drivers/staging/lustre/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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/statfs.h>
  41 #include <linux/types.h>
  42 #include <linux/mm.h>
  43
  44 #include "../include/lustre/lustre_ioctl.h"
  45 #include "../include/lustre_ha.h"
  46 #include "../include/lustre_dlm.h"
  47 #include "../include/lprocfs_status.h"
  48 #include "../include/lustre_disk.h"
  49 #include "../include/lustre_param.h"
  50 #include "../include/lustre_log.h"
  51 #include "../include/cl_object.h"
  52 #include "../include/obd_cksum.h"
  53 #include "llite_internal.h"
  54
  55 struct kmem_cache *ll_file_data_slab;
  56 struct dentry *llite_root;
  57 struct kset *llite_kset;
  58
  59 #ifndef log2
  60 #define log2(n) ffz(~(n))
  61 #endif
  62
  63 static struct ll_sb_info *ll_init_sbi(struct super_block *sb)
  64 {
  65         struct ll_sb_info *sbi = NULL;
  66         unsigned long pages;
  67         unsigned long lru_page_max;
  68         struct sysinfo si;
  69         class_uuid_t uuid;
  70         int i;
  71
  72         sbi = kzalloc(sizeof(*sbi), GFP_NOFS);
  73         if (!sbi)
  74                 return NULL;
  75
  76         spin_lock_init(&sbi->ll_lock);
  77         mutex_init(&sbi->ll_lco.lco_lock);
  78         spin_lock_init(&sbi->ll_pp_extent_lock);
  79         spin_lock_init(&sbi->ll_process_lock);
  80         sbi->ll_rw_stats_on = 0;
  81
  82         si_meminfo(&si);
  83         pages = si.totalram - si.totalhigh;
  84         lru_page_max = pages / 2;
  85
  86         sbi->ll_cache = cl_cache_init(lru_page_max);
  87         if (!sbi->ll_cache) {
  88                 kfree(sbi);
  89                 return NULL;
  90         }
  91
  92         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
  93                                            SBI_DEFAULT_READAHEAD_MAX);
  94         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
  95         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
  96                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
  97
  98         ll_generate_random_uuid(uuid);
  99         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 100         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 101
 102         sbi->ll_flags |= LL_SBI_VERBOSE;
 103         sbi->ll_flags |= LL_SBI_CHECKSUM;
 104
 105         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 106
 107         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 108                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 109                                pp_r_hist.oh_lock);
 110                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 111                                pp_w_hist.oh_lock);
 112         }
 113
 114         /* metadata statahead is enabled by default */
 115         sbi->ll_sa_max = LL_SA_RPC_DEF;
 116         atomic_set(&sbi->ll_sa_total, 0);
 117         atomic_set(&sbi->ll_sa_wrong, 0);
 118         atomic_set(&sbi->ll_sa_running, 0);
 119         atomic_set(&sbi->ll_agl_total, 0);
 120         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 121
 122         /* root squash */
 123         sbi->ll_squash.rsi_uid = 0;
 124         sbi->ll_squash.rsi_gid = 0;
 125         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 126         init_rwsem(&sbi->ll_squash.rsi_sem);
 127
 128         sbi->ll_sb = sb;
 129
 130         return sbi;
 131 }
 132
 133 static void ll_free_sbi(struct super_block *sb)
 134 {
 135         struct ll_sb_info *sbi = ll_s2sbi(sb);
 136
 137         if (sbi->ll_cache) {
 138                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 139                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 140                 cl_cache_decref(sbi->ll_cache);
 141                 sbi->ll_cache = NULL;
 142         }
 143
 144         kfree(sbi);
 145 }
 146
 147 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 148                                     struct vfsmount *mnt)
 149 {
 150         struct inode *root = NULL;
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         struct obd_device *obd;
 153         struct obd_statfs *osfs = NULL;
 154         struct ptlrpc_request *request = NULL;
 155         struct obd_connect_data *data = NULL;
 156         struct obd_uuid *uuid;
 157         struct md_op_data *op_data;
 158         struct lustre_md lmd;
 159         u64 valid;
 160         int size, err, checksum;
 161
 162         obd = class_name2obd(md);
 163         if (!obd) {
 164                 CERROR("MD %s: not setup or attached\n", md);
 165                 return -EINVAL;
 166         }
 167
 168         data = kzalloc(sizeof(*data), GFP_NOFS);
 169         if (!data)
 170                 return -ENOMEM;
 171
 172         osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
 173         if (!osfs) {
 174                 kfree(data);
 175                 return -ENOMEM;
 176         }
 177
 178         /* indicate the features supported by this client */
 179         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 180                                   OBD_CONNECT_ATTRFID  |
 181                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 182                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 183                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 184                                   OBD_CONNECT_VBR       | OBD_CONNECT_FULL20  |
 185                                   OBD_CONNECT_64BITHASH |
 186                                   OBD_CONNECT_EINPROGRESS |
 187                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 188                                   OBD_CONNECT_LAYOUTLOCK |
 189                                   OBD_CONNECT_PINGLESS |
 190                                   OBD_CONNECT_MAX_EASIZE |
 191                                   OBD_CONNECT_FLOCK_DEAD |
 192                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 193                                   OBD_CONNECT_OPEN_BY_FID |
 194                                   OBD_CONNECT_DIR_STRIPE;
 195
 196         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 197                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 198
 199         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 200                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 201 #ifdef CONFIG_FS_POSIX_ACL
 202         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 203 #endif
 204
 205         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 206                 /* flag mdc connection as lightweight, only used for test
 207                  * purpose, use with care
 208                  */
 209                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 210
 211         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 212         data->ocd_version = LUSTRE_VERSION_CODE;
 213
 214         if (sb->s_flags & MS_RDONLY)
 215                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 216         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 217                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 218
 219         if (sbi->ll_flags & LL_SBI_FLOCK)
 220                 sbi->ll_fop = &ll_file_operations_flock;
 221         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 222                 sbi->ll_fop = &ll_file_operations;
 223         else
 224                 sbi->ll_fop = &ll_file_operations_noflock;
 225
 226         /* real client */
 227         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 228
 229         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 230
 231         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
 232                           data, NULL);
 233         if (err == -EBUSY) {
 234                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing recovery, of which this client is not a part. Please wait for recovery to complete, abort, or time out.\n",
 235                                    md);
 236                 goto out;
 237         } else if (err) {
 238                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 239                 goto out;
 240         }
 241
 242         sbi->ll_md_exp->exp_connect_data = *data;
 243
 244         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 245                            LUSTRE_SEQ_METADATA);
 246         if (err) {
 247                 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
 248                        sbi->ll_md_exp->exp_obd->obd_name, err);
 249                 goto out_md;
 250         }
 251
 252         /* For mount, we only need fs info from MDT0, and also in DNE, it
 253          * can make sure the client can be mounted as long as MDT0 is
 254          * available
 255          */
 256         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 257                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 258                          OBD_STATFS_FOR_MDT0);
 259         if (err)
 260                 goto out_md_fid;
 261
 262         /* This needs to be after statfs to ensure connect has finished.
 263          * Note that "data" does NOT contain the valid connect reply.
 264          * If connecting to a 1.8 server there will be no LMV device, so
 265          * we can access the MDC export directly and exp_connect_flags will
 266          * be non-zero, but if accessing an upgraded 2.1 server it will
 267          * have the correct flags filled in.
 268          * XXX: fill in the LMV exp_connect_flags from MDC(s).
 269          */
 270         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 271         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 272             valid != CLIENT_CONNECT_MDT_REQD) {
 273                 char *buf;
 274
 275                 buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
 276                 if (!buf) {
 277                         err = -ENOMEM;
 278                         goto out_md_fid;
 279                 }
 280                 obd_connect_flags2str(buf, PAGE_SIZE,
 281                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 282                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support feature(s) needed for correct operation of this client (%s). Please upgrade server or downgrade client.\n",
 283                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 284                 kfree(buf);
 285                 err = -EPROTO;
 286                 goto out_md_fid;
 287         }
 288
 289         size = sizeof(*data);
 290         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 291                            KEY_CONN_DATA,  &size, data, NULL);
 292         if (err) {
 293                 CERROR("%s: Get connect data failed: rc = %d\n",
 294                        sbi->ll_md_exp->exp_obd->obd_name, err);
 295                 goto out_md_fid;
 296         }
 297
 298         LASSERT(osfs->os_bsize);
 299         sb->s_blocksize = osfs->os_bsize;
 300         sb->s_blocksize_bits = log2(osfs->os_bsize);
 301         sb->s_magic = LL_SUPER_MAGIC;
 302         sb->s_maxbytes = MAX_LFS_FILESIZE;
 303         sbi->ll_namelen = osfs->os_namelen;
 304
 305         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 306             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 307                 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
 308                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 309         }
 310
 311         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 312                 sb->s_flags |= MS_POSIXACL;
 313                 sbi->ll_flags |= LL_SBI_ACL;
 314         } else {
 315                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 316                 sb->s_flags &= ~MS_POSIXACL;
 317                 sbi->ll_flags &= ~LL_SBI_ACL;
 318         }
 319
 320         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 321                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 322
 323         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 324                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_SHIFT;
 325         else
 326                 sbi->ll_md_brw_pages = 1;
 327
 328         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 329                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 330
 331         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 332                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 333                         LCONSOLE_INFO(
 334                                 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
 335                                 dt);
 336                 } else {
 337                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 338                         sbi->ll_xattr_cache_enabled = 1;
 339                 }
 340         }
 341
 342         obd = class_name2obd(dt);
 343         if (!obd) {
 344                 CERROR("DT %s: not setup or attached\n", dt);
 345                 err = -ENODEV;
 346                 goto out_md_fid;
 347         }
 348
 349         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 350                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 351                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 352                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 353                                   OBD_CONNECT_AT        | OBD_CONNECT_OSS_CAPA |
 354                                   OBD_CONNECT_VBR       | OBD_CONNECT_FULL20   |
 355                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 356                                   OBD_CONNECT_EINPROGRESS |
 357                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 358                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 359
 360         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 361                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 362
 363         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 364                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 365                  * disabled by default, because it can still be enabled on the
 366                  * fly via /sys. As a consequence, we still need to come to an
 367                  * agreement on the supported algorithms at connect time
 368                  */
 369                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 370
 371                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 372                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 373                 else
 374                         data->ocd_cksum_types = cksum_types_supported_client();
 375         }
 376
 377         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 378
 379         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
 380                data->ocd_connect_flags,
 381                data->ocd_version, data->ocd_grant);
 382
 383         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 384         obd->obd_upcall.onu_upcall = cl_ocd_update;
 385
 386         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 387
 388         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 389                           NULL);
 390         if (err == -EBUSY) {
 391                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing recovery, of which this client is not a part.  Please wait for recovery to complete, abort, or time out.\n",
 392                                    dt);
 393                 goto out_md;
 394         } else if (err) {
 395                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 396                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 397                 goto out_md;
 398         }
 399
 400         sbi->ll_dt_exp->exp_connect_data = *data;
 401
 402         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 403                            LUSTRE_SEQ_METADATA);
 404         if (err) {
 405                 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
 406                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 407                 goto out_dt;
 408         }
 409
 410         mutex_lock(&sbi->ll_lco.lco_lock);
 411         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 412         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 413         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 414         mutex_unlock(&sbi->ll_lco.lco_lock);
 415
 416         fid_zero(&sbi->ll_root_fid);
 417         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid);
 418         if (err) {
 419                 CERROR("cannot mds_connect: rc = %d\n", err);
 420                 goto out_lock_cn_cb;
 421         }
 422         if (!fid_is_sane(&sbi->ll_root_fid)) {
 423                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 424                        sbi->ll_md_exp->exp_obd->obd_name,
 425                        PFID(&sbi->ll_root_fid));
 426                 err = -EINVAL;
 427                 goto out_lock_cn_cb;
 428         }
 429         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 430
 431         sb->s_op = &lustre_super_operations;
 432         sb->s_xattr = ll_xattr_handlers;
 433 #if THREAD_SIZE >= 8192 /*b=17630*/
 434         sb->s_export_op = &lustre_export_operations;
 435 #endif
 436
 437         /* make root inode
 438          * XXX: move this to after cbd setup?
 439          */
 440         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 441         if (sbi->ll_flags & LL_SBI_ACL)
 442                 valid |= OBD_MD_FLACL;
 443
 444         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
 445         if (!op_data) {
 446                 err = -ENOMEM;
 447                 goto out_lock_cn_cb;
 448         }
 449
 450         op_data->op_fid1 = sbi->ll_root_fid;
 451         op_data->op_mode = 0;
 452         op_data->op_valid = valid;
 453
 454         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 455         kfree(op_data);
 456         if (err) {
 457                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 458                        sbi->ll_md_exp->exp_obd->obd_name, err);
 459                 goto out_lock_cn_cb;
 460         }
 461
 462         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 463                                sbi->ll_md_exp, &lmd);
 464         if (err) {
 465                 CERROR("failed to understand root inode md: rc = %d\n", err);
 466                 ptlrpc_req_finished(request);
 467                 goto out_lock_cn_cb;
 468         }
 469
 470         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 471         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 472                                             sbi->ll_flags & LL_SBI_32BIT_API),
 473                        &lmd);
 474         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 475         ptlrpc_req_finished(request);
 476
 477         if (IS_ERR(root)) {
 478                 if (lmd.lsm)
 479                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 480 #ifdef CONFIG_FS_POSIX_ACL
 481                 if (lmd.posix_acl) {
 482                         posix_acl_release(lmd.posix_acl);
 483                         lmd.posix_acl = NULL;
 484                 }
 485 #endif
 486                 err = -EBADF;
 487                 CERROR("lustre_lite: bad iget4 for root\n");
 488                 goto out_root;
 489         }
 490
 491         err = ll_close_thread_start(&sbi->ll_lcq);
 492         if (err) {
 493                 CERROR("cannot start close thread: rc %d\n", err);
 494                 goto out_root;
 495         }
 496
 497         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 498         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 499                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 500                                  NULL);
 501         if (err) {
 502                 CERROR("%s: Set checksum failed: rc = %d\n",
 503                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 504                 goto out_root;
 505         }
 506         cl_sb_init(sb);
 507
 508         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 509                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 510                                  sbi->ll_cache, NULL);
 511         if (err) {
 512                 CERROR("%s: Set cache_set failed: rc = %d\n",
 513                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 514                 goto out_root;
 515         }
 516
 517         sb->s_root = d_make_root(root);
 518         if (!sb->s_root) {
 519                 CERROR("%s: can't make root dentry\n",
 520                        ll_get_fsname(sb, NULL, 0));
 521                 err = -ENOMEM;
 522                 goto out_lock_cn_cb;
 523         }
 524
 525         sbi->ll_sdev_orig = sb->s_dev;
 526
 527         /* We set sb->s_dev equal on all lustre clients in order to support
 528          * NFS export clustering.  NFSD requires that the FSID be the same
 529          * on all clients.
 530          */
 531         /* s_dev is also used in lt_compare() to compare two fs, but that is
 532          * only a node-local comparison.
 533          */
 534         uuid = obd_get_uuid(sbi->ll_md_exp);
 535         if (uuid) {
 536                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 537                 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
 538         }
 539
 540         kfree(data);
 541         kfree(osfs);
 542
 543         if (llite_root) {
 544                 err = ldebugfs_register_mountpoint(llite_root, sb, dt, md);
 545                 if (err < 0) {
 546                         CERROR("%s: could not register mount in debugfs: "
 547                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 548                         err = 0;
 549                 }
 550         }
 551
 552         return err;
 553 out_root:
 554         iput(root);
 555 out_lock_cn_cb:
 556         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 557 out_dt:
 558         obd_disconnect(sbi->ll_dt_exp);
 559         sbi->ll_dt_exp = NULL;
 560 out_md_fid:
 561         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 562 out_md:
 563         obd_disconnect(sbi->ll_md_exp);
 564         sbi->ll_md_exp = NULL;
 565 out:
 566         kfree(data);
 567         kfree(osfs);
 568         return err;
 569 }
 570
 571 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 572 {
 573         int size, rc;
 574
 575         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 576         size = sizeof(int);
 577         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 578                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 579         if (rc)
 580                 CERROR("Get max mdsize error rc %d\n", rc);
 581
 582         return rc;
 583 }
 584
 585 /**
 586  * Get the value of the default_easize parameter.
 587  *
 588  * \see client_obd::cl_default_mds_easize
 589  *
 590  * \param[in]  sbi      superblock info for this filesystem
 591  * \param[out] lmmsize  pointer to storage location for value
 592  *
 593  * \retval 0            on success
 594  * \retval negative     negated errno on failure
 595  */
 596 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 597 {
 598         int size, rc;
 599
 600         size = sizeof(int);
 601         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 602                           KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 603         if (rc)
 604                 CERROR("Get default mdsize error rc %d\n", rc);
 605
 606         return rc;
 607 }
 608
 609 /**
 610  * Set the default_easize parameter to the given value.
 611  *
 612  * \see client_obd::cl_default_mds_easize
 613  *
 614  * \param[in] sbi       superblock info for this filesystem
 615  * \param[in] lmmsize   the size to set
 616  *
 617  * \retval 0            on success
 618  * \retval negative     negated errno on failure
 619  */
 620 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 621 {
 622         if (lmmsize < sizeof(struct lov_mds_md) ||
 623             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 624                 return -EINVAL;
 625
 626         return obd_set_info_async(NULL, sbi->ll_md_exp,
 627                                   sizeof(KEY_DEFAULT_EASIZE),
 628                                   KEY_DEFAULT_EASIZE,
 629                                   sizeof(int), &lmmsize, NULL);
 630 }
 631
 632 static void client_common_put_super(struct super_block *sb)
 633 {
 634         struct ll_sb_info *sbi = ll_s2sbi(sb);
 635
 636         ll_close_thread_shutdown(sbi->ll_lcq);
 637
 638         cl_sb_fini(sb);
 639
 640         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 641         obd_disconnect(sbi->ll_dt_exp);
 642         sbi->ll_dt_exp = NULL;
 643
 644         ldebugfs_unregister_mountpoint(sbi);
 645
 646         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 647         obd_disconnect(sbi->ll_md_exp);
 648         sbi->ll_md_exp = NULL;
 649 }
 650
 651 void ll_kill_super(struct super_block *sb)
 652 {
 653         struct ll_sb_info *sbi;
 654
 655         /* not init sb ?*/
 656         if (!(sb->s_flags & MS_ACTIVE))
 657                 return;
 658
 659         sbi = ll_s2sbi(sb);
 660         /* we need to restore s_dev from changed for clustered NFS before
 661          * put_super because new kernels have cached s_dev and change sb->s_dev
 662          * in put_super not affected real removing devices
 663          */
 664         if (sbi) {
 665                 sb->s_dev = sbi->ll_sdev_orig;
 666                 sbi->ll_umounting = 1;
 667
 668                 /* wait running statahead threads to quit */
 669                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 670                         set_current_state(TASK_UNINTERRUPTIBLE);
 671                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 672                 }
 673         }
 674 }
 675
 676 static inline int ll_set_opt(const char *opt, char *data, int fl)
 677 {
 678         if (strncmp(opt, data, strlen(opt)) != 0)
 679                 return 0;
 680         else
 681                 return fl;
 682 }
 683
 684 /* non-client-specific mount options are parsed in lmd_parse */
 685 static int ll_options(char *options, int *flags)
 686 {
 687         int tmp;
 688         char *s1 = options, *s2;
 689
 690         if (!options)
 691                 return 0;
 692
 693         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 694
 695         while (*s1) {
 696                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 697                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 698                 if (tmp) {
 699                         *flags |= tmp;
 700                         goto next;
 701                 }
 702                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 703                 if (tmp) {
 704                         *flags |= tmp;
 705                         goto next;
 706                 }
 707                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 708                 if (tmp) {
 709                         *flags |= tmp;
 710                         goto next;
 711                 }
 712                 tmp = ll_set_opt("noflock", s1,
 713                                  LL_SBI_FLOCK | LL_SBI_LOCALFLOCK);
 714                 if (tmp) {
 715                         *flags &= ~tmp;
 716                         goto next;
 717                 }
 718                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 719                 if (tmp) {
 720                         *flags |= tmp;
 721                         goto next;
 722                 }
 723                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 724                 if (tmp) {
 725                         *flags &= ~tmp;
 726                         goto next;
 727                 }
 728                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 729                 if (tmp) {
 730                         *flags |= tmp;
 731                         goto next;
 732                 }
 733                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 734                 if (tmp) {
 735                         *flags &= ~tmp;
 736                         goto next;
 737                 }
 738
 739                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 740                 if (tmp) {
 741                         *flags |= tmp;
 742                         goto next;
 743                 }
 744                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 745                 if (tmp) {
 746                         *flags &= ~tmp;
 747                         goto next;
 748                 }
 749                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 750                 if (tmp) {
 751                         *flags |= tmp;
 752                         goto next;
 753                 }
 754                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 755                 if (tmp) {
 756                         *flags &= ~tmp;
 757                         goto next;
 758                 }
 759                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 760                 if (tmp) {
 761                         *flags |= tmp;
 762                         goto next;
 763                 }
 764                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 765                 if (tmp) {
 766                         *flags &= ~tmp;
 767                         goto next;
 768                 }
 769                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 770                 if (tmp) {
 771                         *flags |= tmp;
 772                         goto next;
 773                 }
 774                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 775                 if (tmp) {
 776                         *flags |= tmp;
 777                         goto next;
 778                 }
 779                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 780                 if (tmp) {
 781                         *flags |= tmp;
 782                         goto next;
 783                 }
 784                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 785                 if (tmp) {
 786                         *flags &= ~tmp;
 787                         goto next;
 788                 }
 789                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 790                                    s1);
 791                 return -EINVAL;
 792
 793 next:
 794                 /* Find next opt */
 795                 s2 = strchr(s1, ',');
 796                 if (!s2)
 797                         break;
 798                 s1 = s2 + 1;
 799         }
 800         return 0;
 801 }
 802
 803 void ll_lli_init(struct ll_inode_info *lli)
 804 {
 805         lli->lli_inode_magic = LLI_INODE_MAGIC;
 806         lli->lli_flags = 0;
 807         lli->lli_ioepoch = 0;
 808         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 809         spin_lock_init(&lli->lli_lock);
 810         lli->lli_posix_acl = NULL;
 811         /* Do not set lli_fid, it has been initialized already. */
 812         fid_zero(&lli->lli_pfid);
 813         INIT_LIST_HEAD(&lli->lli_close_list);
 814         lli->lli_pending_och = NULL;
 815         lli->lli_mds_read_och = NULL;
 816         lli->lli_mds_write_och = NULL;
 817         lli->lli_mds_exec_och = NULL;
 818         lli->lli_open_fd_read_count = 0;
 819         lli->lli_open_fd_write_count = 0;
 820         lli->lli_open_fd_exec_count = 0;
 821         mutex_init(&lli->lli_och_mutex);
 822         spin_lock_init(&lli->lli_agl_lock);
 823         lli->lli_has_smd = false;
 824         spin_lock_init(&lli->lli_layout_lock);
 825         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 826         lli->lli_clob = NULL;
 827
 828         init_rwsem(&lli->lli_xattrs_list_rwsem);
 829         mutex_init(&lli->lli_xattrs_enq_lock);
 830
 831         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 832         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 833                 mutex_init(&lli->lli_readdir_mutex);
 834                 lli->lli_opendir_key = NULL;
 835                 lli->lli_sai = NULL;
 836                 spin_lock_init(&lli->lli_sa_lock);
 837                 lli->lli_opendir_pid = 0;
 838                 lli->lli_sa_enabled = 0;
 839                 lli->lli_def_stripe_offset = -1;
 840         } else {
 841                 mutex_init(&lli->lli_size_mutex);
 842                 lli->lli_symlink_name = NULL;
 843                 init_rwsem(&lli->lli_trunc_sem);
 844                 range_lock_tree_init(&lli->lli_write_tree);
 845                 init_rwsem(&lli->lli_glimpse_sem);
 846                 lli->lli_glimpse_time = 0;
 847                 INIT_LIST_HEAD(&lli->lli_agl_list);
 848                 lli->lli_agl_index = 0;
 849                 lli->lli_async_rc = 0;
 850         }
 851         mutex_init(&lli->lli_layout_mutex);
 852 }
 853
 854 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 855 {
 856         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 857
 858         bdi->name = "lustre";
 859         return bdi_register(bdi, NULL, "lustre-%d",
 860                             atomic_inc_return(&ll_bdi_num));
 861 }
 862
 863 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 864 {
 865         struct lustre_profile *lprof = NULL;
 866         struct lustre_sb_info *lsi = s2lsi(sb);
 867         struct ll_sb_info *sbi;
 868         char  *dt = NULL, *md = NULL;
 869         char  *profilenm = get_profile_name(sb);
 870         struct config_llog_instance *cfg;
 871         int    err;
 872
 873         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 874
 875         cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
 876         if (!cfg)
 877                 return -ENOMEM;
 878
 879         try_module_get(THIS_MODULE);
 880
 881         /* client additional sb info */
 882         sbi = ll_init_sbi(sb);
 883         lsi->lsi_llsbi = sbi;
 884         if (!sbi) {
 885                 module_put(THIS_MODULE);
 886                 kfree(cfg);
 887                 return -ENOMEM;
 888         }
 889
 890         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 891         if (err)
 892                 goto out_free;
 893
 894         err = bdi_init(&lsi->lsi_bdi);
 895         if (err)
 896                 goto out_free;
 897         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 898         lsi->lsi_bdi.capabilities = 0;
 899         err = ll_bdi_register(&lsi->lsi_bdi);
 900         if (err)
 901                 goto out_free;
 902
 903         sb->s_bdi = &lsi->lsi_bdi;
 904         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 905         sb->s_d_op = &ll_d_ops;
 906
 907         /* Generate a string unique to this super, in case some joker tries
 908          * to mount the same fs at two mount points.
 909          * Use the address of the super itself.
 910          */
 911         cfg->cfg_instance = sb;
 912         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
 913         cfg->cfg_callback = class_config_llog_handler;
 914         /* set up client obds */
 915         err = lustre_process_log(sb, profilenm, cfg);
 916         if (err < 0)
 917                 goto out_free;
 918
 919         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
 920         lprof = class_get_profile(profilenm);
 921         if (!lprof) {
 922                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS.  Does that filesystem exist?\n",
 923                                    profilenm);
 924                 err = -EINVAL;
 925                 goto out_free;
 926         }
 927         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
 928                lprof->lp_md, lprof->lp_dt);
 929
 930         dt = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
 931         if (!dt) {
 932                 err = -ENOMEM;
 933                 goto out_free;
 934         }
 935
 936         md = kasprintf(GFP_NOFS, "%s-%p", lprof->lp_md, cfg->cfg_instance);
 937         if (!md) {
 938                 err = -ENOMEM;
 939                 goto out_free;
 940         }
 941
 942         /* connections, registrations, sb setup */
 943         err = client_common_fill_super(sb, md, dt, mnt);
 944
 945 out_free:
 946         kfree(md);
 947         kfree(dt);
 948         if (err)
 949                 ll_put_super(sb);
 950         else if (sbi->ll_flags & LL_SBI_VERBOSE)
 951                 LCONSOLE_WARN("Mounted %s\n", profilenm);
 952
 953         kfree(cfg);
 954         return err;
 955 } /* ll_fill_super */
 956
 957 void ll_put_super(struct super_block *sb)
 958 {
 959         struct config_llog_instance cfg, params_cfg;
 960         struct obd_device *obd;
 961         struct lustre_sb_info *lsi = s2lsi(sb);
 962         struct ll_sb_info *sbi = ll_s2sbi(sb);
 963         char *profilenm = get_profile_name(sb);
 964         int next, force = 1, rc = 0;
 965         long ccc_count;
 966
 967         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
 968
 969         cfg.cfg_instance = sb;
 970         lustre_end_log(sb, profilenm, &cfg);
 971
 972         params_cfg.cfg_instance = sb;
 973         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
 974
 975         if (sbi->ll_md_exp) {
 976                 obd = class_exp2obd(sbi->ll_md_exp);
 977                 if (obd)
 978                         force = obd->obd_force;
 979         }
 980
 981         /* Wait for unstable pages to be committed to stable storage */
 982         if (!force) {
 983                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
 984
 985                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
 986                                   !atomic_long_read(&sbi->ll_cache->ccc_unstable_nr),
 987                                   &lwi);
 988         }
 989
 990         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
 991         if (!force && rc != -EINTR)
 992                 LASSERTF(!ccc_count, "count: %li\n", ccc_count);
 993
 994         /* We need to set force before the lov_disconnect in
 995          * lustre_common_put_super, since l_d cleans up osc's as well.
 996          */
 997         if (force) {
 998                 next = 0;
 999                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1000                                                      &next)) != NULL) {
1001                         obd->obd_force = force;
1002                 }
1003         }
1004
1005         if (sbi->ll_lcq) {
1006                 /* Only if client_common_fill_super succeeded */
1007                 client_common_put_super(sb);
1008         }
1009
1010         next = 0;
1011         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1012                 class_manual_cleanup(obd);
1013
1014         if (sbi->ll_flags & LL_SBI_VERBOSE)
1015                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1016
1017         if (profilenm)
1018                 class_del_profile(profilenm);
1019
1020         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1021                 bdi_destroy(&lsi->lsi_bdi);
1022                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1023         }
1024
1025         ll_free_sbi(sb);
1026         lsi->lsi_llsbi = NULL;
1027
1028         lustre_common_put_super(sb);
1029
1030         cl_env_cache_purge(~0);
1031
1032         module_put(THIS_MODULE);
1033 } /* client_put_super */
1034
1035 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1036 {
1037         struct inode *inode = NULL;
1038
1039         /* NOTE: we depend on atomic igrab() -bzzz */
1040         lock_res_and_lock(lock);
1041         if (lock->l_resource->lr_lvb_inode) {
1042                 struct ll_inode_info *lli;
1043
1044                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1045                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1046                         inode = igrab(lock->l_resource->lr_lvb_inode);
1047                 } else {
1048                         inode = lock->l_resource->lr_lvb_inode;
1049                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1050                                          D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
1051                                          lock->l_resource->lr_lvb_inode,
1052                                          lli->lli_inode_magic);
1053                         inode = NULL;
1054                 }
1055         }
1056         unlock_res_and_lock(lock);
1057         return inode;
1058 }
1059
1060 static void ll_dir_clear_lsm_md(struct inode *inode)
1061 {
1062         struct ll_inode_info *lli = ll_i2info(inode);
1063
1064         LASSERT(S_ISDIR(inode->i_mode));
1065
1066         if (lli->lli_lsm_md) {
1067                 lmv_free_memmd(lli->lli_lsm_md);
1068                 lli->lli_lsm_md = NULL;
1069         }
1070 }
1071
1072 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1073                                       const struct lu_fid *fid,
1074                                       struct lustre_md *md)
1075 {
1076         struct ll_sb_info *sbi = ll_s2sbi(sb);
1077         struct mdt_body *body = md->body;
1078         struct inode *inode;
1079         ino_t ino;
1080
1081         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1082         inode = iget_locked(sb, ino);
1083         if (!inode) {
1084                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1085                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1086                 return ERR_PTR(-ENOENT);
1087         }
1088
1089         if (inode->i_state & I_NEW) {
1090                 struct ll_inode_info *lli = ll_i2info(inode);
1091                 struct lmv_stripe_md *lsm = md->lmv;
1092
1093                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1094                                 (body->mbo_mode & S_IFMT);
1095                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1096                          PFID(fid));
1097
1098                 LTIME_S(inode->i_mtime) = 0;
1099                 LTIME_S(inode->i_atime) = 0;
1100                 LTIME_S(inode->i_ctime) = 0;
1101                 inode->i_rdev = 0;
1102
1103                 inode->i_op = &ll_dir_inode_operations;
1104                 inode->i_fop = &ll_dir_operations;
1105                 lli->lli_fid = *fid;
1106                 ll_lli_init(lli);
1107
1108                 LASSERT(lsm);
1109                 /* master object FID */
1110                 lli->lli_pfid = body->mbo_fid1;
1111                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1112                        lli, PFID(fid), PFID(&lli->lli_pfid));
1113                 unlock_new_inode(inode);
1114         }
1115
1116         return inode;
1117 }
1118
1119 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1120 {
1121         struct lmv_stripe_md *lsm = md->lmv;
1122         struct lu_fid *fid;
1123         int i;
1124
1125         LASSERT(lsm);
1126         /*
1127          * XXX sigh, this lsm_root initialization should be in
1128          * LMV layer, but it needs ll_iget right now, so we
1129          * put this here right now.
1130          */
1131         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1132                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1133                 LASSERT(!lsm->lsm_md_oinfo[i].lmo_root);
1134                 /* Unfortunately ll_iget will call ll_update_inode,
1135                  * where the initialization of slave inode is slightly
1136                  * different, so it reset lsm_md to NULL to avoid
1137                  * initializing lsm for slave inode.
1138                  */
1139                 /* For migrating inode, master stripe and master object will
1140                  * be same, so we only need assign this inode
1141                  */
1142                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && !i)
1143                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1144                 else
1145                         lsm->lsm_md_oinfo[i].lmo_root =
1146                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1147                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1148                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1149
1150                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1151                         return rc;
1152                 }
1153         }
1154
1155         return 0;
1156 }
1157
1158 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1159                                 const struct lmv_stripe_md *lsm_md2)
1160 {
1161         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1162                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1163                lsm_md1->lsm_md_master_mdt_index ==
1164                         lsm_md2->lsm_md_master_mdt_index &&
1165                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1166                lsm_md1->lsm_md_layout_version ==
1167                         lsm_md2->lsm_md_layout_version &&
1168                !strcmp(lsm_md1->lsm_md_pool_name,
1169                        lsm_md2->lsm_md_pool_name);
1170 }
1171
1172 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1173 {
1174         struct ll_inode_info *lli = ll_i2info(inode);
1175         struct lmv_stripe_md *lsm = md->lmv;
1176         int rc;
1177
1178         LASSERT(S_ISDIR(inode->i_mode));
1179         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1180                PFID(ll_inode2fid(inode)));
1181
1182         /* no striped information from request. */
1183         if (!lsm) {
1184                 if (!lli->lli_lsm_md) {
1185                         return 0;
1186                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1187                            LMV_HASH_FLAG_MIGRATION) {
1188                         /*
1189                          * migration is done, the temporay MIGRATE layout has
1190                          * been removed
1191                          */
1192                         CDEBUG(D_INODE, DFID" finish migration.\n",
1193                                PFID(ll_inode2fid(inode)));
1194                         lmv_free_memmd(lli->lli_lsm_md);
1195                         lli->lli_lsm_md = NULL;
1196                         return 0;
1197                 } else {
1198                         /*
1199                          * The lustre_md from req does not include stripeEA,
1200                          * see ll_md_setattr
1201                          */
1202                         return 0;
1203                 }
1204         }
1205
1206         /* set the directory layout */
1207         if (!lli->lli_lsm_md) {
1208                 rc = ll_init_lsm_md(inode, md);
1209                 if (rc)
1210                         return rc;
1211
1212                 lli->lli_lsm_md = lsm;
1213                 /*
1214                  * set lsm_md to NULL, so the following free lustre_md
1215                  * will not free this lsm
1216                  */
1217                 md->lmv = NULL;
1218                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1219                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1220                 return 0;
1221         }
1222
1223         /* Compare the old and new stripe information */
1224         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1225                 struct lmv_stripe_md *old_lsm = lli->lli_lsm_md;
1226                 int idx;
1227
1228                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p) magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1229                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1230                        inode, lsm, old_lsm,
1231                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1232                        lsm->lsm_md_stripe_count,
1233                        old_lsm->lsm_md_stripe_count,
1234                        lsm->lsm_md_master_mdt_index,
1235                        old_lsm->lsm_md_master_mdt_index,
1236                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1237                        lsm->lsm_md_layout_version,
1238                        old_lsm->lsm_md_layout_version,
1239                        lsm->lsm_md_pool_name,
1240                        old_lsm->lsm_md_pool_name);
1241
1242                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1243                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1244                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1245                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1246                 }
1247
1248                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1249                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1250                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1251                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1252                 }
1253
1254                 return -EIO;
1255         }
1256
1257         return 0;
1258 }
1259
1260 void ll_clear_inode(struct inode *inode)
1261 {
1262         struct ll_inode_info *lli = ll_i2info(inode);
1263         struct ll_sb_info *sbi = ll_i2sbi(inode);
1264
1265         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1266                PFID(ll_inode2fid(inode)), inode);
1267
1268         if (S_ISDIR(inode->i_mode)) {
1269                 /* these should have been cleared in ll_file_release */
1270                 LASSERT(!lli->lli_opendir_key);
1271                 LASSERT(!lli->lli_sai);
1272                 LASSERT(lli->lli_opendir_pid == 0);
1273         }
1274
1275         spin_lock(&lli->lli_lock);
1276         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1277         spin_unlock(&lli->lli_lock);
1278         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1279
1280         LASSERT(!lli->lli_open_fd_write_count);
1281         LASSERT(!lli->lli_open_fd_read_count);
1282         LASSERT(!lli->lli_open_fd_exec_count);
1283
1284         if (lli->lli_mds_write_och)
1285                 ll_md_real_close(inode, FMODE_WRITE);
1286         if (lli->lli_mds_exec_och)
1287                 ll_md_real_close(inode, FMODE_EXEC);
1288         if (lli->lli_mds_read_och)
1289                 ll_md_real_close(inode, FMODE_READ);
1290
1291         if (S_ISLNK(inode->i_mode)) {
1292                 kfree(lli->lli_symlink_name);
1293                 lli->lli_symlink_name = NULL;
1294         }
1295
1296         ll_xattr_cache_destroy(inode);
1297
1298 #ifdef CONFIG_FS_POSIX_ACL
1299         if (lli->lli_posix_acl) {
1300                 posix_acl_release(lli->lli_posix_acl);
1301                 lli->lli_posix_acl = NULL;
1302         }
1303 #endif
1304         lli->lli_inode_magic = LLI_INODE_DEAD;
1305
1306         if (S_ISDIR(inode->i_mode))
1307                 ll_dir_clear_lsm_md(inode);
1308         if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1309                 LASSERT(list_empty(&lli->lli_agl_list));
1310
1311         /*
1312          * XXX This has to be done before lsm is freed below, because
1313          * cl_object still uses inode lsm.
1314          */
1315         cl_inode_fini(inode);
1316         lli->lli_has_smd = false;
1317 }
1318
1319 #define TIMES_SET_FLAGS (ATTR_MTIME_SET | ATTR_ATIME_SET | ATTR_TIMES_SET)
1320
1321 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1322                          struct md_open_data **mod)
1323 {
1324         struct lustre_md md;
1325         struct inode *inode = d_inode(dentry);
1326         struct ll_sb_info *sbi = ll_i2sbi(inode);
1327         struct ptlrpc_request *request = NULL;
1328         int rc, ia_valid;
1329
1330         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1331                                      LUSTRE_OPC_ANY, NULL);
1332         if (IS_ERR(op_data))
1333                 return PTR_ERR(op_data);
1334
1335         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1336                         &request, mod);
1337         if (rc) {
1338                 ptlrpc_req_finished(request);
1339                 if (rc == -ENOENT) {
1340                         clear_nlink(inode);
1341                         /* Unlinked special device node? Or just a race?
1342                          * Pretend we did everything.
1343                          */
1344                         if (!S_ISREG(inode->i_mode) &&
1345                             !S_ISDIR(inode->i_mode)) {
1346                                 ia_valid = op_data->op_attr.ia_valid;
1347                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1348                                 rc = simple_setattr(dentry, &op_data->op_attr);
1349                                 op_data->op_attr.ia_valid = ia_valid;
1350                         }
1351                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1352                         CERROR("md_setattr fails: rc = %d\n", rc);
1353                 }
1354                 return rc;
1355         }
1356
1357         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1358                               sbi->ll_md_exp, &md);
1359         if (rc) {
1360                 ptlrpc_req_finished(request);
1361                 return rc;
1362         }
1363
1364         ia_valid = op_data->op_attr.ia_valid;
1365         /* inode size will be in cl_setattr_ost, can't do it now since dirty
1366          * cache is not cleared yet.
1367          */
1368         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1369         rc = simple_setattr(dentry, &op_data->op_attr);
1370         op_data->op_attr.ia_valid = ia_valid;
1371
1372         /* Extract epoch data if obtained. */
1373         op_data->op_handle = md.body->mbo_handle;
1374         op_data->op_ioepoch = md.body->mbo_ioepoch;
1375
1376         rc = ll_update_inode(inode, &md);
1377         ptlrpc_req_finished(request);
1378
1379         return rc;
1380 }
1381
1382 /* Close IO epoch and send Size-on-MDS attribute update. */
1383 static int ll_setattr_done_writing(struct inode *inode,
1384                                    struct md_op_data *op_data,
1385                                    struct md_open_data *mod)
1386 {
1387         struct ll_inode_info *lli = ll_i2info(inode);
1388         int rc = 0;
1389
1390         if (!S_ISREG(inode->i_mode))
1391                 return 0;
1392
1393         CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1394                op_data->op_ioepoch, PFID(&lli->lli_fid));
1395
1396         op_data->op_flags = MF_EPOCH_CLOSE;
1397         ll_done_writing_attr(inode, op_data);
1398         ll_pack_inode2opdata(inode, op_data, NULL);
1399
1400         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1401         if (rc == -EAGAIN)
1402                 /* MDS has instructed us to obtain Size-on-MDS attribute
1403                  * from OSTs and send setattr to back to MDS.
1404                  */
1405                 rc = ll_som_update(inode, op_data);
1406         else if (rc) {
1407                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1408                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1409                        PFID(ll_inode2fid(inode)), rc);
1410         }
1411         return rc;
1412 }
1413
1414 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1415  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1416  * keep these values until such a time that objects are allocated for it.
1417  * We do the MDS operations first, as it is checking permissions for us.
1418  * We don't to the MDS RPC if there is nothing that we want to store there,
1419  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1420  * going to do an RPC anyways.
1421  *
1422  * If we are doing a truncate, we will send the mtime and ctime updates
1423  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1424  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1425  * at the same time.
1426  *
1427  * In case of HSMimport, we only set attr on MDS.
1428  */
1429 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1430 {
1431         struct inode *inode = d_inode(dentry);
1432         struct ll_inode_info *lli = ll_i2info(inode);
1433         struct md_op_data *op_data = NULL;
1434         struct md_open_data *mod = NULL;
1435         bool file_is_released = false;
1436         int rc = 0, rc1 = 0;
1437
1438         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, valid %x, hsm_import %d\n",
1439                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid), inode,
1440                i_size_read(inode), attr->ia_size, attr->ia_valid, hsm_import);
1441
1442         if (attr->ia_valid & ATTR_SIZE) {
1443                 /* Check new size against VFS/VM file size limit and rlimit */
1444                 rc = inode_newsize_ok(inode, attr->ia_size);
1445                 if (rc)
1446                         return rc;
1447
1448                 /* The maximum Lustre file size is variable, based on the
1449                  * OST maximum object size and number of stripes.  This
1450                  * needs another check in addition to the VFS check above.
1451                  */
1452                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1453                         CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1454                                PFID(&lli->lli_fid), attr->ia_size,
1455                                ll_file_maxbytes(inode));
1456                         return -EFBIG;
1457                 }
1458
1459                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1460         }
1461
1462         /* POSIX: check before ATTR_*TIME_SET set (from setattr_prepare) */
1463         if (attr->ia_valid & TIMES_SET_FLAGS) {
1464                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1465                     !capable(CFS_CAP_FOWNER))
1466                         return -EPERM;
1467         }
1468
1469         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1470         if (attr->ia_valid & ATTR_CTIME) {
1471                 attr->ia_ctime = CURRENT_TIME;
1472                 attr->ia_valid |= ATTR_CTIME_SET;
1473         }
1474         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1475             (attr->ia_valid & ATTR_ATIME)) {
1476                 attr->ia_atime = CURRENT_TIME;
1477                 attr->ia_valid |= ATTR_ATIME_SET;
1478         }
1479         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1480             (attr->ia_valid & ATTR_MTIME)) {
1481                 attr->ia_mtime = CURRENT_TIME;
1482                 attr->ia_valid |= ATTR_MTIME_SET;
1483         }
1484
1485         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1486                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %llu\n",
1487                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1488                        (s64)ktime_get_real_seconds());
1489
1490         /* We always do an MDS RPC, even if we're only changing the size;
1491          * only the MDS knows whether truncate() should fail with -ETXTBUSY
1492          */
1493
1494         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1495         if (!op_data)
1496                 return -ENOMEM;
1497
1498         if (!S_ISDIR(inode->i_mode))
1499                 inode_unlock(inode);
1500
1501         /* truncate on a released file must failed with -ENODATA,
1502          * so size must not be set on MDS for released file
1503          * but other attributes must be set
1504          */
1505         if (S_ISREG(inode->i_mode)) {
1506                 struct lov_stripe_md *lsm;
1507                 __u32 gen;
1508
1509                 ll_layout_refresh(inode, &gen);
1510                 lsm = ccc_inode_lsm_get(inode);
1511                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1512                         file_is_released = true;
1513                 ccc_inode_lsm_put(inode, lsm);
1514
1515                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1516                         if (file_is_released) {
1517                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1518                                 if (rc < 0)
1519                                         goto out;
1520
1521                                 file_is_released = false;
1522                                 ll_layout_refresh(inode, &gen);
1523                         }
1524
1525                         /*
1526                          * If we are changing file size, file content is
1527                          * modified, flag it.
1528                          */
1529                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1530                         spin_lock(&lli->lli_lock);
1531                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1532                         spin_unlock(&lli->lli_lock);
1533                         op_data->op_bias |= MDS_DATA_MODIFIED;
1534                 }
1535         }
1536
1537         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1538
1539         /* Open epoch for truncate. */
1540         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1541             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1542                 op_data->op_flags = MF_EPOCH_OPEN;
1543
1544         rc = ll_md_setattr(dentry, op_data, &mod);
1545         if (rc)
1546                 goto out;
1547
1548         /* RPC to MDT is sent, cancel data modification flag */
1549         if (op_data->op_bias & MDS_DATA_MODIFIED) {
1550                 spin_lock(&lli->lli_lock);
1551                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1552                 spin_unlock(&lli->lli_lock);
1553         }
1554
1555         ll_ioepoch_open(lli, op_data->op_ioepoch);
1556         if (!S_ISREG(inode->i_mode) || file_is_released) {
1557                 rc = 0;
1558                 goto out;
1559         }
1560
1561         if (attr->ia_valid & (ATTR_SIZE |
1562                               ATTR_ATIME | ATTR_ATIME_SET |
1563                               ATTR_MTIME | ATTR_MTIME_SET)) {
1564                 /* For truncate and utimes sending attributes to OSTs, setting
1565                  * mtime/atime to the past will be performed under PW [0:EOF]
1566                  * extent lock (new_size:EOF for truncate).  It may seem
1567                  * excessive to send mtime/atime updates to OSTs when not
1568                  * setting times to past, but it is necessary due to possible
1569                  * time de-synchronization between MDT inode and OST objects
1570                  */
1571                 if (attr->ia_valid & ATTR_SIZE)
1572                         down_write(&lli->lli_trunc_sem);
1573                 rc = cl_setattr_ost(inode, attr);
1574                 if (attr->ia_valid & ATTR_SIZE)
1575                         up_write(&lli->lli_trunc_sem);
1576         }
1577 out:
1578         if (op_data->op_ioepoch) {
1579                 rc1 = ll_setattr_done_writing(inode, op_data, mod);
1580                 if (!rc)
1581                         rc = rc1;
1582         }
1583         ll_finish_md_op_data(op_data);
1584
1585         if (!S_ISDIR(inode->i_mode)) {
1586                 inode_lock(inode);
1587                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1588                         inode_dio_wait(inode);
1589         }
1590
1591         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1592                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1593
1594         return rc;
1595 }
1596
1597 int ll_setattr(struct dentry *de, struct iattr *attr)
1598 {
1599         int mode = d_inode(de)->i_mode;
1600
1601         if ((attr->ia_valid & (ATTR_CTIME | ATTR_SIZE | ATTR_MODE)) ==
1602                               (ATTR_CTIME | ATTR_SIZE | ATTR_MODE))
1603                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1604
1605         if (((attr->ia_valid & (ATTR_MODE | ATTR_FORCE | ATTR_SIZE)) ==
1606                                (ATTR_SIZE | ATTR_MODE)) &&
1607             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1608              (((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) &&
1609               !(attr->ia_mode & S_ISGID))))
1610                 attr->ia_valid |= ATTR_FORCE;
1611
1612         if ((attr->ia_valid & ATTR_MODE) &&
1613             (mode & S_ISUID) &&
1614             !(attr->ia_mode & S_ISUID) &&
1615             !(attr->ia_valid & ATTR_KILL_SUID))
1616                 attr->ia_valid |= ATTR_KILL_SUID;
1617
1618         if ((attr->ia_valid & ATTR_MODE) &&
1619             ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) &&
1620             !(attr->ia_mode & S_ISGID) &&
1621             !(attr->ia_valid & ATTR_KILL_SGID))
1622                 attr->ia_valid |= ATTR_KILL_SGID;
1623
1624         return ll_setattr_raw(de, attr, false);
1625 }
1626
1627 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1628                        __u64 max_age, __u32 flags)
1629 {
1630         struct ll_sb_info *sbi = ll_s2sbi(sb);
1631         struct obd_statfs obd_osfs;
1632         int rc;
1633
1634         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1635         if (rc) {
1636                 CERROR("md_statfs fails: rc = %d\n", rc);
1637                 return rc;
1638         }
1639
1640         osfs->os_type = sb->s_magic;
1641
1642         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1643                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1644                osfs->os_files);
1645
1646         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1647                 flags |= OBD_STATFS_NODELAY;
1648
1649         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1650         if (rc) {
1651                 CERROR("obd_statfs fails: rc = %d\n", rc);
1652                 return rc;
1653         }
1654
1655         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1656                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1657                obd_osfs.os_files);
1658
1659         osfs->os_bsize = obd_osfs.os_bsize;
1660         osfs->os_blocks = obd_osfs.os_blocks;
1661         osfs->os_bfree = obd_osfs.os_bfree;
1662         osfs->os_bavail = obd_osfs.os_bavail;
1663
1664         /* If we don't have as many objects free on the OST as inodes
1665          * on the MDS, we reduce the total number of inodes to
1666          * compensate, so that the "inodes in use" number is correct.
1667          */
1668         if (obd_osfs.os_ffree < osfs->os_ffree) {
1669                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1670                         obd_osfs.os_ffree;
1671                 osfs->os_ffree = obd_osfs.os_ffree;
1672         }
1673
1674         return rc;
1675 }
1676
1677 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1678 {
1679         struct super_block *sb = de->d_sb;
1680         struct obd_statfs osfs;
1681         int rc;
1682
1683         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1684         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1685
1686         /* Some amount of caching on the client is allowed */
1687         rc = ll_statfs_internal(sb, &osfs,
1688                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1689                                 0);
1690         if (rc)
1691                 return rc;
1692
1693         statfs_unpack(sfs, &osfs);
1694
1695         /* We need to downshift for all 32-bit kernels, because we can't
1696          * tell if the kernel is being called via sys_statfs64() or not.
1697          * Stop before overflowing f_bsize - in which case it is better
1698          * to just risk EOVERFLOW if caller is using old sys_statfs().
1699          */
1700         if (sizeof(long) < 8) {
1701                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1702                         sfs->f_bsize <<= 1;
1703
1704                         osfs.os_blocks >>= 1;
1705                         osfs.os_bfree >>= 1;
1706                         osfs.os_bavail >>= 1;
1707                 }
1708         }
1709
1710         sfs->f_blocks = osfs.os_blocks;
1711         sfs->f_bfree = osfs.os_bfree;
1712         sfs->f_bavail = osfs.os_bavail;
1713         sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1714         return 0;
1715 }
1716
1717 void ll_inode_size_lock(struct inode *inode)
1718 {
1719         struct ll_inode_info *lli;
1720
1721         LASSERT(!S_ISDIR(inode->i_mode));
1722
1723         lli = ll_i2info(inode);
1724         mutex_lock(&lli->lli_size_mutex);
1725 }
1726
1727 void ll_inode_size_unlock(struct inode *inode)
1728 {
1729         struct ll_inode_info *lli;
1730
1731         lli = ll_i2info(inode);
1732         mutex_unlock(&lli->lli_size_mutex);
1733 }
1734
1735 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1736 {
1737         struct ll_inode_info *lli = ll_i2info(inode);
1738         struct mdt_body *body = md->body;
1739         struct lov_stripe_md *lsm = md->lsm;
1740         struct ll_sb_info *sbi = ll_i2sbi(inode);
1741
1742         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1743         if (lsm) {
1744                 if (!lli->lli_has_smd &&
1745                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1746                         cl_file_inode_init(inode, md);
1747
1748                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1749                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1750                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1751         }
1752
1753         if (S_ISDIR(inode->i_mode)) {
1754                 int rc;
1755
1756                 rc = ll_update_lsm_md(inode, md);
1757                 if (rc)
1758                         return rc;
1759         }
1760
1761 #ifdef CONFIG_FS_POSIX_ACL
1762         if (body->mbo_valid & OBD_MD_FLACL) {
1763                 spin_lock(&lli->lli_lock);
1764                 if (lli->lli_posix_acl)
1765                         posix_acl_release(lli->lli_posix_acl);
1766                 lli->lli_posix_acl = md->posix_acl;
1767                 spin_unlock(&lli->lli_lock);
1768         }
1769 #endif
1770         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1771                                         sbi->ll_flags & LL_SBI_32BIT_API);
1772         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1773
1774         if (body->mbo_valid & OBD_MD_FLATIME) {
1775                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1776                         LTIME_S(inode->i_atime) = body->mbo_atime;
1777                 lli->lli_atime = body->mbo_atime;
1778         }
1779         if (body->mbo_valid & OBD_MD_FLMTIME) {
1780                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1781                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1782                                inode->i_ino, LTIME_S(inode->i_mtime),
1783                                body->mbo_mtime);
1784                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1785                 }
1786                 lli->lli_mtime = body->mbo_mtime;
1787         }
1788         if (body->mbo_valid & OBD_MD_FLCTIME) {
1789                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
1790                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
1791                 lli->lli_ctime = body->mbo_ctime;
1792         }
1793         if (body->mbo_valid & OBD_MD_FLMODE)
1794                 inode->i_mode = (inode->i_mode & S_IFMT) |
1795                                 (body->mbo_mode & ~S_IFMT);
1796         if (body->mbo_valid & OBD_MD_FLTYPE)
1797                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1798                                 (body->mbo_mode & S_IFMT);
1799         LASSERT(inode->i_mode != 0);
1800         if (S_ISREG(inode->i_mode))
1801                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1802                                        LL_MAX_BLKSIZE_BITS);
1803         else
1804                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1805         if (body->mbo_valid & OBD_MD_FLUID)
1806                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1807         if (body->mbo_valid & OBD_MD_FLGID)
1808                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1809         if (body->mbo_valid & OBD_MD_FLFLAGS)
1810                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
1811         if (body->mbo_valid & OBD_MD_FLNLINK)
1812                 set_nlink(inode, body->mbo_nlink);
1813         if (body->mbo_valid & OBD_MD_FLRDEV)
1814                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1815
1816         if (body->mbo_valid & OBD_MD_FLID) {
1817                 /* FID shouldn't be changed! */
1818                 if (fid_is_sane(&lli->lli_fid)) {
1819                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
1820                                  "Trying to change FID "DFID" to the "DFID", inode "DFID"(%p)\n",
1821                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
1822                                  PFID(ll_inode2fid(inode)), inode);
1823                 } else {
1824                         lli->lli_fid = body->mbo_fid1;
1825                 }
1826         }
1827
1828         LASSERT(fid_seq(&lli->lli_fid) != 0);
1829
1830         if (body->mbo_valid & OBD_MD_FLSIZE) {
1831                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1832                     S_ISREG(inode->i_mode)) {
1833                         struct lustre_handle lockh;
1834                         enum ldlm_mode mode;
1835
1836                         /* As it is possible a blocking ast has been processed
1837                          * by this time, we need to check there is an UPDATE
1838                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1839                          * it.
1840                          */
1841                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1842                                                &lockh, LDLM_FL_CBPENDING,
1843                                                LCK_CR | LCK_CW |
1844                                                LCK_PR | LCK_PW);
1845                         if (mode) {
1846                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1847                                                       LLIF_EPOCH_PENDING |
1848                                                       LLIF_SOM_DIRTY)) {
1849                                         CERROR("%s: inode "DFID" flags %u still has size authority! do not trust the size got from MDS\n",
1850                                                sbi->ll_md_exp->exp_obd->obd_name,
1851                                                PFID(ll_inode2fid(inode)),
1852                                                lli->lli_flags);
1853                                 } else {
1854                                         /* Use old size assignment to avoid
1855                                          * deadlock bz14138 & bz14326
1856                                          */
1857                                         i_size_write(inode, body->mbo_size);
1858                                         spin_lock(&lli->lli_lock);
1859                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1860                                         spin_unlock(&lli->lli_lock);
1861                                 }
1862                                 ldlm_lock_decref(&lockh, mode);
1863                         }
1864                 } else {
1865                         /* Use old size assignment to avoid
1866                          * deadlock bz14138 & bz14326
1867                          */
1868                         i_size_write(inode, body->mbo_size);
1869
1870                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1871                                inode->i_ino, (unsigned long long)body->mbo_size);
1872                 }
1873
1874                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
1875                         inode->i_blocks = body->mbo_blocks;
1876         }
1877
1878         if (body->mbo_valid & OBD_MD_TSTATE) {
1879                 if (body->mbo_t_state & MS_RESTORE)
1880                         lli->lli_flags |= LLIF_FILE_RESTORING;
1881         }
1882
1883         return 0;
1884 }
1885
1886 int ll_read_inode2(struct inode *inode, void *opaque)
1887 {
1888         struct lustre_md *md = opaque;
1889         struct ll_inode_info *lli = ll_i2info(inode);
1890         int rc;
1891
1892         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1893                PFID(&lli->lli_fid), inode);
1894
1895         LASSERT(!lli->lli_has_smd);
1896
1897         /* Core attributes from the MDS first.  This is a new inode, and
1898          * the VFS doesn't zero times in the core inode so we have to do
1899          * it ourselves.  They will be overwritten by either MDS or OST
1900          * attributes - we just need to make sure they aren't newer.
1901          */
1902         LTIME_S(inode->i_mtime) = 0;
1903         LTIME_S(inode->i_atime) = 0;
1904         LTIME_S(inode->i_ctime) = 0;
1905         inode->i_rdev = 0;
1906         rc = ll_update_inode(inode, md);
1907         if (rc)
1908                 return rc;
1909
1910         /* OIDEBUG(inode); */
1911
1912         if (S_ISREG(inode->i_mode)) {
1913                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1914
1915                 inode->i_op = &ll_file_inode_operations;
1916                 inode->i_fop = sbi->ll_fop;
1917                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1918         } else if (S_ISDIR(inode->i_mode)) {
1919                 inode->i_op = &ll_dir_inode_operations;
1920                 inode->i_fop = &ll_dir_operations;
1921         } else if (S_ISLNK(inode->i_mode)) {
1922                 inode->i_op = &ll_fast_symlink_inode_operations;
1923         } else {
1924                 inode->i_op = &ll_special_inode_operations;
1925
1926                 init_special_inode(inode, inode->i_mode,
1927                                    inode->i_rdev);
1928         }
1929
1930         return 0;
1931 }
1932
1933 void ll_delete_inode(struct inode *inode)
1934 {
1935         struct ll_inode_info *lli = ll_i2info(inode);
1936
1937         if (S_ISREG(inode->i_mode) && lli->lli_clob)
1938                 /* discard all dirty pages before truncating them, required by
1939                  * osc_extent implementation at LU-1030.
1940                  */
1941                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1942                                    CL_FSYNC_LOCAL, 1);
1943
1944         truncate_inode_pages_final(&inode->i_data);
1945
1946         LASSERTF(!inode->i_data.nrpages,
1947                  "inode=" DFID "(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1948                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
1949
1950         ll_clear_inode(inode);
1951         clear_inode(inode);
1952 }
1953
1954 int ll_iocontrol(struct inode *inode, struct file *file,
1955                  unsigned int cmd, unsigned long arg)
1956 {
1957         struct ll_sb_info *sbi = ll_i2sbi(inode);
1958         struct ptlrpc_request *req = NULL;
1959         int rc, flags = 0;
1960
1961         switch (cmd) {
1962         case FSFILT_IOC_GETFLAGS: {
1963                 struct mdt_body *body;
1964                 struct md_op_data *op_data;
1965
1966                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1967                                              0, 0, LUSTRE_OPC_ANY,
1968                                              NULL);
1969                 if (IS_ERR(op_data))
1970                         return PTR_ERR(op_data);
1971
1972                 op_data->op_valid = OBD_MD_FLFLAGS;
1973                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1974                 ll_finish_md_op_data(op_data);
1975                 if (rc) {
1976                         CERROR("%s: failure inode "DFID": rc = %d\n",
1977                                sbi->ll_md_exp->exp_obd->obd_name,
1978                                PFID(ll_inode2fid(inode)), rc);
1979                         return -abs(rc);
1980                 }
1981
1982                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1983
1984                 flags = body->mbo_flags;
1985
1986                 ptlrpc_req_finished(req);
1987
1988                 return put_user(flags, (int __user *)arg);
1989         }
1990         case FSFILT_IOC_SETFLAGS: {
1991                 struct lov_stripe_md *lsm;
1992                 struct obd_info oinfo = { };
1993                 struct md_op_data *op_data;
1994
1995                 if (get_user(flags, (int __user *)arg))
1996                         return -EFAULT;
1997
1998                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1999                                              LUSTRE_OPC_ANY, NULL);
2000                 if (IS_ERR(op_data))
2001                         return PTR_ERR(op_data);
2002
2003                 op_data->op_attr_flags = flags;
2004                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2005                 rc = md_setattr(sbi->ll_md_exp, op_data,
2006                                 NULL, 0, NULL, 0, &req, NULL);
2007                 ll_finish_md_op_data(op_data);
2008                 ptlrpc_req_finished(req);
2009                 if (rc)
2010                         return rc;
2011
2012                 inode->i_flags = ll_ext_to_inode_flags(flags);
2013
2014                 lsm = ccc_inode_lsm_get(inode);
2015                 if (!lsm_has_objects(lsm)) {
2016                         ccc_inode_lsm_put(inode, lsm);
2017                         return 0;
2018                 }
2019
2020                 oinfo.oi_oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
2021                 if (!oinfo.oi_oa) {
2022                         ccc_inode_lsm_put(inode, lsm);
2023                         return -ENOMEM;
2024                 }
2025                 oinfo.oi_md = lsm;
2026                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2027                 oinfo.oi_oa->o_flags = flags;
2028                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2029                                        OBD_MD_FLGROUP;
2030                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2031                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2032                 kmem_cache_free(obdo_cachep, oinfo.oi_oa);
2033                 ccc_inode_lsm_put(inode, lsm);
2034
2035                 if (rc && rc != -EPERM && rc != -EACCES)
2036                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2037
2038                 return rc;
2039         }
2040         default:
2041                 return -ENOSYS;
2042         }
2043
2044         return 0;
2045 }
2046
2047 int ll_flush_ctx(struct inode *inode)
2048 {
2049         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2050
2051         CDEBUG(D_SEC, "flush context for user %d\n",
2052                from_kuid(&init_user_ns, current_uid()));
2053
2054         obd_set_info_async(NULL, sbi->ll_md_exp,
2055                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2056                            0, NULL, NULL);
2057         obd_set_info_async(NULL, sbi->ll_dt_exp,
2058                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2059                            0, NULL, NULL);
2060         return 0;
2061 }
2062
2063 /* umount -f client means force down, don't save state */
2064 void ll_umount_begin(struct super_block *sb)
2065 {
2066         struct ll_sb_info *sbi = ll_s2sbi(sb);
2067         struct obd_device *obd;
2068         struct obd_ioctl_data *ioc_data;
2069
2070         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2071                sb->s_count, atomic_read(&sb->s_active));
2072
2073         obd = class_exp2obd(sbi->ll_md_exp);
2074         if (!obd) {
2075                 CERROR("Invalid MDC connection handle %#llx\n",
2076                        sbi->ll_md_exp->exp_handle.h_cookie);
2077                 return;
2078         }
2079         obd->obd_force = 1;
2080
2081         obd = class_exp2obd(sbi->ll_dt_exp);
2082         if (!obd) {
2083                 CERROR("Invalid LOV connection handle %#llx\n",
2084                        sbi->ll_dt_exp->exp_handle.h_cookie);
2085                 return;
2086         }
2087         obd->obd_force = 1;
2088
2089         ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
2090         if (ioc_data) {
2091                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2092                               sizeof(*ioc_data), ioc_data, NULL);
2093
2094                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2095                               sizeof(*ioc_data), ioc_data, NULL);
2096
2097                 kfree(ioc_data);
2098         }
2099
2100         /* Really, we'd like to wait until there are no requests outstanding,
2101          * and then continue.  For now, we just invalidate the requests,
2102          * schedule() and sleep one second if needed, and hope.
2103          */
2104         schedule();
2105 }
2106
2107 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2108 {
2109         struct ll_sb_info *sbi = ll_s2sbi(sb);
2110         char *profilenm = get_profile_name(sb);
2111         int err;
2112         __u32 read_only;
2113
2114         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2115                 read_only = *flags & MS_RDONLY;
2116                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2117                                          sizeof(KEY_READ_ONLY),
2118                                          KEY_READ_ONLY, sizeof(read_only),
2119                                          &read_only, NULL);
2120                 if (err) {
2121                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2122                                       profilenm, read_only ?
2123                                       "read-only" : "read-write", err);
2124                         return err;
2125                 }
2126
2127                 if (read_only)
2128                         sb->s_flags |= MS_RDONLY;
2129                 else
2130                         sb->s_flags &= ~MS_RDONLY;
2131
2132                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2133                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2134                                       read_only ?  "read-only" : "read-write");
2135         }
2136         return 0;
2137 }
2138
2139 /**
2140  * Cleanup the open handle that is cached on MDT-side.
2141  *
2142  * For open case, the client side open handling thread may hit error
2143  * after the MDT grant the open. Under such case, the client should
2144  * send close RPC to the MDT as cleanup; otherwise, the open handle
2145  * on the MDT will be leaked there until the client umount or evicted.
2146  *
2147  * In further, if someone unlinked the file, because the open handle
2148  * holds the reference on such file/object, then it will block the
2149  * subsequent threads that want to locate such object via FID.
2150  *
2151  * \param[in] sb        super block for this file-system
2152  * \param[in] open_req  pointer to the original open request
2153  */
2154 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2155 {
2156         struct mdt_body                 *body;
2157         struct md_op_data               *op_data;
2158         struct ptlrpc_request           *close_req = NULL;
2159         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2160
2161         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2162         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2163         if (!op_data)
2164                 return;
2165
2166         op_data->op_fid1 = body->mbo_fid1;
2167         op_data->op_ioepoch = body->mbo_ioepoch;
2168         op_data->op_handle = body->mbo_handle;
2169         op_data->op_mod_time = get_seconds();
2170         md_close(exp, op_data, NULL, &close_req);
2171         ptlrpc_req_finished(close_req);
2172         ll_finish_md_op_data(op_data);
2173 }
2174
2175 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2176                   struct super_block *sb, struct lookup_intent *it)
2177 {
2178         struct ll_sb_info *sbi = NULL;
2179         struct lustre_md md = { NULL };
2180         int rc;
2181
2182         LASSERT(*inode || sb);
2183         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2184         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2185                               sbi->ll_md_exp, &md);
2186         if (rc)
2187                 goto cleanup;
2188
2189         if (*inode) {
2190                 rc = ll_update_inode(*inode, &md);
2191                 if (rc)
2192                         goto out;
2193         } else {
2194                 LASSERT(sb);
2195
2196                 /*
2197                  * At this point server returns to client's same fid as client
2198                  * generated for creating. So using ->fid1 is okay here.
2199                  */
2200                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2201                         CERROR("%s: Fid is insane " DFID "\n",
2202                                ll_get_fsname(sb, NULL, 0),
2203                                PFID(&md.body->mbo_fid1));
2204                         rc = -EINVAL;
2205                         goto out;
2206                 }
2207
2208                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2209                                              sbi->ll_flags & LL_SBI_32BIT_API),
2210                                  &md);
2211                 if (IS_ERR(*inode)) {
2212 #ifdef CONFIG_FS_POSIX_ACL
2213                         if (md.posix_acl) {
2214                                 posix_acl_release(md.posix_acl);
2215                                 md.posix_acl = NULL;
2216                         }
2217 #endif
2218                         rc = -ENOMEM;
2219                         CERROR("new_inode -fatal: rc %d\n", rc);
2220                         goto out;
2221                 }
2222         }
2223
2224         /* Handling piggyback layout lock.
2225          * Layout lock can be piggybacked by getattr and open request.
2226          * The lsm can be applied to inode only if it comes with a layout lock
2227          * otherwise correct layout may be overwritten, for example:
2228          * 1. proc1: mdt returns a lsm but not granting layout
2229          * 2. layout was changed by another client
2230          * 3. proc2: refresh layout and layout lock granted
2231          * 4. proc1: to apply a stale layout
2232          */
2233         if (it && it->it_lock_mode != 0) {
2234                 struct lustre_handle lockh;
2235                 struct ldlm_lock *lock;
2236
2237                 lockh.cookie = it->it_lock_handle;
2238                 lock = ldlm_handle2lock(&lockh);
2239                 LASSERT(lock);
2240                 if (ldlm_has_layout(lock)) {
2241                         struct cl_object_conf conf;
2242
2243                         memset(&conf, 0, sizeof(conf));
2244                         conf.coc_opc = OBJECT_CONF_SET;
2245                         conf.coc_inode = *inode;
2246                         conf.coc_lock = lock;
2247                         conf.u.coc_md = &md;
2248                         (void)ll_layout_conf(*inode, &conf);
2249                 }
2250                 LDLM_LOCK_PUT(lock);
2251         }
2252
2253 out:
2254         if (md.lsm)
2255                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2256         md_free_lustre_md(sbi->ll_md_exp, &md);
2257
2258 cleanup:
2259         if (rc != 0 && it && it->it_op & IT_OPEN)
2260                 ll_open_cleanup(sb ? sb : (*inode)->i_sb, req);
2261
2262         return rc;
2263 }
2264
2265 int ll_obd_statfs(struct inode *inode, void __user *arg)
2266 {
2267         struct ll_sb_info *sbi = NULL;
2268         struct obd_export *exp;
2269         char *buf = NULL;
2270         struct obd_ioctl_data *data = NULL;
2271         __u32 type;
2272         int len = 0, rc;
2273
2274         if (!inode) {
2275                 rc = -EINVAL;
2276                 goto out_statfs;
2277         }
2278
2279         sbi = ll_i2sbi(inode);
2280         if (!sbi) {
2281                 rc = -EINVAL;
2282                 goto out_statfs;
2283         }
2284
2285         rc = obd_ioctl_getdata(&buf, &len, arg);
2286         if (rc)
2287                 goto out_statfs;
2288
2289         data = (void *)buf;
2290         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2291             !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2292                 rc = -EINVAL;
2293                 goto out_statfs;
2294         }
2295
2296         if (data->ioc_inllen1 != sizeof(__u32) ||
2297             data->ioc_inllen2 != sizeof(__u32) ||
2298             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2299             data->ioc_plen2 != sizeof(struct obd_uuid)) {
2300                 rc = -EINVAL;
2301                 goto out_statfs;
2302         }
2303
2304         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2305         if (type & LL_STATFS_LMV) {
2306                 exp = sbi->ll_md_exp;
2307         } else if (type & LL_STATFS_LOV) {
2308                 exp = sbi->ll_dt_exp;
2309         } else {
2310                 rc = -ENODEV;
2311                 goto out_statfs;
2312         }
2313
2314         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2315         if (rc)
2316                 goto out_statfs;
2317 out_statfs:
2318         if (buf)
2319                 obd_ioctl_freedata(buf, len);
2320         return rc;
2321 }
2322
2323 int ll_process_config(struct lustre_cfg *lcfg)
2324 {
2325         char *ptr;
2326         void *sb;
2327         struct lprocfs_static_vars lvars;
2328         unsigned long x;
2329         int rc = 0;
2330
2331         lprocfs_llite_init_vars(&lvars);
2332
2333         /* The instance name contains the sb: lustre-client-aacfe000 */
2334         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2335         if (!ptr || !*(++ptr))
2336                 return -EINVAL;
2337         rc = kstrtoul(ptr, 16, &x);
2338         if (rc != 0)
2339                 return -EINVAL;
2340         sb = (void *)x;
2341         /* This better be a real Lustre superblock! */
2342         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2343
2344         /* Note we have not called client_common_fill_super yet, so
2345          * proc fns must be able to handle that!
2346          */
2347         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2348                                       lcfg, sb);
2349         if (rc > 0)
2350                 rc = 0;
2351         return rc;
2352 }
2353
2354 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2355 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2356                                       struct inode *i1, struct inode *i2,
2357                                       const char *name, size_t namelen,
2358                                       u32 mode, __u32 opc, void *data)
2359 {
2360         if (!name) {
2361                 /* Do not reuse namelen for something else. */
2362                 if (namelen)
2363                         return ERR_PTR(-EINVAL);
2364         } else {
2365                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2366                         return ERR_PTR(-ENAMETOOLONG);
2367
2368                 if (!lu_name_is_valid_2(name, namelen))
2369                         return ERR_PTR(-EINVAL);
2370         }
2371
2372         if (!op_data)
2373                 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2374
2375         if (!op_data)
2376                 return ERR_PTR(-ENOMEM);
2377
2378         ll_i2gids(op_data->op_suppgids, i1, i2);
2379         op_data->op_fid1 = *ll_inode2fid(i1);
2380         op_data->op_default_stripe_offset = -1;
2381         if (S_ISDIR(i1->i_mode)) {
2382                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2383                 op_data->op_default_stripe_offset =
2384                         ll_i2info(i1)->lli_def_stripe_offset;
2385         }
2386
2387         if (i2) {
2388                 op_data->op_fid2 = *ll_inode2fid(i2);
2389                 if (S_ISDIR(i2->i_mode))
2390                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2391         } else {
2392                 fid_zero(&op_data->op_fid2);
2393         }
2394
2395         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2396                 op_data->op_cli_flags |= CLI_HASH64;
2397
2398         if (ll_need_32bit_api(ll_i2sbi(i1)))
2399                 op_data->op_cli_flags |= CLI_API32;
2400
2401         op_data->op_name = name;
2402         op_data->op_namelen = namelen;
2403         op_data->op_mode = mode;
2404         op_data->op_mod_time = ktime_get_real_seconds();
2405         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2406         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2407         op_data->op_cap = cfs_curproc_cap_pack();
2408         op_data->op_bias = 0;
2409         op_data->op_cli_flags = 0;
2410         if ((opc == LUSTRE_OPC_CREATE) && name &&
2411             filename_is_volatile(name, namelen, &op_data->op_mds))
2412                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2413         else
2414                 op_data->op_mds = 0;
2415         op_data->op_data = data;
2416
2417         /* When called by ll_setattr_raw, file is i1. */
2418         if (ll_i2info(i1)->lli_flags & LLIF_DATA_MODIFIED)
2419                 op_data->op_bias |= MDS_DATA_MODIFIED;
2420
2421         return op_data;
2422 }
2423
2424 void ll_finish_md_op_data(struct md_op_data *op_data)
2425 {
2426         kfree(op_data);
2427 }
2428
2429 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2430 {
2431         struct ll_sb_info *sbi;
2432
2433         LASSERT(seq && dentry);
2434         sbi = ll_s2sbi(dentry->d_sb);
2435
2436         if (sbi->ll_flags & LL_SBI_NOLCK)
2437                 seq_puts(seq, ",nolock");
2438
2439         if (sbi->ll_flags & LL_SBI_FLOCK)
2440                 seq_puts(seq, ",flock");
2441
2442         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2443                 seq_puts(seq, ",localflock");
2444
2445         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2446                 seq_puts(seq, ",user_xattr");
2447
2448         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2449                 seq_puts(seq, ",lazystatfs");
2450
2451         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2452                 seq_puts(seq, ",user_fid2path");
2453
2454         return 0;
2455 }
2456
2457 /**
2458  * Get obd name by cmd, and copy out to user space
2459  */
2460 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2461 {
2462         struct ll_sb_info *sbi = ll_i2sbi(inode);
2463         struct obd_device *obd;
2464
2465         if (cmd == OBD_IOC_GETDTNAME)
2466                 obd = class_exp2obd(sbi->ll_dt_exp);
2467         else if (cmd == OBD_IOC_GETMDNAME)
2468                 obd = class_exp2obd(sbi->ll_md_exp);
2469         else
2470                 return -EINVAL;
2471
2472         if (!obd)
2473                 return -ENOENT;
2474
2475         if (copy_to_user((void __user *)arg, obd->obd_name,
2476                          strlen(obd->obd_name) + 1))
2477                 return -EFAULT;
2478
2479         return 0;
2480 }
2481
2482 /**
2483  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2484  * fsname will be returned in this buffer; otherwise, a static buffer will be
2485  * used to store the fsname and returned to caller.
2486  */
2487 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2488 {
2489         static char fsname_static[MTI_NAME_MAXLEN];
2490         struct lustre_sb_info *lsi = s2lsi(sb);
2491         char *ptr;
2492         int len;
2493
2494         if (!buf) {
2495                 /* this means the caller wants to use static buffer
2496                  * and it doesn't care about race. Usually this is
2497                  * in error reporting path
2498                  */
2499                 buf = fsname_static;
2500                 buflen = sizeof(fsname_static);
2501         }
2502
2503         len = strlen(lsi->lsi_lmd->lmd_profile);
2504         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2505         if (ptr && (strcmp(ptr, "-client") == 0))
2506                 len -= 7;
2507
2508         if (unlikely(len >= buflen))
2509                 len = buflen - 1;
2510         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2511         buf[len] = '\0';
2512
2513         return buf;
2514 }
2515
2516 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2517 {
2518         char *buf, *path = NULL;
2519         struct dentry *dentry = NULL;
2520         struct vvp_object *obj = cl_inode2vvp(page->mapping->host);
2521
2522         /* this can be called inside spin lock so use GFP_ATOMIC. */
2523         buf = (char *)__get_free_page(GFP_ATOMIC);
2524         if (buf) {
2525                 dentry = d_find_alias(page->mapping->host);
2526                 if (dentry)
2527                         path = dentry_path_raw(dentry, buf, PAGE_SIZE);
2528         }
2529
2530         CDEBUG(D_WARNING,
2531                "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2532                ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2533                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2534                PFID(&obj->vob_header.coh_lu.loh_fid),
2535                (path && !IS_ERR(path)) ? path : "", ioret);
2536
2537         if (dentry)
2538                 dput(dentry);
2539
2540         if (buf)
2541                 free_page((unsigned long)buf);
2542 }
2543
2544 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2545                         struct lov_user_md **kbuf)
2546 {
2547         struct lov_user_md lum;
2548         ssize_t lum_size;
2549
2550         if (copy_from_user(&lum, md, sizeof(lum))) {
2551                 lum_size = -EFAULT;
2552                 goto no_kbuf;
2553         }
2554
2555         lum_size = ll_lov_user_md_size(&lum);
2556         if (lum_size < 0)
2557                 goto no_kbuf;
2558
2559         *kbuf = kzalloc(lum_size, GFP_NOFS);
2560         if (!*kbuf) {
2561                 lum_size = -ENOMEM;
2562                 goto no_kbuf;
2563         }
2564
2565         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2566                 kfree(*kbuf);
2567                 *kbuf = NULL;
2568                 lum_size = -EFAULT;
2569         }
2570 no_kbuf:
2571         return lum_size;
2572 }
2573
2574 /*
2575  * Compute llite root squash state after a change of root squash
2576  * configuration setting or add/remove of a lnet nid
2577  */
2578 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2579 {
2580         struct root_squash_info *squash = &sbi->ll_squash;
2581         lnet_process_id_t id;
2582         bool matched;
2583         int i;
2584
2585         /* Update norootsquash flag */
2586         down_write(&squash->rsi_sem);
2587         if (list_empty(&squash->rsi_nosquash_nids)) {
2588                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2589         } else {
2590                 /*
2591                  * Do not apply root squash as soon as one of our NIDs is
2592                  * in the nosquash_nids list
2593                  */
2594                 matched = false;
2595                 i = 0;
2596
2597                 while (LNetGetId(i++, &id) != -ENOENT) {
2598                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2599                                 continue;
2600                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2601                                 matched = true;
2602                                 break;
2603                         }
2604                 }
2605                 if (matched)
2606                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2607                 else
2608                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2609         }
2610         up_write(&squash->rsi_sem);
2611 }
2612
2613 /**
2614  * Parse linkea content to extract information about a given hardlink
2615  *
2616  * \param[in]   ldata           - Initialized linkea data
2617  * \param[in]   linkno          - Link identifier
2618  * \param[out]  parent_fid      - The entry's parent FID
2619  * \param[in]   size            - Entry name destination buffer
2620  *
2621  * \retval 0 on success
2622  * \retval Appropriate negative error code on failure
2623  */
2624 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2625                             struct lu_fid *parent_fid, struct lu_name *ln)
2626 {
2627         unsigned int idx;
2628         int rc;
2629
2630         rc = linkea_init(ldata);
2631         if (rc < 0)
2632                 return rc;
2633
2634         if (linkno >= ldata->ld_leh->leh_reccount)
2635                 /* beyond last link */
2636                 return -ENODATA;
2637
2638         linkea_first_entry(ldata);
2639         for (idx = 0; ldata->ld_lee; idx++) {
2640                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2641                                     parent_fid);
2642                 if (idx == linkno)
2643                         break;
2644
2645                 linkea_next_entry(ldata);
2646         }
2647
2648         if (idx < linkno)
2649                 return -ENODATA;
2650
2651         return 0;
2652 }
2653
2654 /**
2655  * Get parent FID and name of an identified link. Operation is performed for
2656  * a given link number, letting the caller iterate over linkno to list one or
2657  * all links of an entry.
2658  *
2659  * \param[in]     file  - File descriptor against which to perform the operation
2660  * \param[in,out] arg   - User-filled structure containing the linkno to operate
2661  *                        on and the available size. It is eventually filled with
2662  *                        the requested information or left untouched on error
2663  *
2664  * \retval - 0 on success
2665  * \retval - Appropriate negative error code on failure
2666  */
2667 int ll_getparent(struct file *file, struct getparent __user *arg)
2668 {
2669         struct inode *inode = file_inode(file);
2670         struct linkea_data *ldata;
2671         struct lu_fid parent_fid;
2672         struct lu_buf buf = {
2673                 .lb_buf = NULL,
2674                 .lb_len = 0
2675         };
2676         struct lu_name ln;
2677         u32 name_size;
2678         u32 linkno;
2679         int rc;
2680
2681         if (!capable(CFS_CAP_DAC_READ_SEARCH) &&
2682             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2683                 return -EPERM;
2684
2685         if (get_user(name_size, &arg->gp_name_size))
2686                 return -EFAULT;
2687
2688         if (get_user(linkno, &arg->gp_linkno))
2689                 return -EFAULT;
2690
2691         if (name_size > PATH_MAX)
2692                 return -EINVAL;
2693
2694         ldata = kzalloc(sizeof(*ldata), GFP_NOFS);
2695         if (!ldata)
2696                 return -ENOMEM;
2697
2698         rc = linkea_data_new(ldata, &buf);
2699         if (rc < 0)
2700                 goto ldata_free;
2701
2702         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2703                            buf.lb_len, OBD_MD_FLXATTR);
2704         if (rc < 0)
2705                 goto lb_free;
2706
2707         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2708         if (rc < 0)
2709                 goto lb_free;
2710
2711         if (ln.ln_namelen >= name_size) {
2712                 rc = -EOVERFLOW;
2713                 goto lb_free;
2714         }
2715
2716         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid))) {
2717                 rc = -EFAULT;
2718                 goto lb_free;
2719         }
2720
2721         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen)) {
2722                 rc = -EFAULT;
2723                 goto lb_free;
2724         }
2725
2726         if (put_user('\0', arg->gp_name + ln.ln_namelen)) {
2727                 rc = -EFAULT;
2728                 goto lb_free;
2729         }
2730
2731 lb_free:
2732         lu_buf_free(&buf);
2733 ldata_free:
2734         kfree(ldata);
2735         return rc;
2736 }