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         /* real client */
 248         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 249
 250         /* always ping even if server suppress_pings */
 251         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 252                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 253
 254         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 255
 256         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 257         if (err == -EBUSY) {
 258                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 259                                    "recovery, of which this client is not a "
 260                                    "part. Please wait for recovery to complete,"
 261                                    " abort, or time out.\n", md);
 262                 GOTO(out, err);
 263         } else if (err) {
 264                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 265                 GOTO(out, err);
 266         }
 267
 268         sbi->ll_md_exp->exp_connect_data = *data;
 269
 270         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 271                            LUSTRE_SEQ_METADATA);
 272         if (err) {
 273                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 274                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 275                 GOTO(out_md, err);
 276         }
 277
 278         /* For mount, we only need fs info from MDT0, and also in DNE, it
 279          * can make sure the client can be mounted as long as MDT0 is
 280          * avaible */
 281         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 282                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 283                         OBD_STATFS_FOR_MDT0);
 284         if (err)
 285                 GOTO(out_md_fid, err);
 286
 287         /* This needs to be after statfs to ensure connect has finished.
 288          * Note that "data" does NOT contain the valid connect reply.
 289          * If connecting to a 1.8 server there will be no LMV device, so
 290          * we can access the MDC export directly and exp_connect_flags will
 291          * be non-zero, but if accessing an upgraded 2.1 server it will
 292          * have the correct flags filled in.
 293          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 294         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 295         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 296             valid != CLIENT_CONNECT_MDT_REQD) {
 297                 char *buf;
 298
 299                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 300                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 301                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 302                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 303                                    "feature(s) needed for correct operation "
 304                                    "of this client (%s). Please upgrade "
 305                                    "server or downgrade client.\n",
 306                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 307                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 308                 GOTO(out_md_fid, err = -EPROTO);
 309         }
 310
 311         size = sizeof(*data);
 312         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 313                            KEY_CONN_DATA,  &size, data);
 314         if (err) {
 315                 CERROR("%s: Get connect data failed: rc = %d\n",
 316                        sbi->ll_md_exp->exp_obd->obd_name, err);
 317                 GOTO(out_md_fid, err);
 318         }
 319
 320         LASSERT(osfs->os_bsize);
 321         sb->s_blocksize = osfs->os_bsize;
 322         sb->s_blocksize_bits = log2(osfs->os_bsize);
 323         sb->s_magic = LL_SUPER_MAGIC;
 324         sb->s_maxbytes = MAX_LFS_FILESIZE;
 325         sbi->ll_namelen = osfs->os_namelen;
 326
 327         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 328             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 329                 LCONSOLE_INFO("Disabling user_xattr feature because "
 330                               "it is not supported on the server\n");
 331                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 332         }
 333
 334         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 335 #ifdef MS_POSIXACL
 336                 sb->s_flags |= MS_POSIXACL;
 337 #endif
 338                 sbi->ll_flags |= LL_SBI_ACL;
 339         } else {
 340                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 341 #ifdef MS_POSIXACL
 342                 sb->s_flags &= ~MS_POSIXACL;
 343 #endif
 344                 sbi->ll_flags &= ~LL_SBI_ACL;
 345         }
 346
 347         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 348                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 349
 350         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 351                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 352         else
 353                 sbi->ll_md_brw_pages = 1;
 354
 355         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 356                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 357
 358         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 359                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 360                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 361                                       "unknown maximum xattr size.\n", dt);
 362                 } else {
 363                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 364                         sbi->ll_xattr_cache_enabled = 1;
 365                 }
 366         }
 367
 368         obd = class_name2obd(dt);
 369         if (!obd) {
 370                 CERROR("DT %s: not setup or attached\n", dt);
 371                 GOTO(out_md_fid, err = -ENODEV);
 372         }
 373
 374         /* pass client page size via ocd_grant_blkbits, the server should report
 375          * back its backend blocksize for grant calculation purpose */
 376         data->ocd_grant_blkbits = PAGE_SHIFT;
 377
 378         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 379                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 380                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 381                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 382                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 383                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 384                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 385                                   OBD_CONNECT_EINPROGRESS |
 386                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 387                                   OBD_CONNECT_LAYOUTLOCK |
 388                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 389                                   OBD_CONNECT_BULK_MBITS;
 390
 391         data->ocd_connect_flags2 = 0;
 392
 393         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 394                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 395
 396         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 397                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 398                  * disabled by default, because it can still be enabled on the
 399                  * fly via /proc. As a consequence, we still need to come to an
 400                  * agreement on the supported algorithms at connect time */
 401                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 402
 403                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 404                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 405                 else
 406                         data->ocd_cksum_types = cksum_types_supported_client();
 407         }
 408
 409 #ifdef HAVE_LRU_RESIZE_SUPPORT
 410         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 411 #endif
 412         /* always ping even if server suppress_pings */
 413         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 414                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 415
 416         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 417                "ocd_grant: %d\n", data->ocd_connect_flags,
 418                data->ocd_version, data->ocd_grant);
 419
 420         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 421         obd->obd_upcall.onu_upcall = cl_ocd_update;
 422
 423         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 424
 425         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 426                           NULL);
 427         if (err == -EBUSY) {
 428                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 429                                    "recovery, of which this client is not a "
 430                                    "part.  Please wait for recovery to "
 431                                    "complete, abort, or time out.\n", dt);
 432                 GOTO(out_md, err);
 433         } else if (err) {
 434                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 435                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 436                 GOTO(out_md, err);
 437         }
 438
 439         sbi->ll_dt_exp->exp_connect_data = *data;
 440
 441         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 442                            LUSTRE_SEQ_METADATA);
 443         if (err) {
 444                 CERROR("%s: Can't init data layer FID infrastructure, "
 445                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 446                 GOTO(out_dt, err);
 447         }
 448
 449         mutex_lock(&sbi->ll_lco.lco_lock);
 450         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 451         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 452         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 453         mutex_unlock(&sbi->ll_lco.lco_lock);
 454
 455         fid_zero(&sbi->ll_root_fid);
 456         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 457                            &sbi->ll_root_fid);
 458         if (err) {
 459                 CERROR("cannot mds_connect: rc = %d\n", err);
 460                 GOTO(out_lock_cn_cb, err);
 461         }
 462         if (!fid_is_sane(&sbi->ll_root_fid)) {
 463                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 464                        sbi->ll_md_exp->exp_obd->obd_name,
 465                        PFID(&sbi->ll_root_fid));
 466                 GOTO(out_lock_cn_cb, err = -EINVAL);
 467         }
 468         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 469
 470         sb->s_op = &lustre_super_operations;
 471 #if THREAD_SIZE >= 8192 /*b=17630*/
 472         sb->s_export_op = &lustre_export_operations;
 473 #endif
 474
 475         /* make root inode
 476          * XXX: move this to after cbd setup? */
 477         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 478         if (sbi->ll_flags & LL_SBI_ACL)
 479                 valid |= OBD_MD_FLACL;
 480
 481         OBD_ALLOC_PTR(op_data);
 482         if (op_data == NULL)
 483                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 484
 485         op_data->op_fid1 = sbi->ll_root_fid;
 486         op_data->op_mode = 0;
 487         op_data->op_valid = valid;
 488
 489         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 490
 491         OBD_FREE_PTR(op_data);
 492         if (err) {
 493                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 494                        sbi->ll_md_exp->exp_obd->obd_name, err);
 495                 GOTO(out_lock_cn_cb, err);
 496         }
 497
 498         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 499                                sbi->ll_md_exp, &lmd);
 500         if (err) {
 501                 CERROR("failed to understand root inode md: rc = %d\n", err);
 502                 ptlrpc_req_finished(request);
 503                 GOTO(out_lock_cn_cb, err);
 504         }
 505
 506         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 507         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 508                                             sbi->ll_flags & LL_SBI_32BIT_API),
 509                        &lmd);
 510         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 511         ptlrpc_req_finished(request);
 512
 513         if (IS_ERR(root)) {
 514 #ifdef CONFIG_FS_POSIX_ACL
 515                 if (lmd.posix_acl) {
 516                         posix_acl_release(lmd.posix_acl);
 517                         lmd.posix_acl = NULL;
 518                 }
 519 #endif
 520                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 521                 root = NULL;
 522                 CERROR("lustre_lite: bad iget4 for root\n");
 523                 GOTO(out_root, err);
 524         }
 525
 526         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 527         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 528                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 529                                  NULL);
 530         if (err) {
 531                 CERROR("%s: Set checksum failed: rc = %d\n",
 532                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 533                 GOTO(out_root, err);
 534         }
 535         cl_sb_init(sb);
 536
 537         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 538                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 539                                  sbi->ll_cache, NULL);
 540         if (err) {
 541                 CERROR("%s: Set cache_set failed: rc = %d\n",
 542                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 543                 GOTO(out_root, err);
 544         }
 545
 546         sb->s_root = d_make_root(root);
 547         if (sb->s_root == NULL) {
 548                 CERROR("%s: can't make root dentry\n",
 549                         ll_get_fsname(sb, NULL, 0));
 550                 GOTO(out_root, err = -ENOMEM);
 551         }
 552 #ifdef HAVE_DCACHE_LOCK
 553         sb->s_root->d_op = &ll_d_ops;
 554 #endif
 555
 556         sbi->ll_sdev_orig = sb->s_dev;
 557
 558         /* We set sb->s_dev equal on all lustre clients in order to support
 559          * NFS export clustering.  NFSD requires that the FSID be the same
 560          * on all clients. */
 561         /* s_dev is also used in lt_compare() to compare two fs, but that is
 562          * only a node-local comparison. */
 563         uuid = obd_get_uuid(sbi->ll_md_exp);
 564         if (uuid != NULL)
 565                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 566
 567         if (data != NULL)
 568                 OBD_FREE_PTR(data);
 569         if (osfs != NULL)
 570                 OBD_FREE_PTR(osfs);
 571         if (proc_lustre_fs_root != NULL) {
 572                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 573                                                   dt, md);
 574                 if (err < 0) {
 575                         CERROR("%s: could not register mount in lprocfs: "
 576                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 577                         err = 0;
 578                 }
 579         }
 580
 581         RETURN(err);
 582 out_root:
 583         if (root)
 584                 iput(root);
 585 out_lock_cn_cb:
 586         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 587 out_dt:
 588         obd_disconnect(sbi->ll_dt_exp);
 589         sbi->ll_dt_exp = NULL;
 590 out_md_fid:
 591         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 592 out_md:
 593         obd_disconnect(sbi->ll_md_exp);
 594         sbi->ll_md_exp = NULL;
 595 out:
 596         if (data != NULL)
 597                 OBD_FREE_PTR(data);
 598         if (osfs != NULL)
 599                 OBD_FREE_PTR(osfs);
 600         return err;
 601 }
 602
 603 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 604 {
 605         int size, rc;
 606
 607         size = sizeof(*lmmsize);
 608         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 609                           KEY_MAX_EASIZE, &size, lmmsize);
 610         if (rc != 0) {
 611                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 612                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 613                 RETURN(rc);
 614         }
 615
 616         size = sizeof(int);
 617         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 618                           KEY_MAX_EASIZE, &size, lmmsize);
 619         if (rc)
 620                 CERROR("Get max mdsize error rc %d\n", rc);
 621
 622         RETURN(rc);
 623 }
 624
 625 /**
 626  * Get the value of the default_easize parameter.
 627  *
 628  * \see client_obd::cl_default_mds_easize
 629  *
 630  * \param[in] sbi       superblock info for this filesystem
 631  * \param[out] lmmsize  pointer to storage location for value
 632  *
 633  * \retval 0            on success
 634  * \retval negative     negated errno on failure
 635  */
 636 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 637 {
 638         int size, rc;
 639
 640         size = sizeof(int);
 641         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 642                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 643         if (rc)
 644                 CERROR("Get default mdsize error rc %d\n", rc);
 645
 646         RETURN(rc);
 647 }
 648
 649 /**
 650  * Set the default_easize parameter to the given value.
 651  *
 652  * \see client_obd::cl_default_mds_easize
 653  *
 654  * \param[in] sbi       superblock info for this filesystem
 655  * \param[in] lmmsize   the size to set
 656  *
 657  * \retval 0            on success
 658  * \retval negative     negated errno on failure
 659  */
 660 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 661 {
 662         int rc;
 663
 664         if (lmmsize < sizeof(struct lov_mds_md) ||
 665             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 666                 return -EINVAL;
 667
 668         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 669                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 670                                 sizeof(int), &lmmsize, NULL);
 671
 672         RETURN(rc);
 673 }
 674
 675 static void ll_dump_inode(struct inode *inode)
 676 {
 677         struct ll_d_hlist_node *tmp;
 678         int dentry_count = 0;
 679
 680         LASSERT(inode != NULL);
 681
 682         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 683                 dentry_count++;
 684
 685         CERROR("%s: inode %p dump: dev=%s fid="DFID
 686                " mode=%o count=%u, %d dentries\n",
 687                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 688                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 689                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 690 }
 691
 692 void lustre_dump_dentry(struct dentry *dentry, int recur)
 693 {
 694         struct list_head *tmp;
 695         int subdirs = 0;
 696
 697         LASSERT(dentry != NULL);
 698
 699         list_for_each(tmp, &dentry->d_subdirs)
 700                 subdirs++;
 701
 702         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 703                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 704                dentry->d_name.len, dentry->d_name.name,
 705                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 706                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 707                dentry->d_flags, dentry->d_fsdata, subdirs);
 708         if (dentry->d_inode != NULL)
 709                 ll_dump_inode(dentry->d_inode);
 710
 711         if (recur == 0)
 712                 return;
 713
 714         list_for_each(tmp, &dentry->d_subdirs) {
 715                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 716                 lustre_dump_dentry(d, recur - 1);
 717         }
 718 }
 719
 720 static void client_common_put_super(struct super_block *sb)
 721 {
 722         struct ll_sb_info *sbi = ll_s2sbi(sb);
 723         ENTRY;
 724
 725         cl_sb_fini(sb);
 726
 727         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 728         obd_disconnect(sbi->ll_dt_exp);
 729         sbi->ll_dt_exp = NULL;
 730
 731         lprocfs_unregister_mountpoint(sbi);
 732
 733         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 734         obd_disconnect(sbi->ll_md_exp);
 735         sbi->ll_md_exp = NULL;
 736
 737         EXIT;
 738 }
 739
 740 void ll_kill_super(struct super_block *sb)
 741 {
 742         struct ll_sb_info *sbi;
 743         ENTRY;
 744
 745         /* not init sb ?*/
 746         if (!(sb->s_flags & MS_ACTIVE))
 747                 return;
 748
 749         sbi = ll_s2sbi(sb);
 750         /* we need restore s_dev from changed for clustred NFS before put_super
 751          * because new kernels have cached s_dev and change sb->s_dev in
 752          * put_super not affected real removing devices */
 753         if (sbi) {
 754                 sb->s_dev = sbi->ll_sdev_orig;
 755                 sbi->ll_umounting = 1;
 756
 757                 /* wait running statahead threads to quit */
 758                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 759                         set_current_state(TASK_UNINTERRUPTIBLE);
 760                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 761                 }
 762         }
 763
 764         EXIT;
 765 }
 766
 767 static inline int ll_set_opt(const char *opt, char *data, int fl)
 768 {
 769         if (strncmp(opt, data, strlen(opt)) != 0)
 770                 return(0);
 771         else
 772                 return(fl);
 773 }
 774
 775 /* non-client-specific mount options are parsed in lmd_parse */
 776 static int ll_options(char *options, int *flags)
 777 {
 778         int tmp;
 779         char *s1 = options, *s2;
 780         ENTRY;
 781
 782         if (!options)
 783                 RETURN(0);
 784
 785         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 786
 787         while (*s1) {
 788                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 789                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 790                 if (tmp) {
 791                         *flags |= tmp;
 792                         goto next;
 793                 }
 794                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 795                 if (tmp) {
 796                         *flags |= tmp;
 797                         goto next;
 798                 }
 799                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 800                 if (tmp) {
 801                         *flags |= tmp;
 802                         goto next;
 803                 }
 804                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 805                 if (tmp) {
 806                         *flags &= ~tmp;
 807                         goto next;
 808                 }
 809                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 810                 if (tmp) {
 811                         *flags |= tmp;
 812                         goto next;
 813                 }
 814                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 815                 if (tmp) {
 816                         *flags &= ~tmp;
 817                         goto next;
 818                 }
 819                 tmp = ll_set_opt("context", s1, 1);
 820                 if (tmp)
 821                         goto next;
 822                 tmp = ll_set_opt("fscontext", s1, 1);
 823                 if (tmp)
 824                         goto next;
 825                 tmp = ll_set_opt("defcontext", s1, 1);
 826                 if (tmp)
 827                         goto next;
 828                 tmp = ll_set_opt("rootcontext", s1, 1);
 829                 if (tmp)
 830                         goto next;
 831                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 832                 if (tmp) {
 833                         *flags |= tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 837                 if (tmp) {
 838                         *flags &= ~tmp;
 839                         goto next;
 840                 }
 841
 842                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 848                 if (tmp) {
 849                         *flags &= ~tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 858                 if (tmp) {
 859                         *flags &= ~tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 863                 if (tmp) {
 864                         *flags |= tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 868                 if (tmp) {
 869                         *flags &= ~tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 873                 if (tmp) {
 874                         *flags |= tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 883                 if (tmp) {
 884                         *flags &= ~tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 888                 if (tmp) {
 889                         *flags |= tmp;
 890                         goto next;
 891                 }
 892                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 893                                    s1);
 894                 RETURN(-EINVAL);
 895
 896 next:
 897                 /* Find next opt */
 898                 s2 = strchr(s1, ',');
 899                 if (s2 == NULL)
 900                         break;
 901                 s1 = s2 + 1;
 902         }
 903         RETURN(0);
 904 }
 905
 906 void ll_lli_init(struct ll_inode_info *lli)
 907 {
 908         lli->lli_inode_magic = LLI_INODE_MAGIC;
 909         lli->lli_flags = 0;
 910         spin_lock_init(&lli->lli_lock);
 911         lli->lli_posix_acl = NULL;
 912         /* Do not set lli_fid, it has been initialized already. */
 913         fid_zero(&lli->lli_pfid);
 914         lli->lli_mds_read_och = NULL;
 915         lli->lli_mds_write_och = NULL;
 916         lli->lli_mds_exec_och = NULL;
 917         lli->lli_open_fd_read_count = 0;
 918         lli->lli_open_fd_write_count = 0;
 919         lli->lli_open_fd_exec_count = 0;
 920         mutex_init(&lli->lli_och_mutex);
 921         spin_lock_init(&lli->lli_agl_lock);
 922         spin_lock_init(&lli->lli_layout_lock);
 923         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 924         lli->lli_clob = NULL;
 925
 926         init_rwsem(&lli->lli_xattrs_list_rwsem);
 927         mutex_init(&lli->lli_xattrs_enq_lock);
 928
 929         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 930         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 931                 mutex_init(&lli->lli_readdir_mutex);
 932                 lli->lli_opendir_key = NULL;
 933                 lli->lli_sai = NULL;
 934                 spin_lock_init(&lli->lli_sa_lock);
 935                 lli->lli_opendir_pid = 0;
 936                 lli->lli_sa_enabled = 0;
 937                 lli->lli_def_stripe_offset = -1;
 938         } else {
 939                 mutex_init(&lli->lli_size_mutex);
 940                 lli->lli_symlink_name = NULL;
 941                 init_rwsem(&lli->lli_trunc_sem);
 942                 range_lock_tree_init(&lli->lli_write_tree);
 943                 init_rwsem(&lli->lli_glimpse_sem);
 944                 lli->lli_glimpse_time = 0;
 945                 INIT_LIST_HEAD(&lli->lli_agl_list);
 946                 lli->lli_agl_index = 0;
 947                 lli->lli_async_rc = 0;
 948         }
 949         mutex_init(&lli->lli_layout_mutex);
 950 }
 951
 952 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 953 {
 954         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 955
 956         bdi->name = "lustre";
 957         return bdi_register(bdi, NULL, "lustre-%d",
 958                             atomic_inc_return(&ll_bdi_num));
 959 }
 960
 961 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 962 {
 963         struct lustre_profile *lprof = NULL;
 964         struct lustre_sb_info *lsi = s2lsi(sb);
 965         struct ll_sb_info *sbi;
 966         char  *dt = NULL, *md = NULL;
 967         char  *profilenm = get_profile_name(sb);
 968         struct config_llog_instance *cfg;
 969         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 970         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 971         int    err;
 972         ENTRY;
 973
 974         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 975
 976         OBD_ALLOC_PTR(cfg);
 977         if (cfg == NULL)
 978                 RETURN(-ENOMEM);
 979
 980         try_module_get(THIS_MODULE);
 981
 982         /* client additional sb info */
 983         lsi->lsi_llsbi = sbi = ll_init_sbi();
 984         if (!sbi) {
 985                 module_put(THIS_MODULE);
 986                 OBD_FREE_PTR(cfg);
 987                 RETURN(-ENOMEM);
 988         }
 989
 990         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 991         if (err)
 992                 GOTO(out_free, err);
 993
 994         err = bdi_init(&lsi->lsi_bdi);
 995         if (err)
 996                 GOTO(out_free, err);
 997         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 998 #ifdef HAVE_BDI_CAP_MAP_COPY
 999         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1000 #else
1001         lsi->lsi_bdi.capabilities = 0;
1002 #endif
1003         err = ll_bdi_register(&lsi->lsi_bdi);
1004         if (err)
1005                 GOTO(out_free, err);
1006
1007         sb->s_bdi = &lsi->lsi_bdi;
1008 #ifndef HAVE_DCACHE_LOCK
1009         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1010         sb->s_d_op = &ll_d_ops;
1011 #endif
1012
1013         /* Generate a string unique to this super, in case some joker tries
1014            to mount the same fs at two mount points.
1015            Use the address of the super itself.*/
1016         cfg->cfg_instance = sb;
1017         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1018         cfg->cfg_callback = class_config_llog_handler;
1019         /* set up client obds */
1020         err = lustre_process_log(sb, profilenm, cfg);
1021         if (err < 0)
1022                 GOTO(out_free, err);
1023
1024         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1025         lprof = class_get_profile(profilenm);
1026         if (lprof == NULL) {
1027                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1028                                    " read from the MGS.  Does that filesystem "
1029                                    "exist?\n", profilenm);
1030                 GOTO(out_free, err = -EINVAL);
1031         }
1032         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1033                lprof->lp_md, lprof->lp_dt);
1034
1035         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1036         if (!dt)
1037                 GOTO(out_free, err = -ENOMEM);
1038         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1039
1040         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1041         if (!md)
1042                 GOTO(out_free, err = -ENOMEM);
1043         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1044
1045         /* connections, registrations, sb setup */
1046         err = client_common_fill_super(sb, md, dt, mnt);
1047         if (err < 0)
1048                 GOTO(out_free, err);
1049
1050         sbi->ll_client_common_fill_super_succeeded = 1;
1051
1052 out_free:
1053         if (md)
1054                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1055         if (dt)
1056                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1057         if (lprof != NULL)
1058                 class_put_profile(lprof);
1059         if (err)
1060                 ll_put_super(sb);
1061         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1062                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1063
1064         OBD_FREE_PTR(cfg);
1065         RETURN(err);
1066 } /* ll_fill_super */
1067
1068 void ll_put_super(struct super_block *sb)
1069 {
1070         struct config_llog_instance cfg, params_cfg;
1071         struct obd_device *obd;
1072         struct lustre_sb_info *lsi = s2lsi(sb);
1073         struct ll_sb_info *sbi = ll_s2sbi(sb);
1074         char *profilenm = get_profile_name(sb);
1075         long ccc_count;
1076         int next, force = 1, rc = 0;
1077         ENTRY;
1078
1079         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1080
1081         cfg.cfg_instance = sb;
1082         lustre_end_log(sb, profilenm, &cfg);
1083
1084         params_cfg.cfg_instance = sb;
1085         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1086
1087         if (sbi->ll_md_exp) {
1088                 obd = class_exp2obd(sbi->ll_md_exp);
1089                 if (obd)
1090                         force = obd->obd_force;
1091         }
1092
1093         /* Wait for unstable pages to be committed to stable storage */
1094         if (force == 0) {
1095                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1096                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1097                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1098                         &lwi);
1099         }
1100
1101         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1102         if (force == 0 && rc != -EINTR)
1103                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1104
1105
1106         /* We need to set force before the lov_disconnect in
1107            lustre_common_put_super, since l_d cleans up osc's as well. */
1108         if (force) {
1109                 next = 0;
1110                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1111                                                      &next)) != NULL) {
1112                         obd->obd_force = force;
1113                 }
1114         }
1115
1116         if (sbi->ll_client_common_fill_super_succeeded) {
1117                 /* Only if client_common_fill_super succeeded */
1118                 client_common_put_super(sb);
1119         }
1120
1121         next = 0;
1122         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1123                 class_manual_cleanup(obd);
1124         }
1125
1126         if (sbi->ll_flags & LL_SBI_VERBOSE)
1127                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1128
1129         if (profilenm)
1130                 class_del_profile(profilenm);
1131
1132         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1133                 bdi_destroy(&lsi->lsi_bdi);
1134                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1135         }
1136
1137         ll_free_sbi(sb);
1138         lsi->lsi_llsbi = NULL;
1139
1140         lustre_common_put_super(sb);
1141
1142         cl_env_cache_purge(~0);
1143
1144         module_put(THIS_MODULE);
1145
1146         EXIT;
1147 } /* client_put_super */
1148
1149 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1150 {
1151         struct inode *inode = NULL;
1152
1153         /* NOTE: we depend on atomic igrab() -bzzz */
1154         lock_res_and_lock(lock);
1155         if (lock->l_resource->lr_lvb_inode) {
1156                 struct ll_inode_info * lli;
1157                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1158                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1159                         inode = igrab(lock->l_resource->lr_lvb_inode);
1160                 } else {
1161                         inode = lock->l_resource->lr_lvb_inode;
1162                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1163                                          D_WARNING, lock, "lr_lvb_inode %p is "
1164                                          "bogus: magic %08x",
1165                                          lock->l_resource->lr_lvb_inode,
1166                                          lli->lli_inode_magic);
1167                         inode = NULL;
1168                 }
1169         }
1170         unlock_res_and_lock(lock);
1171         return inode;
1172 }
1173
1174 void ll_dir_clear_lsm_md(struct inode *inode)
1175 {
1176         struct ll_inode_info *lli = ll_i2info(inode);
1177
1178         LASSERT(S_ISDIR(inode->i_mode));
1179
1180         if (lli->lli_lsm_md != NULL) {
1181                 lmv_free_memmd(lli->lli_lsm_md);
1182                 lli->lli_lsm_md = NULL;
1183         }
1184 }
1185
1186 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1187                                       const struct lu_fid *fid,
1188                                       struct lustre_md *md)
1189 {
1190         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1191         struct mdt_body         *body = md->body;
1192         struct inode            *inode;
1193         ino_t                   ino;
1194         ENTRY;
1195
1196         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1197         inode = iget_locked(sb, ino);
1198         if (inode == NULL) {
1199                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1200                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1201                 RETURN(ERR_PTR(-ENOENT));
1202         }
1203
1204         if (inode->i_state & I_NEW) {
1205                 struct ll_inode_info *lli = ll_i2info(inode);
1206                 struct lmv_stripe_md *lsm = md->lmv;
1207
1208                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1209                                 (body->mbo_mode & S_IFMT);
1210                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1211                          PFID(fid));
1212
1213                 LTIME_S(inode->i_mtime) = 0;
1214                 LTIME_S(inode->i_atime) = 0;
1215                 LTIME_S(inode->i_ctime) = 0;
1216                 inode->i_rdev = 0;
1217
1218 #ifdef HAVE_BACKING_DEV_INFO
1219                 /* initializing backing dev info. */
1220                 inode->i_mapping->backing_dev_info =
1221                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1222 #endif
1223                 inode->i_op = &ll_dir_inode_operations;
1224                 inode->i_fop = &ll_dir_operations;
1225                 lli->lli_fid = *fid;
1226                 ll_lli_init(lli);
1227
1228                 LASSERT(lsm != NULL);
1229                 /* master object FID */
1230                 lli->lli_pfid = body->mbo_fid1;
1231                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1232                        lli, PFID(fid), PFID(&lli->lli_pfid));
1233                 unlock_new_inode(inode);
1234         }
1235
1236         RETURN(inode);
1237 }
1238
1239 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1240 {
1241         struct lu_fid *fid;
1242         struct lmv_stripe_md *lsm = md->lmv;
1243         int i;
1244
1245         LASSERT(lsm != NULL);
1246         /* XXX sigh, this lsm_root initialization should be in
1247          * LMV layer, but it needs ll_iget right now, so we
1248          * put this here right now. */
1249         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1250                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1251                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1252                 /* Unfortunately ll_iget will call ll_update_inode,
1253                  * where the initialization of slave inode is slightly
1254                  * different, so it reset lsm_md to NULL to avoid
1255                  * initializing lsm for slave inode. */
1256                 /* For migrating inode, master stripe and master object will
1257                  * be same, so we only need assign this inode */
1258                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1259                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1260                 else
1261                         lsm->lsm_md_oinfo[i].lmo_root =
1262                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1263
1264                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1265                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1266
1267                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1268                         return rc;
1269                 }
1270         }
1271
1272         return 0;
1273 }
1274
1275 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1276                                 const struct lmv_stripe_md *lsm_md2)
1277 {
1278         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1279                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1280                lsm_md1->lsm_md_master_mdt_index ==
1281                                         lsm_md2->lsm_md_master_mdt_index &&
1282                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1283                lsm_md1->lsm_md_layout_version ==
1284                                         lsm_md2->lsm_md_layout_version &&
1285                strcmp(lsm_md1->lsm_md_pool_name,
1286                       lsm_md2->lsm_md_pool_name) == 0;
1287 }
1288
1289 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1290 {
1291         struct ll_inode_info *lli = ll_i2info(inode);
1292         struct lmv_stripe_md *lsm = md->lmv;
1293         int     rc;
1294         ENTRY;
1295
1296         LASSERT(S_ISDIR(inode->i_mode));
1297         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1298                PFID(ll_inode2fid(inode)));
1299
1300         /* no striped information from request. */
1301         if (lsm == NULL) {
1302                 if (lli->lli_lsm_md == NULL) {
1303                         RETURN(0);
1304                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1305                                                 LMV_HASH_FLAG_MIGRATION) {
1306                         /* migration is done, the temporay MIGRATE layout has
1307                          * been removed */
1308                         CDEBUG(D_INODE, DFID" finish migration.\n",
1309                                PFID(ll_inode2fid(inode)));
1310                         lmv_free_memmd(lli->lli_lsm_md);
1311                         lli->lli_lsm_md = NULL;
1312                         RETURN(0);
1313                 } else {
1314                         /* The lustre_md from req does not include stripeEA,
1315                          * see ll_md_setattr */
1316                         RETURN(0);
1317                 }
1318         }
1319
1320         /* set the directory layout */
1321         if (lli->lli_lsm_md == NULL) {
1322                 struct cl_attr  *attr;
1323
1324                 rc = ll_init_lsm_md(inode, md);
1325                 if (rc != 0)
1326                         RETURN(rc);
1327
1328                 /* set md->lmv to NULL, so the following free lustre_md
1329                  * will not free this lsm */
1330                 md->lmv = NULL;
1331                 lli->lli_lsm_md = lsm;
1332
1333                 OBD_ALLOC_PTR(attr);
1334                 if (attr == NULL)
1335                         RETURN(-ENOMEM);
1336
1337                 /* validate the lsm */
1338                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1339                                    ll_md_blocking_ast);
1340                 if (rc != 0) {
1341                         OBD_FREE_PTR(attr);
1342                         RETURN(rc);
1343                 }
1344
1345                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1346                         md->body->mbo_nlink = attr->cat_nlink;
1347                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1348                         md->body->mbo_size = attr->cat_size;
1349                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1350                         md->body->mbo_atime = attr->cat_atime;
1351                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1352                         md->body->mbo_ctime = attr->cat_ctime;
1353                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1354                         md->body->mbo_mtime = attr->cat_mtime;
1355
1356                 OBD_FREE_PTR(attr);
1357
1358                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1359                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1360                 RETURN(0);
1361         }
1362
1363         /* Compare the old and new stripe information */
1364         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1365                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1366                 int                     idx;
1367
1368                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1369                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1370                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1371                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1372                        inode, lsm, old_lsm,
1373                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1374                        lsm->lsm_md_stripe_count,
1375                        old_lsm->lsm_md_stripe_count,
1376                        lsm->lsm_md_master_mdt_index,
1377                        old_lsm->lsm_md_master_mdt_index,
1378                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1379                        lsm->lsm_md_layout_version,
1380                        old_lsm->lsm_md_layout_version,
1381                        lsm->lsm_md_pool_name,
1382                        old_lsm->lsm_md_pool_name);
1383
1384                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1385                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1386                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1387                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1388                 }
1389
1390                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1391                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1392                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1393                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1394                 }
1395
1396                 RETURN(-EIO);
1397         }
1398
1399         RETURN(0);
1400 }
1401
1402 void ll_clear_inode(struct inode *inode)
1403 {
1404         struct ll_inode_info *lli = ll_i2info(inode);
1405         struct ll_sb_info *sbi = ll_i2sbi(inode);
1406         ENTRY;
1407
1408         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1409                PFID(ll_inode2fid(inode)), inode);
1410
1411         if (S_ISDIR(inode->i_mode)) {
1412                 /* these should have been cleared in ll_file_release */
1413                 LASSERT(lli->lli_opendir_key == NULL);
1414                 LASSERT(lli->lli_sai == NULL);
1415                 LASSERT(lli->lli_opendir_pid == 0);
1416         }
1417
1418         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1419
1420         LASSERT(!lli->lli_open_fd_write_count);
1421         LASSERT(!lli->lli_open_fd_read_count);
1422         LASSERT(!lli->lli_open_fd_exec_count);
1423
1424         if (lli->lli_mds_write_och)
1425                 ll_md_real_close(inode, FMODE_WRITE);
1426         if (lli->lli_mds_exec_och)
1427                 ll_md_real_close(inode, FMODE_EXEC);
1428         if (lli->lli_mds_read_och)
1429                 ll_md_real_close(inode, FMODE_READ);
1430
1431         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1432                 OBD_FREE(lli->lli_symlink_name,
1433                          strlen(lli->lli_symlink_name) + 1);
1434                 lli->lli_symlink_name = NULL;
1435         }
1436
1437         ll_xattr_cache_destroy(inode);
1438
1439 #ifdef CONFIG_FS_POSIX_ACL
1440         if (lli->lli_posix_acl) {
1441                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1442                 posix_acl_release(lli->lli_posix_acl);
1443                 lli->lli_posix_acl = NULL;
1444         }
1445 #endif
1446         lli->lli_inode_magic = LLI_INODE_DEAD;
1447
1448         if (S_ISDIR(inode->i_mode))
1449                 ll_dir_clear_lsm_md(inode);
1450         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1451                 LASSERT(list_empty(&lli->lli_agl_list));
1452
1453         /*
1454          * XXX This has to be done before lsm is freed below, because
1455          * cl_object still uses inode lsm.
1456          */
1457         cl_inode_fini(inode);
1458
1459         EXIT;
1460 }
1461
1462 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1463 {
1464         struct lustre_md md;
1465         struct inode *inode = dentry->d_inode;
1466         struct ll_sb_info *sbi = ll_i2sbi(inode);
1467         struct ptlrpc_request *request = NULL;
1468         int rc, ia_valid;
1469         ENTRY;
1470
1471         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1472                                      LUSTRE_OPC_ANY, NULL);
1473         if (IS_ERR(op_data))
1474                 RETURN(PTR_ERR(op_data));
1475
1476         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1477         if (rc) {
1478                 ptlrpc_req_finished(request);
1479                 if (rc == -ENOENT) {
1480                         clear_nlink(inode);
1481                         /* Unlinked special device node? Or just a race?
1482                          * Pretend we done everything. */
1483                         if (!S_ISREG(inode->i_mode) &&
1484                             !S_ISDIR(inode->i_mode)) {
1485                                 ia_valid = op_data->op_attr.ia_valid;
1486                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1487                                 rc = simple_setattr(dentry, &op_data->op_attr);
1488                                 op_data->op_attr.ia_valid = ia_valid;
1489                         }
1490                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1491                         CERROR("md_setattr fails: rc = %d\n", rc);
1492                 }
1493                 RETURN(rc);
1494         }
1495
1496         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1497                               sbi->ll_md_exp, &md);
1498         if (rc) {
1499                 ptlrpc_req_finished(request);
1500                 RETURN(rc);
1501         }
1502
1503         ia_valid = op_data->op_attr.ia_valid;
1504         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1505          * cache is not cleared yet. */
1506         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1507         if (S_ISREG(inode->i_mode))
1508                 inode_lock(inode);
1509         rc = simple_setattr(dentry, &op_data->op_attr);
1510         if (S_ISREG(inode->i_mode))
1511                 inode_unlock(inode);
1512         op_data->op_attr.ia_valid = ia_valid;
1513
1514         rc = ll_update_inode(inode, &md);
1515         ptlrpc_req_finished(request);
1516
1517         RETURN(rc);
1518 }
1519
1520 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1521  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1522  * keep these values until such a time that objects are allocated for it.
1523  * We do the MDS operations first, as it is checking permissions for us.
1524  * We don't to the MDS RPC if there is nothing that we want to store there,
1525  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1526  * going to do an RPC anyways.
1527  *
1528  * If we are doing a truncate, we will send the mtime and ctime updates
1529  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1530  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1531  * at the same time.
1532  *
1533  * In case of HSMimport, we only set attr on MDS.
1534  */
1535 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1536 {
1537         struct inode *inode = dentry->d_inode;
1538         struct ll_inode_info *lli = ll_i2info(inode);
1539         struct md_op_data *op_data = NULL;
1540         int rc = 0;
1541         ENTRY;
1542
1543         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1544                "valid %x, hsm_import %d\n",
1545                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1546                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1547                hsm_import);
1548
1549         if (attr->ia_valid & ATTR_SIZE) {
1550                 /* Check new size against VFS/VM file size limit and rlimit */
1551                 rc = inode_newsize_ok(inode, attr->ia_size);
1552                 if (rc)
1553                         RETURN(rc);
1554
1555                 /* The maximum Lustre file size is variable, based on the
1556                  * OST maximum object size and number of stripes.  This
1557                  * needs another check in addition to the VFS check above. */
1558                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1559                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1560                                PFID(&lli->lli_fid), attr->ia_size,
1561                                ll_file_maxbytes(inode));
1562                         RETURN(-EFBIG);
1563                 }
1564
1565                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1566         }
1567
1568         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1569         if (attr->ia_valid & TIMES_SET_FLAGS) {
1570                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1571                     !cfs_capable(CFS_CAP_FOWNER))
1572                         RETURN(-EPERM);
1573         }
1574
1575         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1576         if (!(attr->ia_valid & ATTR_CTIME_SET) &&
1577             (attr->ia_valid & ATTR_CTIME)) {
1578                 attr->ia_ctime = CURRENT_TIME;
1579                 attr->ia_valid |= ATTR_CTIME_SET;
1580         }
1581         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1582             (attr->ia_valid & ATTR_ATIME)) {
1583                 attr->ia_atime = CURRENT_TIME;
1584                 attr->ia_valid |= ATTR_ATIME_SET;
1585         }
1586         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1587             (attr->ia_valid & ATTR_MTIME)) {
1588                 attr->ia_mtime = CURRENT_TIME;
1589                 attr->ia_valid |= ATTR_MTIME_SET;
1590         }
1591
1592         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1593                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1594                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1595                        cfs_time_current_sec());
1596
1597         if (S_ISREG(inode->i_mode)) {
1598                 if (attr->ia_valid & ATTR_SIZE)
1599                         inode_dio_write_done(inode);
1600                 inode_unlock(inode);
1601         }
1602
1603         /* We always do an MDS RPC, even if we're only changing the size;
1604          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1605
1606         OBD_ALLOC_PTR(op_data);
1607         if (op_data == NULL)
1608                 GOTO(out, rc = -ENOMEM);
1609
1610         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1611                 /* If we are changing file size, file content is
1612                  * modified, flag it. */
1613                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1614                 op_data->op_bias |= MDS_DATA_MODIFIED;
1615                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1616         }
1617
1618         op_data->op_attr = *attr;
1619
1620         rc = ll_md_setattr(dentry, op_data);
1621         if (rc)
1622                 GOTO(out, rc);
1623
1624         if (!S_ISREG(inode->i_mode) || hsm_import)
1625                 GOTO(out, rc = 0);
1626
1627         if (attr->ia_valid & (ATTR_SIZE |
1628                               ATTR_ATIME | ATTR_ATIME_SET |
1629                               ATTR_MTIME | ATTR_MTIME_SET |
1630                               ATTR_CTIME | ATTR_CTIME_SET)) {
1631                 /* For truncate and utimes sending attributes to OSTs, setting
1632                  * mtime/atime to the past will be performed under PW [0:EOF]
1633                  * extent lock (new_size:EOF for truncate).  It may seem
1634                  * excessive to send mtime/atime updates to OSTs when not
1635                  * setting times to past, but it is necessary due to possible
1636                  * time de-synchronization between MDT inode and OST objects */
1637                 rc = cl_setattr_ost(lli->lli_clob, attr, 0);
1638         }
1639
1640         /* If the file was restored, it needs to set dirty flag.
1641          *
1642          * We've already sent MDS_DATA_MODIFIED flag in
1643          * ll_md_setattr() for truncate. However, the MDT refuses to
1644          * set the HS_DIRTY flag on released files, so we have to set
1645          * it again if the file has been restored. Please check how
1646          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1647          *
1648          * Please notice that if the file is not released, the previous
1649          * MDS_DATA_MODIFIED has taken effect and usually
1650          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1651          * This way we can save an RPC for common open + trunc
1652          * operation. */
1653         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1654                 struct hsm_state_set hss = {
1655                         .hss_valid = HSS_SETMASK,
1656                         .hss_setmask = HS_DIRTY,
1657                 };
1658                 int rc2;
1659
1660                 rc2 = ll_hsm_state_set(inode, &hss);
1661                 /* truncate and write can happen at the same time, so that
1662                  * the file can be set modified even though the file is not
1663                  * restored from released state, and ll_hsm_state_set() is
1664                  * not applicable for the file, and rc2 < 0 is normal in this
1665                  * case. */
1666                 if (rc2 < 0)
1667                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1668                                PFID(ll_inode2fid(inode)), rc2);
1669         }
1670
1671         EXIT;
1672 out:
1673         if (op_data != NULL)
1674                 ll_finish_md_op_data(op_data);
1675
1676         if (S_ISREG(inode->i_mode)) {
1677                 inode_lock(inode);
1678                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1679                         inode_dio_wait(inode);
1680         }
1681
1682         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1683                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1684
1685         return rc;
1686 }
1687
1688 int ll_setattr(struct dentry *de, struct iattr *attr)
1689 {
1690         int mode = de->d_inode->i_mode;
1691
1692         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1693                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1694                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1695
1696         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1697                                (ATTR_SIZE|ATTR_MODE)) &&
1698             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1699              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1700               !(attr->ia_mode & S_ISGID))))
1701                 attr->ia_valid |= ATTR_FORCE;
1702
1703         if ((attr->ia_valid & ATTR_MODE) &&
1704             (mode & S_ISUID) &&
1705             !(attr->ia_mode & S_ISUID) &&
1706             !(attr->ia_valid & ATTR_KILL_SUID))
1707                 attr->ia_valid |= ATTR_KILL_SUID;
1708
1709         if ((attr->ia_valid & ATTR_MODE) &&
1710             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1711             !(attr->ia_mode & S_ISGID) &&
1712             !(attr->ia_valid & ATTR_KILL_SGID))
1713                 attr->ia_valid |= ATTR_KILL_SGID;
1714
1715         return ll_setattr_raw(de, attr, false);
1716 }
1717
1718 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1719                        __u64 max_age, __u32 flags)
1720 {
1721         struct ll_sb_info *sbi = ll_s2sbi(sb);
1722         struct obd_statfs obd_osfs;
1723         int rc;
1724         ENTRY;
1725
1726         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1727         if (rc) {
1728                 CERROR("md_statfs fails: rc = %d\n", rc);
1729                 RETURN(rc);
1730         }
1731
1732         osfs->os_type = sb->s_magic;
1733
1734         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1735                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1736
1737         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1738                 flags |= OBD_STATFS_NODELAY;
1739
1740         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1741         if (rc) {
1742                 CERROR("obd_statfs fails: rc = %d\n", rc);
1743                 RETURN(rc);
1744         }
1745
1746         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1747                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1748                obd_osfs.os_files);
1749
1750         osfs->os_bsize = obd_osfs.os_bsize;
1751         osfs->os_blocks = obd_osfs.os_blocks;
1752         osfs->os_bfree = obd_osfs.os_bfree;
1753         osfs->os_bavail = obd_osfs.os_bavail;
1754
1755         /* If we don't have as many objects free on the OST as inodes
1756          * on the MDS, we reduce the total number of inodes to
1757          * compensate, so that the "inodes in use" number is correct.
1758          */
1759         if (obd_osfs.os_ffree < osfs->os_ffree) {
1760                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1761                         obd_osfs.os_ffree;
1762                 osfs->os_ffree = obd_osfs.os_ffree;
1763         }
1764
1765         RETURN(rc);
1766 }
1767 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1768 {
1769         struct super_block *sb = de->d_sb;
1770         struct obd_statfs osfs;
1771         __u64 fsid = huge_encode_dev(sb->s_dev);
1772         int rc;
1773
1774         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1775         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1776
1777         /* Some amount of caching on the client is allowed */
1778         rc = ll_statfs_internal(sb, &osfs,
1779                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1780                                 0);
1781         if (rc)
1782                 return rc;
1783
1784         statfs_unpack(sfs, &osfs);
1785
1786         /* We need to downshift for all 32-bit kernels, because we can't
1787          * tell if the kernel is being called via sys_statfs64() or not.
1788          * Stop before overflowing f_bsize - in which case it is better
1789          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1790         if (sizeof(long) < 8) {
1791                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1792                         sfs->f_bsize <<= 1;
1793
1794                         osfs.os_blocks >>= 1;
1795                         osfs.os_bfree >>= 1;
1796                         osfs.os_bavail >>= 1;
1797                 }
1798         }
1799
1800         sfs->f_blocks = osfs.os_blocks;
1801         sfs->f_bfree = osfs.os_bfree;
1802         sfs->f_bavail = osfs.os_bavail;
1803         sfs->f_fsid.val[0] = (__u32)fsid;
1804         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1805         return 0;
1806 }
1807
1808 void ll_inode_size_lock(struct inode *inode)
1809 {
1810         struct ll_inode_info *lli;
1811
1812         LASSERT(!S_ISDIR(inode->i_mode));
1813
1814         lli = ll_i2info(inode);
1815         mutex_lock(&lli->lli_size_mutex);
1816 }
1817
1818 void ll_inode_size_unlock(struct inode *inode)
1819 {
1820         struct ll_inode_info *lli;
1821
1822         lli = ll_i2info(inode);
1823         mutex_unlock(&lli->lli_size_mutex);
1824 }
1825
1826 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1827 {
1828         struct ll_inode_info *lli = ll_i2info(inode);
1829         struct mdt_body *body = md->body;
1830         struct ll_sb_info *sbi = ll_i2sbi(inode);
1831
1832         if (body->mbo_valid & OBD_MD_FLEASIZE)
1833                 cl_file_inode_init(inode, md);
1834
1835         if (S_ISDIR(inode->i_mode)) {
1836                 int     rc;
1837
1838                 rc = ll_update_lsm_md(inode, md);
1839                 if (rc != 0)
1840                         return rc;
1841         }
1842
1843 #ifdef CONFIG_FS_POSIX_ACL
1844         if (body->mbo_valid & OBD_MD_FLACL) {
1845                 spin_lock(&lli->lli_lock);
1846                 if (lli->lli_posix_acl)
1847                         posix_acl_release(lli->lli_posix_acl);
1848                 lli->lli_posix_acl = md->posix_acl;
1849                 spin_unlock(&lli->lli_lock);
1850         }
1851 #endif
1852         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1853                                         sbi->ll_flags & LL_SBI_32BIT_API);
1854         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1855
1856         if (body->mbo_valid & OBD_MD_FLATIME) {
1857                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1858                         LTIME_S(inode->i_atime) = body->mbo_atime;
1859                 lli->lli_atime = body->mbo_atime;
1860         }
1861
1862         if (body->mbo_valid & OBD_MD_FLMTIME) {
1863                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1864                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1865                                "to "LPU64"\n", inode->i_ino,
1866                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1867                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1868                 }
1869                 lli->lli_mtime = body->mbo_mtime;
1870         }
1871
1872         if (body->mbo_valid & OBD_MD_FLCTIME) {
1873                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1874                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1875                 lli->lli_ctime = body->mbo_ctime;
1876         }
1877
1878         if (body->mbo_valid & OBD_MD_FLMODE)
1879                 inode->i_mode = (inode->i_mode & S_IFMT) |
1880                                 (body->mbo_mode & ~S_IFMT);
1881
1882         if (body->mbo_valid & OBD_MD_FLTYPE)
1883                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1884                                 (body->mbo_mode & S_IFMT);
1885
1886         LASSERT(inode->i_mode != 0);
1887         if (S_ISREG(inode->i_mode))
1888                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1889                                        LL_MAX_BLKSIZE_BITS);
1890         else
1891                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1892
1893         if (body->mbo_valid & OBD_MD_FLUID)
1894                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1895         if (body->mbo_valid & OBD_MD_FLGID)
1896                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1897         if (body->mbo_valid & OBD_MD_FLFLAGS)
1898                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1899         if (body->mbo_valid & OBD_MD_FLNLINK)
1900                 set_nlink(inode, body->mbo_nlink);
1901         if (body->mbo_valid & OBD_MD_FLRDEV)
1902                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1903
1904         if (body->mbo_valid & OBD_MD_FLID) {
1905                 /* FID shouldn't be changed! */
1906                 if (fid_is_sane(&lli->lli_fid)) {
1907                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1908                                  "Trying to change FID "DFID
1909                                  " to the "DFID", inode "DFID"(%p)\n",
1910                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1911                                  PFID(ll_inode2fid(inode)), inode);
1912                 } else {
1913                         lli->lli_fid = body->mbo_fid1;
1914                 }
1915         }
1916
1917         LASSERT(fid_seq(&lli->lli_fid) != 0);
1918
1919         if (body->mbo_valid & OBD_MD_FLSIZE) {
1920                 i_size_write(inode, body->mbo_size);
1921
1922                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
1923                        PFID(ll_inode2fid(inode)),
1924                        (unsigned long long)body->mbo_size);
1925
1926                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1927                         inode->i_blocks = body->mbo_blocks;
1928         }
1929
1930         if (body->mbo_valid & OBD_MD_TSTATE) {
1931                 /* Set LLIF_FILE_RESTORING if restore ongoing and
1932                  * clear it when done to ensure to start again
1933                  * glimpsing updated attrs
1934                  */
1935                 if (body->mbo_t_state & MS_RESTORE)
1936                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
1937                 else
1938                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
1939         }
1940
1941         return 0;
1942 }
1943
1944 int ll_read_inode2(struct inode *inode, void *opaque)
1945 {
1946         struct lustre_md *md = opaque;
1947         struct ll_inode_info *lli = ll_i2info(inode);
1948         int     rc;
1949         ENTRY;
1950
1951         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1952                PFID(&lli->lli_fid), inode);
1953
1954         /* Core attributes from the MDS first.  This is a new inode, and
1955          * the VFS doesn't zero times in the core inode so we have to do
1956          * it ourselves.  They will be overwritten by either MDS or OST
1957          * attributes - we just need to make sure they aren't newer. */
1958         LTIME_S(inode->i_mtime) = 0;
1959         LTIME_S(inode->i_atime) = 0;
1960         LTIME_S(inode->i_ctime) = 0;
1961         inode->i_rdev = 0;
1962         rc = ll_update_inode(inode, md);
1963         if (rc != 0)
1964                 RETURN(rc);
1965
1966         /* OIDEBUG(inode); */
1967
1968 #ifdef HAVE_BACKING_DEV_INFO
1969         /* initializing backing dev info. */
1970         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1971 #endif
1972         if (S_ISREG(inode->i_mode)) {
1973                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1974                 inode->i_op = &ll_file_inode_operations;
1975                 inode->i_fop = sbi->ll_fop;
1976                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1977                 EXIT;
1978         } else if (S_ISDIR(inode->i_mode)) {
1979                 inode->i_op = &ll_dir_inode_operations;
1980                 inode->i_fop = &ll_dir_operations;
1981                 EXIT;
1982         } else if (S_ISLNK(inode->i_mode)) {
1983                 inode->i_op = &ll_fast_symlink_inode_operations;
1984                 EXIT;
1985         } else {
1986                 inode->i_op = &ll_special_inode_operations;
1987
1988                 init_special_inode(inode, inode->i_mode,
1989                                    inode->i_rdev);
1990
1991                 EXIT;
1992         }
1993
1994         return 0;
1995 }
1996
1997 void ll_delete_inode(struct inode *inode)
1998 {
1999         struct ll_inode_info *lli = ll_i2info(inode);
2000         ENTRY;
2001
2002         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2003                 /* It is last chance to write out dirty pages,
2004                  * otherwise we may lose data while umount */
2005                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2006
2007         truncate_inode_pages_final(&inode->i_data);
2008
2009         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2010                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2011                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2012
2013 #ifdef HAVE_SBOPS_EVICT_INODE
2014         ll_clear_inode(inode);
2015 #endif
2016         clear_inode(inode);
2017
2018         EXIT;
2019 }
2020
2021 int ll_iocontrol(struct inode *inode, struct file *file,
2022                  unsigned int cmd, unsigned long arg)
2023 {
2024         struct ll_sb_info *sbi = ll_i2sbi(inode);
2025         struct ptlrpc_request *req = NULL;
2026         int rc, flags = 0;
2027         ENTRY;
2028
2029         switch(cmd) {
2030         case FSFILT_IOC_GETFLAGS: {
2031                 struct mdt_body *body;
2032                 struct md_op_data *op_data;
2033
2034                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2035                                              0, 0, LUSTRE_OPC_ANY,
2036                                              NULL);
2037                 if (IS_ERR(op_data))
2038                         RETURN(PTR_ERR(op_data));
2039
2040                 op_data->op_valid = OBD_MD_FLFLAGS;
2041                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2042                 ll_finish_md_op_data(op_data);
2043                 if (rc) {
2044                         CERROR("%s: failure inode "DFID": rc = %d\n",
2045                                sbi->ll_md_exp->exp_obd->obd_name,
2046                                PFID(ll_inode2fid(inode)), rc);
2047                         RETURN(-abs(rc));
2048                 }
2049
2050                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2051
2052                 flags = body->mbo_flags;
2053
2054                 ptlrpc_req_finished(req);
2055
2056                 RETURN(put_user(flags, (int __user *)arg));
2057         }
2058         case FSFILT_IOC_SETFLAGS: {
2059                 struct iattr *attr;
2060                 struct md_op_data *op_data;
2061                 struct cl_object *obj;
2062
2063                 if (get_user(flags, (int __user *)arg))
2064                         RETURN(-EFAULT);
2065
2066                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2067                                              LUSTRE_OPC_ANY, NULL);
2068                 if (IS_ERR(op_data))
2069                         RETURN(PTR_ERR(op_data));
2070
2071                 op_data->op_attr_flags = flags;
2072                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2073                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2074                 ll_finish_md_op_data(op_data);
2075                 ptlrpc_req_finished(req);
2076                 if (rc)
2077                         RETURN(rc);
2078
2079                 inode->i_flags = ll_ext_to_inode_flags(flags);
2080
2081                 obj = ll_i2info(inode)->lli_clob;
2082                 if (obj == NULL)
2083                         RETURN(0);
2084
2085                 OBD_ALLOC_PTR(attr);
2086                 if (attr == NULL)
2087                         RETURN(-ENOMEM);
2088
2089                 attr->ia_valid = ATTR_ATTR_FLAG;
2090                 rc = cl_setattr_ost(obj, attr, flags);
2091
2092                 OBD_FREE_PTR(attr);
2093                 RETURN(rc);
2094         }
2095         default:
2096                 RETURN(-ENOSYS);
2097         }
2098
2099         RETURN(0);
2100 }
2101
2102 int ll_flush_ctx(struct inode *inode)
2103 {
2104         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2105
2106         CDEBUG(D_SEC, "flush context for user %d\n",
2107                from_kuid(&init_user_ns, current_uid()));
2108
2109         obd_set_info_async(NULL, sbi->ll_md_exp,
2110                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2111                            0, NULL, NULL);
2112         obd_set_info_async(NULL, sbi->ll_dt_exp,
2113                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2114                            0, NULL, NULL);
2115         return 0;
2116 }
2117
2118 /* umount -f client means force down, don't save state */
2119 void ll_umount_begin(struct super_block *sb)
2120 {
2121         struct ll_sb_info *sbi = ll_s2sbi(sb);
2122         struct obd_device *obd;
2123         struct obd_ioctl_data *ioc_data;
2124         ENTRY;
2125
2126         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2127                sb->s_count, atomic_read(&sb->s_active));
2128
2129         obd = class_exp2obd(sbi->ll_md_exp);
2130         if (obd == NULL) {
2131                 CERROR("Invalid MDC connection handle "LPX64"\n",
2132                        sbi->ll_md_exp->exp_handle.h_cookie);
2133                 EXIT;
2134                 return;
2135         }
2136         obd->obd_force = 1;
2137
2138         obd = class_exp2obd(sbi->ll_dt_exp);
2139         if (obd == NULL) {
2140                 CERROR("Invalid LOV connection handle "LPX64"\n",
2141                        sbi->ll_dt_exp->exp_handle.h_cookie);
2142                 EXIT;
2143                 return;
2144         }
2145         obd->obd_force = 1;
2146
2147         OBD_ALLOC_PTR(ioc_data);
2148         if (ioc_data) {
2149                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2150                               sizeof *ioc_data, ioc_data, NULL);
2151
2152                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2153                               sizeof *ioc_data, ioc_data, NULL);
2154
2155                 OBD_FREE_PTR(ioc_data);
2156         }
2157
2158         /* Really, we'd like to wait until there are no requests outstanding,
2159          * and then continue.  For now, we just invalidate the requests,
2160          * schedule() and sleep one second if needed, and hope.
2161          */
2162         schedule();
2163         EXIT;
2164 }
2165
2166 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2167 {
2168         struct ll_sb_info *sbi = ll_s2sbi(sb);
2169         char *profilenm = get_profile_name(sb);
2170         int err;
2171         __u32 read_only;
2172
2173         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2174                 read_only = *flags & MS_RDONLY;
2175                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2176                                          sizeof(KEY_READ_ONLY),
2177                                          KEY_READ_ONLY, sizeof(read_only),
2178                                          &read_only, NULL);
2179                 if (err) {
2180                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2181                                       profilenm, read_only ?
2182                                       "read-only" : "read-write", err);
2183                         return err;
2184                 }
2185
2186                 if (read_only)
2187                         sb->s_flags |= MS_RDONLY;
2188                 else
2189                         sb->s_flags &= ~MS_RDONLY;
2190
2191                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2192                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2193                                       read_only ?  "read-only" : "read-write");
2194         }
2195         return 0;
2196 }
2197
2198 /**
2199  * Cleanup the open handle that is cached on MDT-side.
2200  *
2201  * For open case, the client side open handling thread may hit error
2202  * after the MDT grant the open. Under such case, the client should
2203  * send close RPC to the MDT as cleanup; otherwise, the open handle
2204  * on the MDT will be leaked there until the client umount or evicted.
2205  *
2206  * In further, if someone unlinked the file, because the open handle
2207  * holds the reference on such file/object, then it will block the
2208  * subsequent threads that want to locate such object via FID.
2209  *
2210  * \param[in] sb        super block for this file-system
2211  * \param[in] open_req  pointer to the original open request
2212  */
2213 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2214 {
2215         struct mdt_body                 *body;
2216         struct md_op_data               *op_data;
2217         struct ptlrpc_request           *close_req = NULL;
2218         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2219         ENTRY;
2220
2221         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2222         OBD_ALLOC_PTR(op_data);
2223         if (op_data == NULL) {
2224                 CWARN("%s: cannot allocate op_data to release open handle for "
2225                       DFID"\n",
2226                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2227
2228                 RETURN_EXIT;
2229         }
2230
2231         op_data->op_fid1 = body->mbo_fid1;
2232         op_data->op_handle = body->mbo_handle;
2233         op_data->op_mod_time = cfs_time_current_sec();
2234         md_close(exp, op_data, NULL, &close_req);
2235         ptlrpc_req_finished(close_req);
2236         ll_finish_md_op_data(op_data);
2237
2238         EXIT;
2239 }
2240
2241 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2242                   struct super_block *sb, struct lookup_intent *it)
2243 {
2244         struct ll_sb_info *sbi = NULL;
2245         struct lustre_md md = { NULL };
2246         int rc;
2247         ENTRY;
2248
2249         LASSERT(*inode || sb);
2250         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2251         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2252                               sbi->ll_md_exp, &md);
2253         if (rc != 0)
2254                 GOTO(cleanup, rc);
2255
2256         if (*inode) {
2257                 rc = ll_update_inode(*inode, &md);
2258                 if (rc != 0)
2259                         GOTO(out, rc);
2260         } else {
2261                 LASSERT(sb != NULL);
2262
2263                 /*
2264                  * At this point server returns to client's same fid as client
2265                  * generated for creating. So using ->fid1 is okay here.
2266                  */
2267                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2268                         CERROR("%s: Fid is insane "DFID"\n",
2269                                 ll_get_fsname(sb, NULL, 0),
2270                                 PFID(&md.body->mbo_fid1));
2271                         GOTO(out, rc = -EINVAL);
2272                 }
2273
2274                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2275                                              sbi->ll_flags & LL_SBI_32BIT_API),
2276                                  &md);
2277                 if (IS_ERR(*inode)) {
2278 #ifdef CONFIG_FS_POSIX_ACL
2279                         if (md.posix_acl) {
2280                                 posix_acl_release(md.posix_acl);
2281                                 md.posix_acl = NULL;
2282                         }
2283 #endif
2284                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2285                         *inode = NULL;
2286                         CERROR("new_inode -fatal: rc %d\n", rc);
2287                         GOTO(out, rc);
2288                 }
2289         }
2290
2291         /* Handling piggyback layout lock.
2292          * Layout lock can be piggybacked by getattr and open request.
2293          * The lsm can be applied to inode only if it comes with a layout lock
2294          * otherwise correct layout may be overwritten, for example:
2295          * 1. proc1: mdt returns a lsm but not granting layout
2296          * 2. layout was changed by another client
2297          * 3. proc2: refresh layout and layout lock granted
2298          * 4. proc1: to apply a stale layout */
2299         if (it != NULL && it->it_lock_mode != 0) {
2300                 struct lustre_handle lockh;
2301                 struct ldlm_lock *lock;
2302
2303                 lockh.cookie = it->it_lock_handle;
2304                 lock = ldlm_handle2lock(&lockh);
2305                 LASSERT(lock != NULL);
2306                 if (ldlm_has_layout(lock)) {
2307                         struct cl_object_conf conf;
2308
2309                         memset(&conf, 0, sizeof(conf));
2310                         conf.coc_opc = OBJECT_CONF_SET;
2311                         conf.coc_inode = *inode;
2312                         conf.coc_lock = lock;
2313                         conf.u.coc_layout = md.layout;
2314                         (void)ll_layout_conf(*inode, &conf);
2315                 }
2316                 LDLM_LOCK_PUT(lock);
2317         }
2318
2319         GOTO(out, rc = 0);
2320
2321 out:
2322         md_free_lustre_md(sbi->ll_md_exp, &md);
2323
2324 cleanup:
2325         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2326                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2327
2328         return rc;
2329 }
2330
2331 int ll_obd_statfs(struct inode *inode, void __user *arg)
2332 {
2333         struct ll_sb_info *sbi = NULL;
2334         struct obd_export *exp;
2335         char *buf = NULL;
2336         struct obd_ioctl_data *data = NULL;
2337         __u32 type;
2338         int len = 0, rc;
2339
2340         if (!inode || !(sbi = ll_i2sbi(inode)))
2341                 GOTO(out_statfs, rc = -EINVAL);
2342
2343         rc = obd_ioctl_getdata(&buf, &len, arg);
2344         if (rc)
2345                 GOTO(out_statfs, rc);
2346
2347         data = (void*)buf;
2348         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2349             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2350                 GOTO(out_statfs, rc = -EINVAL);
2351
2352         if (data->ioc_inllen1 != sizeof(__u32) ||
2353             data->ioc_inllen2 != sizeof(__u32) ||
2354             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2355             data->ioc_plen2 != sizeof(struct obd_uuid))
2356                 GOTO(out_statfs, rc = -EINVAL);
2357
2358         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2359         if (type & LL_STATFS_LMV)
2360                 exp = sbi->ll_md_exp;
2361         else if (type & LL_STATFS_LOV)
2362                 exp = sbi->ll_dt_exp;
2363         else
2364                 GOTO(out_statfs, rc = -ENODEV);
2365
2366         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2367         if (rc)
2368                 GOTO(out_statfs, rc);
2369 out_statfs:
2370         if (buf)
2371                 obd_ioctl_freedata(buf, len);
2372         return rc;
2373 }
2374
2375 int ll_process_config(struct lustre_cfg *lcfg)
2376 {
2377         struct super_block *sb;
2378         unsigned long x;
2379         int rc = 0;
2380         char *ptr;
2381
2382         /* The instance name contains the sb: lustre-client-aacfe000 */
2383         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2384         if (!ptr || !*(++ptr))
2385                 return -EINVAL;
2386         if (sscanf(ptr, "%lx", &x) != 1)
2387                 return -EINVAL;
2388         sb = (struct super_block *)x;
2389         /* This better be a real Lustre superblock! */
2390         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2391
2392         /* Note we have not called client_common_fill_super yet, so
2393            proc fns must be able to handle that! */
2394         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2395                                       lcfg, sb);
2396         if (rc > 0)
2397                 rc = 0;
2398         return rc;
2399 }
2400
2401 /* this function prepares md_op_data hint for passing it down to MD stack. */
2402 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2403                                       struct inode *i1, struct inode *i2,
2404                                       const char *name, size_t namelen,
2405                                       __u32 mode, __u32 opc, void *data)
2406 {
2407         LASSERT(i1 != NULL);
2408
2409         if (name == NULL) {
2410                 /* Do not reuse namelen for something else. */
2411                 if (namelen != 0)
2412                         return ERR_PTR(-EINVAL);
2413         } else {
2414                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2415                         return ERR_PTR(-ENAMETOOLONG);
2416
2417                 if (!lu_name_is_valid_2(name, namelen))
2418                         return ERR_PTR(-EINVAL);
2419         }
2420
2421         if (op_data == NULL)
2422                 OBD_ALLOC_PTR(op_data);
2423
2424         if (op_data == NULL)
2425                 return ERR_PTR(-ENOMEM);
2426
2427         ll_i2gids(op_data->op_suppgids, i1, i2);
2428         op_data->op_fid1 = *ll_inode2fid(i1);
2429         op_data->op_default_stripe_offset = -1;
2430         if (S_ISDIR(i1->i_mode)) {
2431                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2432                 if (opc == LUSTRE_OPC_MKDIR)
2433                         op_data->op_default_stripe_offset =
2434                                    ll_i2info(i1)->lli_def_stripe_offset;
2435         }
2436
2437         if (i2) {
2438                 op_data->op_fid2 = *ll_inode2fid(i2);
2439                 if (S_ISDIR(i2->i_mode))
2440                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2441         } else {
2442                 fid_zero(&op_data->op_fid2);
2443         }
2444
2445         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2446                 op_data->op_cli_flags |= CLI_HASH64;
2447
2448         if (ll_need_32bit_api(ll_i2sbi(i1)))
2449                 op_data->op_cli_flags |= CLI_API32;
2450
2451         op_data->op_name = name;
2452         op_data->op_namelen = namelen;
2453         op_data->op_mode = mode;
2454         op_data->op_mod_time = cfs_time_current_sec();
2455         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2456         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2457         op_data->op_cap = cfs_curproc_cap_pack();
2458         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2459              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2460                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2461         } else {
2462                 op_data->op_mds = 0;
2463         }
2464         op_data->op_data = data;
2465
2466         return op_data;
2467 }
2468
2469 void ll_finish_md_op_data(struct md_op_data *op_data)
2470 {
2471         OBD_FREE_PTR(op_data);
2472 }
2473
2474 #ifdef HAVE_SUPEROPS_USE_DENTRY
2475 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2476 #else
2477 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2478 #endif
2479 {
2480         struct ll_sb_info *sbi;
2481
2482 #ifdef HAVE_SUPEROPS_USE_DENTRY
2483         LASSERT((seq != NULL) && (dentry != NULL));
2484         sbi = ll_s2sbi(dentry->d_sb);
2485 #else
2486         LASSERT((seq != NULL) && (vfs != NULL));
2487         sbi = ll_s2sbi(vfs->mnt_sb);
2488 #endif
2489
2490         if (sbi->ll_flags & LL_SBI_NOLCK)
2491                 seq_puts(seq, ",nolock");
2492
2493         if (sbi->ll_flags & LL_SBI_FLOCK)
2494                 seq_puts(seq, ",flock");
2495
2496         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2497                 seq_puts(seq, ",localflock");
2498
2499         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2500                 seq_puts(seq, ",user_xattr");
2501
2502         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2503                 seq_puts(seq, ",lazystatfs");
2504
2505         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2506                 seq_puts(seq, ",user_fid2path");
2507
2508         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2509                 seq_puts(seq, ",always_ping");
2510
2511         RETURN(0);
2512 }
2513
2514 /**
2515  * Get obd name by cmd, and copy out to user space
2516  */
2517 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2518 {
2519         struct ll_sb_info *sbi = ll_i2sbi(inode);
2520         struct obd_device *obd;
2521         ENTRY;
2522
2523         if (cmd == OBD_IOC_GETDTNAME)
2524                 obd = class_exp2obd(sbi->ll_dt_exp);
2525         else if (cmd == OBD_IOC_GETMDNAME)
2526                 obd = class_exp2obd(sbi->ll_md_exp);
2527         else
2528                 RETURN(-EINVAL);
2529
2530         if (!obd)
2531                 RETURN(-ENOENT);
2532
2533         if (copy_to_user((void __user *)arg, obd->obd_name,
2534                          strlen(obd->obd_name) + 1))
2535                 RETURN(-EFAULT);
2536
2537         RETURN(0);
2538 }
2539
2540 /**
2541  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2542  * fsname will be returned in this buffer; otherwise, a static buffer will be
2543  * used to store the fsname and returned to caller.
2544  */
2545 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2546 {
2547         static char fsname_static[MTI_NAME_MAXLEN];
2548         struct lustre_sb_info *lsi = s2lsi(sb);
2549         char *ptr;
2550         int len;
2551
2552         if (buf == NULL) {
2553                 /* this means the caller wants to use static buffer
2554                  * and it doesn't care about race. Usually this is
2555                  * in error reporting path */
2556                 buf = fsname_static;
2557                 buflen = sizeof(fsname_static);
2558         }
2559
2560         len = strlen(lsi->lsi_lmd->lmd_profile);
2561         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2562         if (ptr && (strcmp(ptr, "-client") == 0))
2563                 len -= 7;
2564
2565         if (unlikely(len >= buflen))
2566                 len = buflen - 1;
2567         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2568         buf[len] = '\0';
2569
2570         return buf;
2571 }
2572
2573 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2574 {
2575         char *path = NULL;
2576
2577         struct path p;
2578
2579         p.dentry = dentry;
2580         p.mnt = current->fs->root.mnt;
2581         path_get(&p);
2582         path = d_path(&p, buf, bufsize);
2583         path_put(&p);
2584         return path;
2585 }
2586
2587 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2588 {
2589         char *buf, *path = NULL;
2590         struct dentry *dentry = NULL;
2591         struct inode *inode = page->mapping->host;
2592
2593         /* this can be called inside spin lock so use GFP_ATOMIC. */
2594         buf = (char *)__get_free_page(GFP_ATOMIC);
2595         if (buf != NULL) {
2596                 dentry = d_find_alias(page->mapping->host);
2597                 if (dentry != NULL)
2598                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2599         }
2600
2601         CDEBUG(D_WARNING,
2602                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2603                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2604                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2605                PFID(ll_inode2fid(inode)),
2606                (path && !IS_ERR(path)) ? path : "", ioret);
2607
2608         if (dentry != NULL)
2609                 dput(dentry);
2610
2611         if (buf != NULL)
2612                 free_page((unsigned long)buf);
2613 }
2614
2615 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2616                         struct lov_user_md **kbuf)
2617 {
2618         struct lov_user_md      lum;
2619         ssize_t                 lum_size;
2620         ENTRY;
2621
2622         if (copy_from_user(&lum, md, sizeof(lum)))
2623                 RETURN(-EFAULT);
2624
2625         lum_size = ll_lov_user_md_size(&lum);
2626         if (lum_size < 0)
2627                 RETURN(lum_size);
2628
2629         OBD_ALLOC(*kbuf, lum_size);
2630         if (*kbuf == NULL)
2631                 RETURN(-ENOMEM);
2632
2633         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2634                 OBD_FREE(*kbuf, lum_size);
2635                 RETURN(-EFAULT);
2636         }
2637
2638         RETURN(lum_size);
2639 }
2640
2641 /*
2642  * Compute llite root squash state after a change of root squash
2643  * configuration setting or add/remove of a lnet nid
2644  */
2645 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2646 {
2647         struct root_squash_info *squash = &sbi->ll_squash;
2648         int i;
2649         bool matched;
2650         lnet_process_id_t id;
2651
2652         /* Update norootsquash flag */
2653         down_write(&squash->rsi_sem);
2654         if (list_empty(&squash->rsi_nosquash_nids))
2655                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2656         else {
2657                 /* Do not apply root squash as soon as one of our NIDs is
2658                  * in the nosquash_nids list */
2659                 matched = false;
2660                 i = 0;
2661                 while (LNetGetId(i++, &id) != -ENOENT) {
2662                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2663                                 continue;
2664                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2665                                 matched = true;
2666                                 break;
2667                         }
2668                 }
2669                 if (matched)
2670                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2671                 else
2672                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2673         }
2674         up_write(&squash->rsi_sem);
2675 }
2676
2677 /**
2678  * Parse linkea content to extract information about a given hardlink
2679  *
2680  * \param[in]   ldata      - Initialized linkea data
2681  * \param[in]   linkno     - Link identifier
2682  * \param[out]  parent_fid - The entry's parent FID
2683  * \param[out]  ln         - Entry name destination buffer
2684  *
2685  * \retval 0 on success
2686  * \retval Appropriate negative error code on failure
2687  */
2688 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2689                             struct lu_fid *parent_fid, struct lu_name *ln)
2690 {
2691         unsigned int    idx;
2692         int             rc;
2693         ENTRY;
2694
2695         rc = linkea_init(ldata);
2696         if (rc < 0)
2697                 RETURN(rc);
2698
2699         if (linkno >= ldata->ld_leh->leh_reccount)
2700                 /* beyond last link */
2701                 RETURN(-ENODATA);
2702
2703         linkea_first_entry(ldata);
2704         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2705                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2706                                     parent_fid);
2707                 if (idx == linkno)
2708                         break;
2709
2710                 linkea_next_entry(ldata);
2711         }
2712
2713         if (idx < linkno)
2714                 RETURN(-ENODATA);
2715
2716         RETURN(0);
2717 }
2718
2719 /**
2720  * Get parent FID and name of an identified link. Operation is performed for
2721  * a given link number, letting the caller iterate over linkno to list one or
2722  * all links of an entry.
2723  *
2724  * \param[in]     file - File descriptor against which to perform the operation
2725  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2726  *                       on and the available size. It is eventually filled with
2727  *                       the requested information or left untouched on error
2728  *
2729  * \retval - 0 on success
2730  * \retval - Appropriate negative error code on failure
2731  */
2732 int ll_getparent(struct file *file, struct getparent __user *arg)
2733 {
2734         struct dentry           *dentry = file_dentry(file);
2735         struct inode            *inode = file_inode(file);
2736         struct linkea_data      *ldata;
2737         struct lu_buf            buf = LU_BUF_NULL;
2738         struct lu_name           ln;
2739         struct lu_fid            parent_fid;
2740         __u32                    linkno;
2741         __u32                    name_size;
2742         int                      rc;
2743
2744         ENTRY;
2745
2746         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2747             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2748                 RETURN(-EPERM);
2749
2750         if (get_user(name_size, &arg->gp_name_size))
2751                 RETURN(-EFAULT);
2752
2753         if (get_user(linkno, &arg->gp_linkno))
2754                 RETURN(-EFAULT);
2755
2756         if (name_size > PATH_MAX)
2757                 RETURN(-EINVAL);
2758
2759         OBD_ALLOC(ldata, sizeof(*ldata));
2760         if (ldata == NULL)
2761                 RETURN(-ENOMEM);
2762
2763         rc = linkea_data_new(ldata, &buf);
2764         if (rc < 0)
2765                 GOTO(ldata_free, rc);
2766
2767         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2768         if (rc < 0)
2769                 GOTO(lb_free, rc);
2770
2771         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2772         if (rc < 0)
2773                 GOTO(lb_free, rc);
2774
2775         if (ln.ln_namelen >= name_size)
2776                 GOTO(lb_free, rc = -EOVERFLOW);
2777
2778         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2779                 GOTO(lb_free, rc = -EFAULT);
2780
2781         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2782                 GOTO(lb_free, rc = -EFAULT);
2783
2784         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2785                 GOTO(lb_free, rc = -EFAULT);
2786
2787 lb_free:
2788         lu_buf_free(&buf);
2789 ldata_free:
2790         OBD_FREE(ldata, sizeof(*ldata));
2791
2792         RETURN(rc);
2793 }