dev;
 554
 555         /* We set sb->s_dev equal on all lustre clients in order to support
 556          * NFS export clustering.  NFSD requires that the FSID be the same
 557          * on all clients. */
 558         /* s_dev is also used in lt_compare() to compare two fs, but that is
 559          * only a node-local comparison. */
 560         uuid = obd_get_uuid(sbi->ll_md_exp);
 561         if (uuid != NULL)
 562                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 563
 564         if (data != NULL)
 565                 OBD_FREE_PTR(data);
 566         if (osfs != NULL)
 567                 OBD_FREE_PTR(osfs);
 568         if (proc_lustre_fs_root != NULL) {
 569                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 570                                                   dt, md);
 571                 if (err < 0) {
 572                         CERROR("%s: could not register mount in lprocfs: "
 573                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 574                         err = 0;
 575                 }
 576         }
 577
 578         RETURN(err);
 579 out_root:
 580         if (root)
 581                 iput(root);
 582 out_lock_cn_cb:
 583         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 584 out_dt:
 585         obd_disconnect(sbi->ll_dt_exp);
 586         sbi->ll_dt_exp = NULL;
 587 out_md_fid:
 588         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 589 out_md:
 590         obd_disconnect(sbi->ll_md_exp);
 591         sbi->ll_md_exp = NULL;
 592 out:
 593         if (data != NULL)
 594                 OBD_FREE_PTR(data);
 595         if (osfs != NULL)
 596                 OBD_FREE_PTR(osfs);
 597         return err;
 598 }
 599
 600 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 601 {
 602         int size, rc;
 603
 604         size = sizeof(*lmmsize);
 605         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 606                           KEY_MAX_EASIZE, &size, lmmsize);
 607         if (rc != 0) {
 608                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 609                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 610                 RETURN(rc);
 611         }
 612
 613         size = sizeof(int);
 614         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 615                           KEY_MAX_EASIZE, &size, lmmsize);
 616         if (rc)
 617                 CERROR("Get max mdsize error rc %d\n", rc);
 618
 619         RETURN(rc);
 620 }
 621
 622 /**
 623  * Get the value of the default_easize parameter.
 624  *
 625  * \see client_obd::cl_default_mds_easize
 626  *
 627  * \param[in] sbi       superblock info for this filesystem
 628  * \param[out] lmmsize  pointer to storage location for value
 629  *
 630  * \retval 0            on success
 631  * \retval negative     negated errno on failure
 632  */
 633 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 634 {
 635         int size, rc;
 636
 637         size = sizeof(int);
 638         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 639                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 640         if (rc)
 641                 CERROR("Get default mdsize error rc %d\n", rc);
 642
 643         RETURN(rc);
 644 }
 645
 646 /**
 647  * Set the default_easize parameter to the given value.
 648  *
 649  * \see client_obd::cl_default_mds_easize
 650  *
 651  * \param[in] sbi       superblock info for this filesystem
 652  * \param[in] lmmsize   the size to set
 653  *
 654  * \retval 0            on success
 655  * \retval negative     negated errno on failure
 656  */
 657 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 658 {
 659         int rc;
 660
 661         if (lmmsize < sizeof(struct lov_mds_md) ||
 662             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 663                 return -EINVAL;
 664
 665         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 666                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 667                                 sizeof(int), &lmmsize, NULL);
 668
 669         RETURN(rc);
 670 }
 671
 672 static void ll_dump_inode(struct inode *inode)
 673 {
 674         struct ll_d_hlist_node *tmp;
 675         int dentry_count = 0;
 676
 677         LASSERT(inode != NULL);
 678
 679         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 680                 dentry_count++;
 681
 682         CERROR("%s: inode %p dump: dev=%s fid="DFID
 683                " mode=%o count=%u, %d dentries\n",
 684                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 685                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 686                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 687 }
 688
 689 void lustre_dump_dentry(struct dentry *dentry, int recur)
 690 {
 691         struct list_head *tmp;
 692         int subdirs = 0;
 693
 694         LASSERT(dentry != NULL);
 695
 696         list_for_each(tmp, &dentry->d_subdirs)
 697                 subdirs++;
 698
 699         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 700                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 701                dentry->d_name.len, dentry->d_name.name,
 702                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 703                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 704                dentry->d_flags, dentry->d_fsdata, subdirs);
 705         if (dentry->d_inode != NULL)
 706                 ll_dump_inode(dentry->d_inode);
 707
 708         if (recur == 0)
 709                 return;
 710
 711         list_for_each(tmp, &dentry->d_subdirs) {
 712                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 713                 lustre_dump_dentry(d, recur - 1);
 714         }
 715 }
 716
 717 static void client_common_put_super(struct super_block *sb)
 718 {
 719         struct ll_sb_info *sbi = ll_s2sbi(sb);
 720         ENTRY;
 721
 722         cl_sb_fini(sb);
 723
 724         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 725         obd_disconnect(sbi->ll_dt_exp);
 726         sbi->ll_dt_exp = NULL;
 727
 728         lprocfs_unregister_mountpoint(sbi);
 729
 730         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 731         obd_disconnect(sbi->ll_md_exp);
 732         sbi->ll_md_exp = NULL;
 733
 734         EXIT;
 735 }
 736
 737 void ll_kill_super(struct super_block *sb)
 738 {
 739         struct ll_sb_info *sbi;
 740         ENTRY;
 741
 742         /* not init sb ?*/
 743         if (!(sb->s_flags & MS_ACTIVE))
 744                 return;
 745
 746         sbi = ll_s2sbi(sb);
 747         /* we need restore s_dev from changed for clustred NFS before put_super
 748          * because new kernels have cached s_dev and change sb->s_dev in
 749          * put_super not affected real removing devices */
 750         if (sbi) {
 751                 sb->s_dev = sbi->ll_sdev_orig;
 752                 sbi->ll_umounting = 1;
 753
 754                 /* wait running statahead threads to quit */
 755                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 756                         set_current_state(TASK_UNINTERRUPTIBLE);
 757                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 758                 }
 759         }
 760
 761         EXIT;
 762 }
 763
 764 static inline int ll_set_opt(const char *opt, char *data, int fl)
 765 {
 766         if (strncmp(opt, data, strlen(opt)) != 0)
 767                 return(0);
 768         else
 769                 return(fl);
 770 }
 771
 772 /* non-client-specific mount options are parsed in lmd_parse */
 773 static int ll_options(char *options, int *flags)
 774 {
 775         int tmp;
 776         char *s1 = options, *s2;
 777         ENTRY;
 778
 779         if (!options)
 780                 RETURN(0);
 781
 782         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 783
 784         while (*s1) {
 785                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 786                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 787                 if (tmp) {
 788                         *flags |= tmp;
 789                         goto next;
 790                 }
 791                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 792                 if (tmp) {
 793                         *flags |= tmp;
 794                         goto next;
 795                 }
 796                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 797                 if (tmp) {
 798                         *flags |= tmp;
 799                         goto next;
 800                 }
 801                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 802                 if (tmp) {
 803                         *flags &= ~tmp;
 804                         goto next;
 805                 }
 806                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 807                 if (tmp) {
 808                         *flags |= tmp;
 809                         goto next;
 810                 }
 811                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 812                 if (tmp) {
 813                         *flags &= ~tmp;
 814                         goto next;
 815                 }
 816                 tmp = ll_set_opt("context", s1, 1);
 817                 if (tmp)
 818                         goto next;
 819                 tmp = ll_set_opt("fscontext", s1, 1);
 820                 if (tmp)
 821                         goto next;
 822                 tmp = ll_set_opt("defcontext", s1, 1);
 823                 if (tmp)
 824                         goto next;
 825                 tmp = ll_set_opt("rootcontext", s1, 1);
 826                 if (tmp)
 827                         goto next;
 828                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838
 839                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 840                 if (tmp) {
 841                         *flags |= tmp;
 842                         goto next;
 843                 }
 844                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 845                 if (tmp) {
 846                         *flags &= ~tmp;
 847                         goto next;
 848                 }
 849                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 850                 if (tmp) {
 851                         *flags |= tmp;
 852                         goto next;
 853                 }
 854                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 855                 if (tmp) {
 856                         *flags &= ~tmp;
 857                         goto next;
 858                 }
 859                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 860                 if (tmp) {
 861                         *flags |= tmp;
 862                         goto next;
 863                 }
 864                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 865                 if (tmp) {
 866                         *flags &= ~tmp;
 867                         goto next;
 868                 }
 869                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 870                 if (tmp) {
 871                         *flags |= tmp;
 872                         goto next;
 873                 }
 874                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 875                 if (tmp) {
 876                         *flags |= tmp;
 877                         goto next;
 878                 }
 879                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 880                 if (tmp) {
 881                         *flags &= ~tmp;
 882                         goto next;
 883                 }
 884                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 885                 if (tmp) {
 886                         *flags |= tmp;
 887                         goto next;
 888                 }
 889                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 890                                    s1);
 891                 RETURN(-EINVAL);
 892
 893 next:
 894                 /* Find next opt */
 895                 s2 = strchr(s1, ',');
 896                 if (s2 == NULL)
 897                         break;
 898                 s1 = s2 + 1;
 899         }
 900         RETURN(0);
 901 }
 902
 903 void ll_lli_init(struct ll_inode_info *lli)
 904 {
 905         lli->lli_inode_magic = LLI_INODE_MAGIC;
 906         lli->lli_flags = 0;
 907         spin_lock_init(&lli->lli_lock);
 908         lli->lli_posix_acl = NULL;
 909         /* Do not set lli_fid, it has been initialized already. */
 910         fid_zero(&lli->lli_pfid);
 911         lli->lli_mds_read_och = NULL;
 912         lli->lli_mds_write_och = NULL;
 913         lli->lli_mds_exec_och = NULL;
 914         lli->lli_open_fd_read_count = 0;
 915         lli->lli_open_fd_write_count = 0;
 916         lli->lli_open_fd_exec_count = 0;
 917         mutex_init(&lli->lli_och_mutex);
 918         spin_lock_init(&lli->lli_agl_lock);
 919         spin_lock_init(&lli->lli_layout_lock);
 920         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 921         lli->lli_clob = NULL;
 922
 923         init_rwsem(&lli->lli_xattrs_list_rwsem);
 924         mutex_init(&lli->lli_xattrs_enq_lock);
 925
 926         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 927         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 928                 mutex_init(&lli->lli_readdir_mutex);
 929                 lli->lli_opendir_key = NULL;
 930                 lli->lli_sai = NULL;
 931                 spin_lock_init(&lli->lli_sa_lock);
 932                 lli->lli_opendir_pid = 0;
 933                 lli->lli_sa_enabled = 0;
 934                 lli->lli_def_stripe_offset = -1;
 935         } else {
 936                 mutex_init(&lli->lli_size_mutex);
 937                 lli->lli_symlink_name = NULL;
 938                 init_rwsem(&lli->lli_trunc_sem);
 939                 range_lock_tree_init(&lli->lli_write_tree);
 940                 init_rwsem(&lli->lli_glimpse_sem);
 941                 lli->lli_glimpse_time = 0;
 942                 INIT_LIST_HEAD(&lli->lli_agl_list);
 943                 lli->lli_agl_index = 0;
 944                 lli->lli_async_rc = 0;
 945         }
 946         mutex_init(&lli->lli_layout_mutex);
 947 }
 948
 949 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 950 {
 951         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 952
 953         bdi->name = "lustre";
 954         return bdi_register(bdi, NULL, "lustre-%d",
 955                             atomic_inc_return(&ll_bdi_num));
 956 }
 957
 958 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 959 {
 960         struct lustre_profile *lprof = NULL;
 961         struct lustre_sb_info *lsi = s2lsi(sb);
 962         struct ll_sb_info *sbi;
 963         char  *dt = NULL, *md = NULL;
 964         char  *profilenm = get_profile_name(sb);
 965         struct config_llog_instance *cfg;
 966         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 967         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 968         int    err;
 969         ENTRY;
 970
 971         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 972
 973         OBD_ALLOC_PTR(cfg);
 974         if (cfg == NULL)
 975                 RETURN(-ENOMEM);
 976
 977         try_module_get(THIS_MODULE);
 978
 979         /* client additional sb info */
 980         lsi->lsi_llsbi = sbi = ll_init_sbi();
 981         if (!sbi) {
 982                 module_put(THIS_MODULE);
 983                 OBD_FREE_PTR(cfg);
 984                 RETURN(-ENOMEM);
 985         }
 986
 987         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 988         if (err)
 989                 GOTO(out_free, err);
 990
 991         err = bdi_init(&lsi->lsi_bdi);
 992         if (err)
 993                 GOTO(out_free, err);
 994         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 995 #ifdef HAVE_BDI_CAP_MAP_COPY
 996         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 997 #else
 998         lsi->lsi_bdi.capabilities = 0;
 999 #endif
1000         err = ll_bdi_register(&lsi->lsi_bdi);
1001         if (err)
1002                 GOTO(out_free, err);
1003
1004         sb->s_bdi = &lsi->lsi_bdi;
1005 #ifndef HAVE_DCACHE_LOCK
1006         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1007         sb->s_d_op = &ll_d_ops;
1008 #endif
1009
1010         /* Generate a string unique to this super, in case some joker tries
1011            to mount the same fs at two mount points.
1012            Use the address of the super itself.*/
1013         cfg->cfg_instance = sb;
1014         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1015         cfg->cfg_callback = class_config_llog_handler;
1016         /* set up client obds */
1017         err = lustre_process_log(sb, profilenm, cfg);
1018         if (err < 0)
1019                 GOTO(out_free, err);
1020
1021         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1022         lprof = class_get_profile(profilenm);
1023         if (lprof == NULL) {
1024                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1025                                    " read from the MGS.  Does that filesystem "
1026                                    "exist?\n", profilenm);
1027                 GOTO(out_free, err = -EINVAL);
1028         }
1029         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1030                lprof->lp_md, lprof->lp_dt);
1031
1032         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1033         if (!dt)
1034                 GOTO(out_free, err = -ENOMEM);
1035         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1036
1037         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1038         if (!md)
1039                 GOTO(out_free, err = -ENOMEM);
1040         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1041
1042         /* connections, registrations, sb setup */
1043         err = client_common_fill_super(sb, md, dt, mnt);
1044         if (err < 0)
1045                 GOTO(out_free, err);
1046
1047         sbi->ll_client_common_fill_super_succeeded = 1;
1048
1049 out_free:
1050         if (md)
1051                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1052         if (dt)
1053                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1054         if (lprof != NULL)
1055                 class_put_profile(lprof);
1056         if (err)
1057                 ll_put_super(sb);
1058         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1059                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1060
1061         OBD_FREE_PTR(cfg);
1062         RETURN(err);
1063 } /* ll_fill_super */
1064
1065 void ll_put_super(struct super_block *sb)
1066 {
1067         struct config_llog_instance cfg, params_cfg;
1068         struct obd_device *obd;
1069         struct lustre_sb_info *lsi = s2lsi(sb);
1070         struct ll_sb_info *sbi = ll_s2sbi(sb);
1071         char *profilenm = get_profile_name(sb);
1072         long ccc_count;
1073         int next, force = 1, rc = 0;
1074         ENTRY;
1075
1076         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1077
1078         cfg.cfg_instance = sb;
1079         lustre_end_log(sb, profilenm, &cfg);
1080
1081         params_cfg.cfg_instance = sb;
1082         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1083
1084         if (sbi->ll_md_exp) {
1085                 obd = class_exp2obd(sbi->ll_md_exp);
1086                 if (obd)
1087                         force = obd->obd_force;
1088         }
1089
1090         /* Wait for unstable pages to be committed to stable storage */
1091         if (force == 0) {
1092                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1093                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1094                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1095                         &lwi);
1096         }
1097
1098         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1099         if (force == 0 && rc != -EINTR)
1100                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1101
1102
1103         /* We need to set force before the lov_disconnect in
1104            lustre_common_put_super, since l_d cleans up osc's as well. */
1105         if (force) {
1106                 next = 0;
1107                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1108                                                      &next)) != NULL) {
1109                         obd->obd_force = force;
1110                 }
1111         }
1112
1113         if (sbi->ll_client_common_fill_super_succeeded) {
1114                 /* Only if client_common_fill_super succeeded */
1115                 client_common_put_super(sb);
1116         }
1117
1118         next = 0;
1119         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1120                 class_manual_cleanup(obd);
1121         }
1122
1123         if (sbi->ll_flags & LL_SBI_VERBOSE)
1124                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1125
1126         if (profilenm)
1127                 class_del_profile(profilenm);
1128
1129         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1130                 bdi_destroy(&lsi->lsi_bdi);
1131                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1132         }
1133
1134         ll_free_sbi(sb);
1135         lsi->lsi_llsbi = NULL;
1136
1137         lustre_common_put_super(sb);
1138
1139         cl_env_cache_purge(~0);
1140
1141         module_put(THIS_MODULE);
1142
1143         EXIT;
1144 } /* client_put_super */
1145
1146 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1147 {
1148         struct inode *inode = NULL;
1149
1150         /* NOTE: we depend on atomic igrab() -bzzz */
1151         lock_res_and_lock(lock);
1152         if (lock->l_resource->lr_lvb_inode) {
1153                 struct ll_inode_info * lli;
1154                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1155                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1156                         inode = igrab(lock->l_resource->lr_lvb_inode);
1157                 } else {
1158                         inode = lock->l_resource->lr_lvb_inode;
1159                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1160                                          D_WARNING, lock, "lr_lvb_inode %p is "
1161                                          "bogus: magic %08x",
1162                                          lock->l_resource->lr_lvb_inode,
1163                                          lli->lli_inode_magic);
1164                         inode = NULL;
1165                 }
1166         }
1167         unlock_res_and_lock(lock);
1168         return inode;
1169 }
1170
1171 void ll_dir_clear_lsm_md(struct inode *inode)
1172 {
1173         struct ll_inode_info *lli = ll_i2info(inode);
1174
1175         LASSERT(S_ISDIR(inode->i_mode));
1176
1177         if (lli->lli_lsm_md != NULL) {
1178                 lmv_free_memmd(lli->lli_lsm_md);
1179                 lli->lli_lsm_md = NULL;
1180         }
1181 }
1182
1183 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1184                                       const struct lu_fid *fid,
1185                                       struct lustre_md *md)
1186 {
1187         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1188         struct mdt_body         *body = md->body;
1189         struct inode            *inode;
1190         ino_t                   ino;
1191         ENTRY;
1192
1193         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1194         inode = iget_locked(sb, ino);
1195         if (inode == NULL) {
1196                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1197                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1198                 RETURN(ERR_PTR(-ENOENT));
1199         }
1200
1201         if (inode->i_state & I_NEW) {
1202                 struct ll_inode_info *lli = ll_i2info(inode);
1203                 struct lmv_stripe_md *lsm = md->lmv;
1204
1205                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1206                                 (body->mbo_mode & S_IFMT);
1207                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1208                          PFID(fid));
1209
1210                 LTIME_S(inode->i_mtime) = 0;
1211                 LTIME_S(inode->i_atime) = 0;
1212                 LTIME_S(inode->i_ctime) = 0;
1213                 inode->i_rdev = 0;
1214
1215 #ifdef HAVE_BACKING_DEV_INFO
1216                 /* initializing backing dev info. */
1217                 inode->i_mapping->backing_dev_info =
1218                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1219 #endif
1220                 inode->i_op = &ll_dir_inode_operations;
1221                 inode->i_fop = &ll_dir_operations;
1222                 lli->lli_fid = *fid;
1223                 ll_lli_init(lli);
1224
1225                 LASSERT(lsm != NULL);
1226                 /* master object FID */
1227                 lli->lli_pfid = body->mbo_fid1;
1228                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1229                        lli, PFID(fid), PFID(&lli->lli_pfid));
1230                 unlock_new_inode(inode);
1231         }
1232
1233         RETURN(inode);
1234 }
1235
1236 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1237 {
1238         struct lu_fid *fid;
1239         struct lmv_stripe_md *lsm = md->lmv;
1240         int i;
1241
1242         LASSERT(lsm != NULL);
1243         /* XXX sigh, this lsm_root initialization should be in
1244          * LMV layer, but it needs ll_iget right now, so we
1245          * put this here right now. */
1246         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1247                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1248                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1249                 /* Unfortunately ll_iget will call ll_update_inode,
1250                  * where the initialization of slave inode is slightly
1251                  * different, so it reset lsm_md to NULL to avoid
1252                  * initializing lsm for slave inode. */
1253                 /* For migrating inode, master stripe and master object will
1254                  * be same, so we only need assign this inode */
1255                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1256                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1257                 else
1258                         lsm->lsm_md_oinfo[i].lmo_root =
1259                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1260
1261                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1262                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1263
1264                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1265                         return rc;
1266                 }
1267         }
1268
1269         return 0;
1270 }
1271
1272 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1273                                 const struct lmv_stripe_md *lsm_md2)
1274 {
1275         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1276                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1277                lsm_md1->lsm_md_master_mdt_index ==
1278                                         lsm_md2->lsm_md_master_mdt_index &&
1279                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1280                lsm_md1->lsm_md_layout_version ==
1281                                         lsm_md2->lsm_md_layout_version &&
1282                strcmp(lsm_md1->lsm_md_pool_name,
1283                       lsm_md2->lsm_md_pool_name) == 0;
1284 }
1285
1286 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1287 {
1288         struct ll_inode_info *lli = ll_i2info(inode);
1289         struct lmv_stripe_md *lsm = md->lmv;
1290         int     rc;
1291         ENTRY;
1292
1293         LASSERT(S_ISDIR(inode->i_mode));
1294         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1295                PFID(ll_inode2fid(inode)));
1296
1297         /* no striped information from request. */
1298         if (lsm == NULL) {
1299                 if (lli->lli_lsm_md == NULL) {
1300                         RETURN(0);
1301                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1302                                                 LMV_HASH_FLAG_MIGRATION) {
1303                         /* migration is done, the temporay MIGRATE layout has
1304                          * been removed */
1305                         CDEBUG(D_INODE, DFID" finish migration.\n",
1306                                PFID(ll_inode2fid(inode)));
1307                         lmv_free_memmd(lli->lli_lsm_md);
1308                         lli->lli_lsm_md = NULL;
1309                         RETURN(0);
1310                 } else {
1311                         /* The lustre_md from req does not include stripeEA,
1312                          * see ll_md_setattr */
1313                         RETURN(0);
1314                 }
1315         }
1316
1317         /* set the directory layout */
1318         if (lli->lli_lsm_md == NULL) {
1319                 struct cl_attr  *attr;
1320
1321                 rc = ll_init_lsm_md(inode, md);
1322                 if (rc != 0)
1323                         RETURN(rc);
1324
1325                 /* set md->lmv to NULL, so the following free lustre_md
1326                  * will not free this lsm */
1327                 md->lmv = NULL;
1328                 lli->lli_lsm_md = lsm;
1329
1330                 OBD_ALLOC_PTR(attr);
1331                 if (attr == NULL)
1332                         RETURN(-ENOMEM);
1333
1334                 /* validate the lsm */
1335                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1336                                    ll_md_blocking_ast);
1337                 if (rc != 0) {
1338                         OBD_FREE_PTR(attr);
1339                         RETURN(rc);
1340                 }
1341
1342                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1343                         md->body->mbo_nlink = attr->cat_nlink;
1344                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1345                         md->body->mbo_size = attr->cat_size;
1346                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1347                         md->body->mbo_atime = attr->cat_atime;
1348                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1349                         md->body->mbo_ctime = attr->cat_ctime;
1350                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1351                         md->body->mbo_mtime = attr->cat_mtime;
1352
1353                 OBD_FREE_PTR(attr);
1354
1355                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1356                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1357                 RETURN(0);
1358         }
1359
1360         /* Compare the old and new stripe information */
1361         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1362                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1363                 int                     idx;
1364
1365                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1366                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1367                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1368                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1369                        inode, lsm, old_lsm,
1370                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1371                        lsm->lsm_md_stripe_count,
1372                        old_lsm->lsm_md_stripe_count,
1373                        lsm->lsm_md_master_mdt_index,
1374                        old_lsm->lsm_md_master_mdt_index,
1375                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1376                        lsm->lsm_md_layout_version,
1377                        old_lsm->lsm_md_layout_version,
1378                        lsm->lsm_md_pool_name,
1379                        old_lsm->lsm_md_pool_name);
1380
1381                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1382                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1383                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1384                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1385                 }
1386
1387                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1388                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1389                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1390                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1391                 }
1392
1393                 RETURN(-EIO);
1394         }
1395
1396         RETURN(0);
1397 }
1398
1399 void ll_clear_inode(struct inode *inode)
1400 {
1401         struct ll_inode_info *lli = ll_i2info(inode);
1402         struct ll_sb_info *sbi = ll_i2sbi(inode);
1403         ENTRY;
1404
1405         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1406                PFID(ll_inode2fid(inode)), inode);
1407
1408         if (S_ISDIR(inode->i_mode)) {
1409                 /* these should have been cleared in ll_file_release */
1410                 LASSERT(lli->lli_opendir_key == NULL);
1411                 LASSERT(lli->lli_sai == NULL);
1412                 LASSERT(lli->lli_opendir_pid == 0);
1413         }
1414
1415         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1416
1417         LASSERT(!lli->lli_open_fd_write_count);
1418         LASSERT(!lli->lli_open_fd_read_count);
1419         LASSERT(!lli->lli_open_fd_exec_count);
1420
1421         if (lli->lli_mds_write_och)
1422                 ll_md_real_close(inode, FMODE_WRITE);
1423         if (lli->lli_mds_exec_och)
1424                 ll_md_real_close(inode, FMODE_EXEC);
1425         if (lli->lli_mds_read_och)
1426                 ll_md_real_close(inode, FMODE_READ);
1427
1428         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1429                 OBD_FREE(lli->lli_symlink_name,
1430                          strlen(lli->lli_symlink_name) + 1);
1431                 lli->lli_symlink_name = NULL;
1432         }
1433
1434         ll_xattr_cache_destroy(inode);
1435
1436 #ifdef CONFIG_FS_POSIX_ACL
1437         if (lli->lli_posix_acl) {
1438                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1439                 posix_acl_release(lli->lli_posix_acl);
1440                 lli->lli_posix_acl = NULL;
1441         }
1442 #endif
1443         lli->lli_inode_magic = LLI_INODE_DEAD;
1444
1445         if (S_ISDIR(inode->i_mode))
1446                 ll_dir_clear_lsm_md(inode);
1447         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1448                 LASSERT(list_empty(&lli->lli_agl_list));
1449
1450         /*
1451          * XXX This has to be done before lsm is freed below, because
1452          * cl_object still uses inode lsm.
1453          */
1454         cl_inode_fini(inode);
1455
1456         EXIT;
1457 }
1458
1459 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1460 {
1461         struct lustre_md md;
1462         struct inode *inode = dentry->d_inode;
1463         struct ll_sb_info *sbi = ll_i2sbi(inode);
1464         struct ptlrpc_request *request = NULL;
1465         int rc, ia_valid;
1466         ENTRY;
1467
1468         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1469                                      LUSTRE_OPC_ANY, NULL);
1470         if (IS_ERR(op_data))
1471                 RETURN(PTR_ERR(op_data));
1472
1473         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1474         if (rc) {
1475                 ptlrpc_req_finished(request);
1476                 if (rc == -ENOENT) {
1477                         clear_nlink(inode);
1478                         /* Unlinked special device node? Or just a race?
1479                          * Pretend we done everything. */
1480                         if (!S_ISREG(inode->i_mode) &&
1481                             !S_ISDIR(inode->i_mode)) {
1482                                 ia_valid = op_data->op_attr.ia_valid;
1483                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1484                                 rc = simple_setattr(dentry, &op_data->op_attr);
1485                                 op_data->op_attr.ia_valid = ia_valid;
1486                         }
1487                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1488                         CERROR("md_setattr fails: rc = %d\n", rc);
1489                 }
1490                 RETURN(rc);
1491         }
1492
1493         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1494                               sbi->ll_md_exp, &md);
1495         if (rc) {
1496                 ptlrpc_req_finished(request);
1497                 RETURN(rc);
1498         }
1499
1500         ia_valid = op_data->op_attr.ia_valid;
1501         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1502          * cache is not cleared yet. */
1503         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1504         if (S_ISREG(inode->i_mode))
1505                 inode_lock(inode);
1506         rc = simple_setattr(dentry, &op_data->op_attr);
1507         if (S_ISREG(inode->i_mode))
1508                 inode_unlock(inode);
1509         op_data->op_attr.ia_valid = ia_valid;
1510
1511         rc = ll_update_inode(inode, &md);
1512         ptlrpc_req_finished(request);
1513
1514         RETURN(rc);
1515 }
1516
1517 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1518  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1519  * keep these values until such a time that objects are allocated for it.
1520  * We do the MDS operations first, as it is checking permissions for us.
1521  * We don't to the MDS RPC if there is nothing that we want to store there,
1522  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1523  * going to do an RPC anyways.
1524  *
1525  * If we are doing a truncate, we will send the mtime and ctime updates
1526  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1527  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1528  * at the same time.
1529  *
1530  * In case of HSMimport, we only set attr on MDS.
1531  */
1532 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1533 {
1534         struct inode *inode = dentry->d_inode;
1535         struct ll_inode_info *lli = ll_i2info(inode);
1536         struct md_op_data *op_data = NULL;
1537         int rc = 0;
1538         ENTRY;
1539
1540         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1541                "valid %x, hsm_import %d\n",
1542                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1543                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1544                hsm_import);
1545
1546         if (attr->ia_valid & ATTR_SIZE) {
1547                 /* Check new size against VFS/VM file size limit and rlimit */
1548                 rc = inode_newsize_ok(inode, attr->ia_size);
1549                 if (rc)
1550                         RETURN(rc);
1551
1552                 /* The maximum Lustre file size is variable, based on the
1553                  * OST maximum object size and number of stripes.  This
1554                  * needs another check in addition to the VFS check above. */
1555                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1556                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1557                                PFID(&lli->lli_fid), attr->ia_size,
1558                                ll_file_maxbytes(inode));
1559                         RETURN(-EFBIG);
1560                 }
1561
1562                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1563         }
1564
1565         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1566         if (attr->ia_valid & TIMES_SET_FLAGS) {
1567                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1568                     !cfs_capable(CFS_CAP_FOWNER))
1569                         RETURN(-EPERM);
1570         }
1571
1572         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1573         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1574             (attr->ia_valid & ATTR_CTIME)) {
1575                 attr->ia_ctime = CURRENT_TIME;
1576                 attr->ia_valid |= ATTR_CTIME_SET;
1577         }
1578         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1579             (attr->ia_valid & ATTR_ATIME)) {
1580                 attr->ia_atime = CURRENT_TIME;
1581                 attr->ia_valid |= ATTR_ATIME_SET;
1582         }
1583         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1584             (attr->ia_valid & ATTR_MTIME)) {
1585                 attr->ia_mtime = CURRENT_TIME;
1586                 attr->ia_valid |= ATTR_MTIME_SET;
1587         }
1588
1589         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1590                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1591                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1592                        cfs_time_current_sec());
1593
1594         if (S_ISREG(inode->i_mode)) {
1595                 if (attr->ia_valid & ATTR_SIZE)
1596                         inode_dio_write_done(inode);
1597                 inode_unlock(inode);
1598         }
1599
1600         /* We always do an MDS RPC, even if we're only changing the size;
1601          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1602
1603         OBD_ALLOC_PTR(op_data);
1604         if (op_data == NULL)
1605                 GOTO(out, rc = -ENOMEM);
1606
1607         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1608                 /* If we are changing file size, file content is
1609                  * modified, flag it. */
1610                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1611                 op_data->op_bias |= MDS_DATA_MODIFIED;
1612                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1613         }
1614
1615         op_data->op_attr = *attr;
1616
1617         rc = ll_md_setattr(dentry, op_data);
1618         if (rc)
1619                 GOTO(out, rc);
1620
1621         if (!S_ISREG(inode->i_mode) || hsm_import)
1622                 GOTO(out, rc = 0);
1623
1624         if (attr->ia_valid & (ATTR_SIZE |
1625                               ATTR_ATIME | ATTR_ATIME_SET |
1626                               ATTR_MTIME | ATTR_MTIME_SET |
1627                               ATTR_CTIME | ATTR_CTIME_SET)) {
1628                 /* For truncate and utimes sending attributes to OSTs, setting
1629                  * mtime/atime to the past will be performed under PW [0:EOF]
1630                  * extent lock (new_size:EOF for truncate).  It may seem
1631                  * excessive to send mtime/atime updates to OSTs when not
1632                  * setting times to past, but it is necessary due to possible
1633                  * time de-synchronization between MDT inode and OST objects */
1634                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1635         }
1636
1637         /* If the file was restored, it needs to set dirty flag.
1638          *
1639          * We've already sent MDS_DATA_MODIFIED flag in
1640          * ll_md_setattr() for truncate. However, the MDT refuses to
1641          * set the HS_DIRTY flag on released files, so we have to set
1642          * it again if the file has been restored. Please check how
1643          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1644          *
1645          * Please notice that if the file is not released, the previous
1646          * MDS_DATA_MODIFIED has taken effect and usually
1647          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1648          * This way we can save an RPC for common open + trunc
1649          * operation. */
1650         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1651                 struct hsm_state_set hss = {
1652                         .hss_valid = HSS_SETMASK,
1653                         .hss_setmask = HS_DIRTY,
1654                 };
1655                 int rc2;
1656
1657                 rc2 = ll_hsm_state_set(inode, &hss);
1658                 /* truncate and write can happen at the same time, so that
1659                  * the file can be set modified even though the file is not
1660                  * restored from released state, and ll_hsm_state_set() is
1661                  * not applicable for the file, and rc2 < 0 is normal in this
1662                  * case. */
1663                 if (rc2 < 0)
1664                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1665                                PFID(ll_inode2fid(inode)), rc2);
1666         }
1667
1668         EXIT;
1669 out:
1670         if (op_data != NULL)
1671                 ll_finish_md_op_data(op_data);
1672
1673         if (S_ISREG(inode->i_mode)) {
1674                 inode_lock(inode);
1675                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1676                         inode_dio_wait(inode);
1677         }
1678
1679         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1680                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1681
1682         return rc;
1683 }
1684
1685 int ll_setattr(struct dentry *de, struct iattr *attr)
1686 {
1687         int mode = de->d_inode->i_mode;
1688
1689         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1690                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1691                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1692
1693         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1694                                (ATTR_SIZE|ATTR_MODE)) &&
1695             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1696              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1697               !(attr->ia_mode & S_ISGID))))
1698                 attr->ia_valid |= ATTR_FORCE;
1699
1700         if ((attr->ia_valid & ATTR_MODE) &&
1701             (mode & S_ISUID) &&
1702             !(attr->ia_mode & S_ISUID) &&
1703             !(attr->ia_valid & ATTR_KILL_SUID))
1704                 attr->ia_valid |= ATTR_KILL_SUID;
1705
1706         if ((attr->ia_valid & ATTR_MODE) &&
1707             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1708             !(attr->ia_mode & S_ISGID) &&
1709             !(attr->ia_valid & ATTR_KILL_SGID))
1710                 attr->ia_valid |= ATTR_KILL_SGID;
1711
1712         return ll_setattr_raw(de, attr, false);
1713 }
1714
1715 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1716                        __u64 max_age, __u32 flags)
1717 {
1718         struct ll_sb_info *sbi = ll_s2sbi(sb);
1719         struct obd_statfs obd_osfs;
1720         int rc;
1721         ENTRY;
1722
1723         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1724         if (rc) {
1725                 CERROR("md_statfs fails: rc = %d\n", rc);
1726                 RETURN(rc);
1727         }
1728
1729         osfs->os_type = sb->s_magic;
1730
1731         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1732                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1733
1734         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1735                 flags |= OBD_STATFS_NODELAY;
1736
1737         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1738         if (rc) {
1739                 CERROR("obd_statfs fails: rc = %d\n", rc);
1740                 RETURN(rc);
1741         }
1742
1743         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1744                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1745                obd_osfs.os_files);
1746
1747         osfs->os_bsize = obd_osfs.os_bsize;
1748         osfs->os_blocks = obd_osfs.os_blocks;
1749         osfs->os_bfree = obd_osfs.os_bfree;
1750         osfs->os_bavail = obd_osfs.os_bavail;
1751
1752         /* If we don't have as many objects free on the OST as inodes
1753          * on the MDS, we reduce the total number of inodes to
1754          * compensate, so that the "inodes in use" number is correct.
1755          */
1756         if (obd_osfs.os_ffree < osfs->os_ffree) {
1757                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1758                         obd_osfs.os_ffree;
1759                 osfs->os_ffree = obd_osfs.os_ffree;
1760         }
1761
1762         RETURN(rc);
1763 }
1764 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1765 {
1766         struct super_block *sb = de->d_sb;
1767         struct obd_statfs osfs;
1768         __u64 fsid = huge_encode_dev(sb->s_dev);
1769         int rc;
1770
1771         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1772         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1773
1774         /* Some amount of caching on the client is allowed */
1775         rc = ll_statfs_internal(sb, &osfs,
1776                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1777                                 0);
1778         if (rc)
1779                 return rc;
1780
1781         statfs_unpack(sfs, &osfs);
1782
1783         /* We need to downshift for all 32-bit kernels, because we can't
1784          * tell if the kernel is being called via sys_statfs64() or not.
1785          * Stop before overflowing f_bsize - in which case it is better
1786          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1787         if (sizeof(long) < 8) {
1788                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1789                         sfs->f_bsize <<= 1;
1790
1791                         osfs.os_blocks >>= 1;
1792                         osfs.os_bfree >>= 1;
1793                         osfs.os_bavail >>= 1;
1794                 }
1795         }
1796
1797         sfs->f_blocks = osfs.os_blocks;
1798         sfs->f_bfree = osfs.os_bfree;
1799         sfs->f_bavail = osfs.os_bavail;
1800         sfs->f_fsid.val[0] = (__u32)fsid;
1801         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1802         return 0;
1803 }
1804
1805 void ll_inode_size_lock(struct inode *inode)
1806 {
1807         struct ll_inode_info *lli;
1808
1809         LASSERT(!S_ISDIR(inode->i_mode));
1810
1811         lli = ll_i2info(inode);
1812         mutex_lock(&lli->lli_size_mutex);
1813 }
1814
1815 void ll_inode_size_unlock(struct inode *inode)
1816 {
1817         struct ll_inode_info *lli;
1818
1819         lli = ll_i2info(inode);
1820         mutex_unlock(&lli->lli_size_mutex);
1821 }
1822
1823 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1824 {
1825         struct ll_inode_info *lli = ll_i2info(inode);
1826         struct mdt_body *body = md->body;
1827         struct ll_sb_info *sbi = ll_i2sbi(inode);
1828
1829         if (body->mbo_valid & OBD_MD_FLEASIZE)
1830                 cl_file_inode_init(inode, md);
1831
1832         if (S_ISDIR(inode->i_mode)) {
1833                 int     rc;
1834
1835                 rc = ll_update_lsm_md(inode, md);
1836                 if (rc != 0)
1837                         return rc;
1838         }
1839
1840 #ifdef CONFIG_FS_POSIX_ACL
1841         if (body->mbo_valid & OBD_MD_FLACL) {
1842                 spin_lock(&lli->lli_lock);
1843                 if (lli->lli_posix_acl)
1844                         posix_acl_release(lli->lli_posix_acl);
1845                 lli->lli_posix_acl = md->posix_acl;
1846                 spin_unlock(&lli->lli_lock);
1847         }
1848 #endif
1849         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1850                                         sbi->ll_flags & LL_SBI_32BIT_API);
1851         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1852
1853         if (body->mbo_valid & OBD_MD_FLATIME) {
1854                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1855                         LTIME_S(inode->i_atime) = body->mbo_atime;
1856                 lli->lli_atime = body->mbo_atime;
1857         }
1858
1859         if (body->mbo_valid & OBD_MD_FLMTIME) {
1860                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1861                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1862                                "to "LPU64"\n", inode->i_ino,
1863                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1864                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1865                 }
1866                 lli->lli_mtime = body->mbo_mtime;
1867         }
1868
1869         if (body->mbo_valid & OBD_MD_FLCTIME) {
1870                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1871                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1872                 lli->lli_ctime = body->mbo_ctime;
1873         }
1874
1875         if (body->mbo_valid & OBD_MD_FLMODE)
1876                 inode->i_mode = (inode->i_mode & S_IFMT) |
1877                                 (body->mbo_mode & ~S_IFMT);
1878
1879         if (body->mbo_valid & OBD_MD_FLTYPE)
1880                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1881                                 (body->mbo_mode & S_IFMT);
1882
1883         LASSERT(inode->i_mode != 0);
1884         if (S_ISREG(inode->i_mode))
1885                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1886                                        LL_MAX_BLKSIZE_BITS);
1887         else
1888                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1889
1890         if (body->mbo_valid & OBD_MD_FLUID)
1891                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1892         if (body->mbo_valid & OBD_MD_FLGID)
1893                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1894         if (body->mbo_valid & OBD_MD_FLFLAGS)
1895                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1896         if (body->mbo_valid & OBD_MD_FLNLINK)
1897                 set_nlink(inode, body->mbo_nlink);
1898         if (body->mbo_valid & OBD_MD_FLRDEV)
1899                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1900
1901         if (body->mbo_valid & OBD_MD_FLID) {
1902                 /* FID shouldn't be changed! */
1903                 if (fid_is_sane(&lli->lli_fid)) {
1904                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1905                                  "Trying to change FID "DFID
1906                                  " to the "DFID", inode "DFID"(%p)\n",
1907                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1908                                  PFID(ll_inode2fid(inode)), inode);
1909                 } else {
1910                         lli->lli_fid = body->mbo_fid1;
1911                 }
1912         }
1913
1914         LASSERT(fid_seq(&lli->lli_fid) != 0);
1915
1916         if (body->mbo_valid & OBD_MD_FLSIZE) {
1917                 i_size_write(inode, body->mbo_size);
1918
1919                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1920                        PFID(ll_inode2fid(inode)),
1921                        (unsigned long long)body->mbo_size);
1922
1923                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1924                         inode->i_blocks = body->mbo_blocks;
1925         }
1926
1927         if (body->mbo_valid & OBD_MD_TSTATE) {
1928                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1929                  * clear it when done to ensure to start again
1930                  * glimpsing updated attrs
1931                  */
1932                 if (body->mbo_t_state & MS_RESTORE)
1933                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1934                 else
1935                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1936         }
1937
1938         return 0;
1939 }
1940
1941 int ll_read_inode2(struct inode *inode, void *opaque)
1942 {
1943         struct lustre_md *md = opaque;
1944         struct ll_inode_info *lli = ll_i2info(inode);
1945         int     rc;
1946         ENTRY;
1947
1948         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1949                PFID(&lli->lli_fid), inode);
1950
1951         /* Core attributes from the MDS first.  This is a new inode, and
1952          * the VFS doesn't zero times in the core inode so we have to do
1953          * it ourselves.  They will be overwritten by either MDS or OST
1954          * attributes - we just need to make sure they aren't newer. */
1955         LTIME_S(inode->i_mtime) = 0;
1956         LTIME_S(inode->i_atime) = 0;
1957         LTIME_S(inode->i_ctime) = 0;
1958         inode->i_rdev = 0;
1959         rc = ll_update_inode(inode, md);
1960         if (rc != 0)
1961                 RETURN(rc);
1962
1963         /* OIDEBUG(inode); */
1964
1965 #ifdef HAVE_BACKING_DEV_INFO
1966         /* initializing backing dev info. */
1967         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1968 #endif
1969         if (S_ISREG(inode->i_mode)) {
1970                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1971                 inode->i_op = &ll_file_inode_operations;
1972                 inode->i_fop = sbi->ll_fop;
1973                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1974                 EXIT;
1975         } else if (S_ISDIR(inode->i_mode)) {
1976                 inode->i_op = &ll_dir_inode_operations;
1977                 inode->i_fop = &ll_dir_operations;
1978                 EXIT;
1979         } else if (S_ISLNK(inode->i_mode)) {
1980                 inode->i_op = &ll_fast_symlink_inode_operations;
1981                 EXIT;
1982         } else {
1983                 inode->i_op = &ll_special_inode_operations;
1984
1985                 init_special_inode(inode, inode->i_mode,
1986                                    inode->i_rdev);
1987
1988                 EXIT;
1989         }
1990
1991         return 0;
1992 }
1993
1994 void ll_delete_inode(struct inode *inode)
1995 {
1996         struct ll_inode_info *lli = ll_i2info(inode);
1997         ENTRY;
1998
1999         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2000                 /* It is last chance to write out dirty pages,
2001                  * otherwise we may lose data while umount */
2002                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2003
2004         truncate_inode_pages_final(&inode->i_data);
2005
2006         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2007                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2008                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2009
2010 #ifdef HAVE_SBOPS_EVICT_INODE
2011         ll_clear_inode(inode);
2012 #endif
2013         clear_inode(inode);
2014
2015         EXIT;
2016 }
2017
2018 int ll_iocontrol(struct inode *inode, struct file *file,
2019                  unsigned int cmd, unsigned long arg)
2020 {
2021         struct ll_sb_info *sbi = ll_i2sbi(inode);
2022         struct ptlrpc_request *req = NULL;
2023         int rc, flags = 0;
2024         ENTRY;
2025
2026         switch(cmd) {
2027         case FSFILT_IOC_GETFLAGS: {
2028                 struct mdt_body *body;
2029                 struct md_op_data *op_data;
2030
2031                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2032                                              0, 0, LUSTRE_OPC_ANY,
2033                                              NULL);
2034                 if (IS_ERR(op_data))
2035                         RETURN(PTR_ERR(op_data));
2036
2037                 op_data->op_valid = OBD_MD_FLFLAGS;
2038                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2039                 ll_finish_md_op_data(op_data);
2040                 if (rc) {
2041                         CERROR("%s: failure inode "DFID": rc = %d\n",
2042                                sbi->ll_md_exp->exp_obd->obd_name,
2043                                PFID(ll_inode2fid(inode)), rc);
2044                         RETURN(-abs(rc));
2045                 }
2046
2047                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2048
2049                 flags = body->mbo_flags;
2050
2051                 ptlrpc_req_finished(req);
2052
2053                 RETURN(put_user(flags, (int __user *)arg));
2054         }
2055         case FSFILT_IOC_SETFLAGS: {
2056                 struct iattr *attr;
2057                 struct md_op_data *op_data;
2058                 struct cl_object *obj;
2059
2060                 if (get_user(flags, (int __user *)arg))
2061                         RETURN(-EFAULT);
2062
2063                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2064                                              LUSTRE_OPC_ANY, NULL);
2065                 if (IS_ERR(op_data))
2066                         RETURN(PTR_ERR(op_data));
2067
2068                 op_data->op_attr_flags = flags;
2069                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2070                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2071                 ll_finish_md_op_data(op_data);
2072                 ptlrpc_req_finished(req);
2073                 if (rc)
2074                         RETURN(rc);
2075
2076                 inode->i_flags = ll_ext_to_inode_flags(flags);
2077
2078                 obj = ll_i2info(inode)->lli_clob;
2079                 if (obj == NULL)
2080                         RETURN(0);
2081
2082                 OBD_ALLOC_PTR(attr);
2083                 if (attr == NULL)
2084                         RETURN(-ENOMEM);
2085
2086                 attr->ia_valid = ATTR_ATTR_FLAG;
2087                 rc = cl_setattr_ost(obj, attr, flags);
2088
2089                 OBD_FREE_PTR(attr);
2090                 RETURN(rc);
2091         }
2092         default:
2093                 RETURN(-ENOSYS);
2094         }
2095
2096         RETURN(0);
2097 }
2098
2099 int ll_flush_ctx(struct inode *inode)
2100 {
2101         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2102
2103         CDEBUG(D_SEC, "flush context for user %d\n",
2104                from_kuid(&init_user_ns, current_uid()));
2105
2106         obd_set_info_async(NULL, sbi->ll_md_exp,
2107                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2108                            0, NULL, NULL);
2109         obd_set_info_async(NULL, sbi->ll_dt_exp,
2110                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2111                            0, NULL, NULL);
2112         return 0;
2113 }
2114
2115 /* umount -f client means force down, don't save state */
2116 void ll_umount_begin(struct super_block *sb)
2117 {
2118         struct ll_sb_info *sbi = ll_s2sbi(sb);
2119         struct obd_device *obd;
2120         struct obd_ioctl_data *ioc_data;
2121         ENTRY;
2122
2123         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2124                sb->s_count, atomic_read(&sb->s_active));
2125
2126         obd = class_exp2obd(sbi->ll_md_exp);
2127         if (obd == NULL) {
2128                 CERROR("Invalid MDC connection handle "LPX64"\n",
2129                        sbi->ll_md_exp->exp_handle.h_cookie);
2130                 EXIT;
2131                 return;
2132         }
2133         obd->obd_force = 1;
2134
2135         obd = class_exp2obd(sbi->ll_dt_exp);
2136         if (obd == NULL) {
2137                 CERROR("Invalid LOV connection handle "LPX64"\n",
2138                        sbi->ll_dt_exp->exp_handle.h_cookie);
2139                 EXIT;
2140                 return;
2141         }
2142         obd->obd_force = 1;
2143
2144         OBD_ALLOC_PTR(ioc_data);
2145         if (ioc_data) {
2146                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2147                               sizeof *ioc_data, ioc_data, NULL);
2148
2149                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2150                               sizeof *ioc_data, ioc_data, NULL);
2151
2152                 OBD_FREE_PTR(ioc_data);
2153         }
2154
2155         /* Really, we'd like to wait until there are no requests outstanding,
2156          * and then continue.  For now, we just invalidate the requests,
2157          * schedule() and sleep one second if needed, and hope.
2158          */
2159         schedule();
2160         EXIT;
2161 }
2162
2163 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2164 {
2165         struct ll_sb_info *sbi = ll_s2sbi(sb);
2166         char *profilenm = get_profile_name(sb);
2167         int err;
2168         __u32 read_only;
2169
2170         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2171                 read_only = *flags & MS_RDONLY;
2172                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2173                                          sizeof(KEY_READ_ONLY),
2174                                          KEY_READ_ONLY, sizeof(read_only),
2175                                          &read_only, NULL);
2176                 if (err) {
2177                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2178                                       profilenm, read_only ?
2179                                       "read-only" : "read-write", err);
2180                         return err;
2181                 }
2182
2183                 if (read_only)
2184                         sb->s_flags |= MS_RDONLY;
2185                 else
2186                         sb->s_flags &= ~MS_RDONLY;
2187
2188                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2189                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2190                                       read_only ?  "read-only" : "read-write");
2191         }
2192         return 0;
2193 }
2194
2195 /**
2196  * Cleanup the open handle that is cached on MDT-side.
2197  *
2198  * For open case, the client side open handling thread may hit error
2199  * after the MDT grant the open. Under such case, the client should
2200  * send close RPC to the MDT as cleanup; otherwise, the open handle
2201  * on the MDT will be leaked there until the client umount or evicted.
2202  *
2203  * In further, if someone unlinked the file, because the open handle
2204  * holds the reference on such file/object, then it will block the
2205  * subsequent threads that want to locate such object via FID.
2206  *
2207  * \param[in] sb        super block for this file-system
2208  * \param[in] open_req  pointer to the original open request
2209  */
2210 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2211 {
2212         struct mdt_body                 *body;
2213         struct md_op_data               *op_data;
2214         struct ptlrpc_request           *close_req = NULL;
2215         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2216         ENTRY;
2217
2218         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2219         OBD_ALLOC_PTR(op_data);
2220         if (op_data == NULL) {
2221                 CWARN("%s: cannot allocate op_data to release open handle for "
2222                       DFID"\n",
2223                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2224
2225                 RETURN_EXIT;
2226         }
2227
2228         op_data->op_fid1 = body->mbo_fid1;
2229         op_data->op_handle = body->mbo_handle;
2230         op_data->op_mod_time = cfs_time_current_sec();
2231         md_close(exp, op_data, NULL, &close_req);
2232         ptlrpc_req_finished(close_req);
2233         ll_finish_md_op_data(op_data);
2234
2235         EXIT;
2236 }
2237
2238 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2239                   struct super_block *sb, struct lookup_intent *it)
2240 {
2241         struct ll_sb_info *sbi = NULL;
2242         struct lustre_md md = { NULL };
2243         int rc;
2244         ENTRY;
2245
2246         LASSERT(*inode || sb);
2247         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2248         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2249                               sbi->ll_md_exp, &md);
2250         if (rc != 0)
2251                 GOTO(cleanup, rc);
2252
2253         if (*inode) {
2254                 rc = ll_update_inode(*inode, &md);
2255                 if (rc != 0)
2256                         GOTO(out, rc);
2257         } else {
2258                 LASSERT(sb != NULL);
2259
2260                 /*
2261                  * At this point server returns to client's same fid as client
2262                  * generated for creating. So using ->fid1 is okay here.
2263                  */
2264                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2265                         CERROR("%s: Fid is insane "DFID"\n",
2266                                 ll_get_fsname(sb, NULL, 0),
2267                                 PFID(&md.body->mbo_fid1));
2268                         GOTO(out, rc = -EINVAL);
2269                 }
2270
2271                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2272                                              sbi->ll_flags & LL_SBI_32BIT_API),
2273                                  &md);
2274                 if (IS_ERR(*inode)) {
2275 #ifdef CONFIG_FS_POSIX_ACL
2276                         if (md.posix_acl) {
2277                                 posix_acl_release(md.posix_acl);
2278                                 md.posix_acl = NULL;
2279                         }
2280 #endif
2281                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2282                         *inode = NULL;
2283                         CERROR("new_inode -fatal: rc %d\n", rc);
2284                         GOTO(out, rc);
2285                 }
2286         }
2287
2288         /* Handling piggyback layout lock.
2289          * Layout lock can be piggybacked by getattr and open request.
2290          * The lsm can be applied to inode only if it comes with a layout lock
2291          * otherwise correct layout may be overwritten, for example:
2292          * 1. proc1: mdt returns a lsm but not granting layout
2293          * 2. layout was changed by another client
2294          * 3. proc2: refresh layout and layout lock granted
2295          * 4. proc1: to apply a stale layout */
2296         if (it != NULL && it->it_lock_mode != 0) {
2297                 struct lustre_handle lockh;
2298                 struct ldlm_lock *lock;
2299
2300                 lockh.cookie = it->it_lock_handle;
2301                 lock = ldlm_handle2lock(&lockh);
2302                 LASSERT(lock != NULL);
2303                 if (ldlm_has_layout(lock)) {
2304                         struct cl_object_conf conf;
2305
2306                         memset(&conf, 0, sizeof(conf));
2307                         conf.coc_opc = OBJECT_CONF_SET;
2308                         conf.coc_inode = *inode;
2309                         conf.coc_lock = lock;
2310                         conf.u.coc_layout = md.layout;
2311                         (void)ll_layout_conf(*inode, &conf);
2312                 }
2313                 LDLM_LOCK_PUT(lock);
2314         }
2315
2316         GOTO(out, rc = 0);
2317
2318 out:
2319         md_free_lustre_md(sbi->ll_md_exp, &md);
2320
2321 cleanup:
2322         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2323                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2324
2325         return rc;
2326 }
2327
2328 int ll_obd_statfs(struct inode *inode, void __user *arg)
2329 {
2330         struct ll_sb_info *sbi = NULL;
2331         struct obd_export *exp;
2332         char *buf = NULL;
2333         struct obd_ioctl_data *data = NULL;
2334         __u32 type;
2335         int len = 0, rc;
2336
2337         if (!inode || !(sbi = ll_i2sbi(inode)))
2338                 GOTO(out_statfs, rc = -EINVAL);
2339
2340         rc = obd_ioctl_getdata(&buf, &len, arg);
2341         if (rc)
2342                 GOTO(out_statfs, rc);
2343
2344         data = (void*)buf;
2345         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2346             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2347                 GOTO(out_statfs, rc = -EINVAL);
2348
2349         if (data->ioc_inllen1 != sizeof(__u32) ||
2350             data->ioc_inllen2 != sizeof(__u32) ||
2351             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2352             data->ioc_plen2 != sizeof(struct obd_uuid))
2353                 GOTO(out_statfs, rc = -EINVAL);
2354
2355         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2356         if (type & LL_STATFS_LMV)
2357                 exp = sbi->ll_md_exp;
2358         else if (type & LL_STATFS_LOV)
2359                 exp = sbi->ll_dt_exp;
2360         else
2361                 GOTO(out_statfs, rc = -ENODEV);
2362
2363         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2364         if (rc)
2365                 GOTO(out_statfs, rc);
2366 out_statfs:
2367         if (buf)
2368                 obd_ioctl_freedata(buf, len);
2369         return rc;
2370 }
2371
2372 int ll_process_config(struct lustre_cfg *lcfg)
2373 {
2374         struct super_block *sb;
2375         unsigned long x;
2376         int rc = 0;
2377         char *ptr;
2378
2379         /* The instance name contains the sb: lustre-client-aacfe000 */
2380         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2381         if (!ptr || !*(++ptr))
2382                 return -EINVAL;
2383         if (sscanf(ptr, "%lx", &x) != 1)
2384                 return -EINVAL;
2385         sb = (struct super_block *)x;
2386         /* This better be a real Lustre superblock! */
2387         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2388
2389         /* Note we have not called client_common_fill_super yet, so
2390            proc fns must be able to handle that! */
2391         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2392                                       lcfg, sb);
2393         if (rc > 0)
2394                 rc = 0;
2395         return rc;
2396 }
2397
2398 /* this function prepares md_op_data hint for passing it down to MD stack. */
2399 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2400                                       struct inode *i1, struct inode *i2,
2401                                       const char *name, size_t namelen,
2402                                       __u32 mode, __u32 opc, void *data)
2403 {
2404         LASSERT(i1 != NULL);
2405
2406         if (name == NULL) {
2407                 /* Do not reuse namelen for something else. */
2408                 if (namelen != 0)
2409                         return ERR_PTR(-EINVAL);
2410         } else {
2411                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2412                         return ERR_PTR(-ENAMETOOLONG);
2413
2414                 if (!lu_name_is_valid_2(name, namelen))
2415                         return ERR_PTR(-EINVAL);
2416         }
2417
2418         if (op_data == NULL)
2419                 OBD_ALLOC_PTR(op_data);
2420
2421         if (op_data == NULL)
2422                 return ERR_PTR(-ENOMEM);
2423
2424         ll_i2gids(op_data->op_suppgids, i1, i2);
2425         op_data->op_fid1 = *ll_inode2fid(i1);
2426         op_data->op_default_stripe_offset = -1;
2427         if (S_ISDIR(i1->i_mode)) {
2428                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2429                 if (opc == LUSTRE_OPC_MKDIR)
2430                         op_data->op_default_stripe_offset =
2431                                    ll_i2info(i1)->lli_def_stripe_offset;
2432         }
2433
2434         if (i2) {
2435                 op_data->op_fid2 = *ll_inode2fid(i2);
2436                 if (S_ISDIR(i2->i_mode))
2437                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2438         } else {
2439                 fid_zero(&op_data->op_fid2);
2440         }
2441
2442         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2443                 op_data->op_cli_flags |= CLI_HASH64;
2444
2445         if (ll_need_32bit_api(ll_i2sbi(i1)))
2446                 op_data->op_cli_flags |= CLI_API32;
2447
2448         op_data->op_name = name;
2449         op_data->op_namelen = namelen;
2450         op_data->op_mode = mode;
2451         op_data->op_mod_time = cfs_time_current_sec();
2452         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2453         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2454         op_data->op_cap = cfs_curproc_cap_pack();
2455         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2456              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2457                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2458         } else {
2459                 op_data->op_mds = 0;
2460         }
2461         op_data->op_data = data;
2462
2463         return op_data;
2464 }
2465
2466 void ll_finish_md_op_data(struct md_op_data *op_data)
2467 {
2468         OBD_FREE_PTR(op_data);
2469 }
2470
2471 #ifdef HAVE_SUPEROPS_USE_DENTRY
2472 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2473 #else
2474 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2475 #endif
2476 {
2477         struct ll_sb_info *sbi;
2478
2479 #ifdef HAVE_SUPEROPS_USE_DENTRY
2480         LASSERT((seq != NULL) && (dentry != NULL));
2481         sbi = ll_s2sbi(dentry->d_sb);
2482 #else
2483         LASSERT((seq != NULL) && (vfs != NULL));
2484         sbi = ll_s2sbi(vfs->mnt_sb);
2485 #endif
2486
2487         if (sbi->ll_flags & LL_SBI_NOLCK)
2488                 seq_puts(seq, ",nolock");
2489
2490         if (sbi->ll_flags & LL_SBI_FLOCK)
2491                 seq_puts(seq, ",flock");
2492
2493         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2494                 seq_puts(seq, ",localflock");
2495
2496         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2497                 seq_puts(seq, ",user_xattr");
2498
2499         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2500                 seq_puts(seq, ",lazystatfs");
2501
2502         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2503                 seq_puts(seq, ",user_fid2path");
2504
2505         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2506                 seq_puts(seq, ",always_ping");
2507
2508         RETURN(0);
2509 }
2510
2511 /**
2512  * Get obd name by cmd, and copy out to user space
2513  */
2514 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2515 {
2516         struct ll_sb_info *sbi = ll_i2sbi(inode);
2517         struct obd_device *obd;
2518         ENTRY;
2519
2520         if (cmd == OBD_IOC_GETDTNAME)
2521                 obd = class_exp2obd(sbi->ll_dt_exp);
2522         else if (cmd == OBD_IOC_GETMDNAME)
2523                 obd = class_exp2obd(sbi->ll_md_exp);
2524         else
2525                 RETURN(-EINVAL);
2526
2527         if (!obd)
2528                 RETURN(-ENOENT);
2529
2530         if (copy_to_user((void __user *)arg, obd->obd_name,
2531                          strlen(obd->obd_name) + 1))
2532                 RETURN(-EFAULT);
2533
2534         RETURN(0);
2535 }
2536
2537 /**
2538  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2539  * fsname will be returned in this buffer; otherwise, a static buffer will be
2540  * used to store the fsname and returned to caller.
2541  */
2542 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2543 {
2544         static char fsname_static[MTI_NAME_MAXLEN];
2545         struct lustre_sb_info *lsi = s2lsi(sb);
2546         char *ptr;
2547         int len;
2548
2549         if (buf == NULL) {
2550                 /* this means the caller wants to use static buffer
2551                  * and it doesn't care about race. Usually this is
2552                  * in error reporting path */
2553                 buf = fsname_static;
2554                 buflen = sizeof(fsname_static);
2555         }
2556
2557         len = strlen(lsi->lsi_lmd->lmd_profile);
2558         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2559         if (ptr && (strcmp(ptr, "-client") == 0))
2560                 len -= 7;
2561
2562         if (unlikely(len >= buflen))
2563                 len = buflen - 1;
2564         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2565         buf[len] = '\0';
2566
2567         return buf;
2568 }
2569
2570 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2571 {
2572         char *path = NULL;
2573
2574         struct path p;
2575
2576         p.dentry = dentry;
2577         p.mnt = current->fs->root.mnt;
2578         path_get(&p);
2579         path = d_path(&p, buf, bufsize);
2580         path_put(&p);
2581         return path;
2582 }
2583
2584 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2585 {
2586         char *buf, *path = NULL;
2587         struct dentry *dentry = NULL;
2588         struct inode *inode = page->mapping->host;
2589
2590         /* this can be called inside spin lock so use GFP_ATOMIC. */
2591         buf = (char *)__get_free_page(GFP_ATOMIC);
2592         if (buf != NULL) {
2593                 dentry = d_find_alias(page->mapping->host);
2594                 if (dentry != NULL)
2595                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2596         }
2597
2598         CDEBUG(D_WARNING,
2599                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2600                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2601                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2602                PFID(ll_inode2fid(inode)),
2603                (path && !IS_ERR(path)) ? path : "", ioret);
2604
2605         if (dentry != NULL)
2606                 dput(dentry);
2607
2608         if (buf != NULL)
2609                 free_page((unsigned long)buf);
2610 }
2611
2612 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2613                         struct lov_user_md **kbuf)
2614 {
2615         struct lov_user_md      lum;
2616         ssize_t                 lum_size;
2617         ENTRY;
2618
2619         if (copy_from_user(&lum, md, sizeof(lum)))
2620                 RETURN(-EFAULT);
2621
2622         lum_size = ll_lov_user_md_size(&lum);
2623         if (lum_size < 0)
2624                 RETURN(lum_size);
2625
2626         OBD_ALLOC(*kbuf, lum_size);
2627         if (*kbuf == NULL)
2628                 RETURN(-ENOMEM);
2629
2630         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2631                 OBD_FREE(*kbuf, lum_size);
2632                 RETURN(-EFAULT);
2633         }
2634
2635         RETURN(lum_size);
2636 }
2637
2638 /*
2639  * Compute llite root squash state after a change of root squash
2640  * configuration setting or add/remove of a lnet nid
2641  */
2642 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2643 {
2644         struct root_squash_info *squash = &sbi->ll_squash;
2645         int i;
2646         bool matched;
2647         lnet_process_id_t id;
2648
2649         /* Update norootsquash flag */
2650         down_write(&squash->rsi_sem);
2651         if (list_empty(&squash->rsi_nosquash_nids))
2652                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2653         else {
2654                 /* Do not apply root squash as soon as one of our NIDs is
2655                  * in the nosquash_nids list */
2656                 matched = false;
2657                 i = 0;
2658                 while (LNetGetId(i++, &id) != -ENOENT) {
2659                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2660                                 continue;
2661                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2662                                 matched = true;
2663                                 break;
2664                         }
2665                 }
2666                 if (matched)
2667                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2668                 else
2669                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2670         }
2671         up_write(&squash->rsi_sem);
2672 }
2673
2674 /**
2675  * Parse linkea content to extract information about a given hardlink
2676  *
2677  * \param[in]   ldata      - Initialized linkea data
2678  * \param[in]   linkno     - Link identifier
2679  * \param[out]  parent_fid - The entry's parent FID
2680  * \param[out]  ln         - Entry name destination buffer
2681  *
2682  * \retval 0 on success
2683  * \retval Appropriate negative error code on failure
2684  */
2685 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2686                             struct lu_fid *parent_fid, struct lu_name *ln)
2687 {
2688         unsigned int    idx;
2689         int             rc;
2690         ENTRY;
2691
2692         rc = linkea_init(ldata);
2693         if (rc < 0)
2694                 RETURN(rc);
2695
2696         if (linkno >= ldata->ld_leh->leh_reccount)
2697                 /* beyond last link */
2698                 RETURN(-ENODATA);
2699
2700         linkea_first_entry(ldata);
2701         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2702                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2703                                     parent_fid);
2704                 if (idx == linkno)
2705                         break;
2706
2707                 linkea_next_entry(ldata);
2708         }
2709
2710         if (idx < linkno)
2711                 RETURN(-ENODATA);
2712
2713         RETURN(0);
2714 }
2715
2716 /**
2717  * Get parent FID and name of an identified link. Operation is performed for
2718  * a given link number, letting the caller iterate over linkno to list one or
2719  * all links of an entry.
2720  *
2721  * \param[in]     file - File descriptor against which to perform the operation
2722  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2723  *                       on and the available size. It is eventually filled with
2724  *                       the requested information or left untouched on error
2725  *
2726  * \retval - 0 on success
2727  * \retval - Appropriate negative error code on failure
2728  */
2729 int ll_getparent(struct file *file, struct getparent __user *arg)
2730 {
2731         struct dentry           *dentry = file_dentry(file);
2732         struct inode            *inode = file_inode(file);
2733         struct linkea_data      *ldata;
2734         struct lu_buf            buf = LU_BUF_NULL;
2735         struct lu_name           ln;
2736         struct lu_fid            parent_fid;
2737         __u32                    linkno;
2738         __u32                    name_size;
2739         int                      rc;
2740
2741         ENTRY;
2742
2743         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2744             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2745                 RETURN(-EPERM);
2746
2747         if (get_user(name_size, &arg->gp_name_size))
2748                 RETURN(-EFAULT);
2749
2750         if (get_user(linkno, &arg->gp_linkno))
2751                 RETURN(-EFAULT);
2752
2753         if (name_size > PATH_MAX)
2754                 RETURN(-EINVAL);
2755
2756         OBD_ALLOC(ldata, sizeof(*ldata));
2757         if (ldata == NULL)
2758                 RETURN(-ENOMEM);
2759
2760         rc = linkea_data_new(ldata, &buf);
2761         if (rc < 0)
2762                 GOTO(ldata_free, rc);
2763
2764         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2765         if (rc < 0)
2766                 GOTO(lb_free, rc);
2767
2768         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2769         if (rc < 0)
2770                 GOTO(lb_free, rc);
2771
2772         if (ln.ln_namelen >= name_size)
2773                 GOTO(lb_free, rc = -EOVERFLOW);
2774
2775         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2776                 GOTO(lb_free, rc = -EFAULT);
2777
2778         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2779                 GOTO(lb_free, rc = -EFAULT);
2780
2781         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2782                 GOTO(lb_free, rc = -EFAULT);
2783
2784 lb_free:
2785         lu_buf_free(&buf);
2786 ldata_free:
2787         OBD_FREE(ldata, sizeof(*ldata));
2788
2789         RETURN(rc);
2790 }