lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/llite_lib.c
  37  *
  38  * Lustre Light Super operations
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_LLITE
  42
  43 #include <linux/module.h>
  44 #include <linux/statfs.h>
  45 #include <linux/types.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48
  49 #include <lustre_ioctl.h>
  50 #include <lustre_ha.h>
  51 #include <lustre_dlm.h>
  52 #include <lprocfs_status.h>
  53 #include <lustre_disk.h>
  54 #include <lustre_param.h>
  55 #include <lustre_log.h>
  56 #include <cl_object.h>
  57 #include <obd_cksum.h>
  58 #include "llite_internal.h"
  59
  60 struct kmem_cache *ll_file_data_slab;
  61
  62 static struct list_head ll_super_blocks = LIST_HEAD_INIT(ll_super_blocks);
  63 static DEFINE_SPINLOCK(ll_sb_lock);
  64
  65 #ifndef log2
  66 #define log2(n) ffz(~(n))
  67 #endif
  68
  69 static struct ll_sb_info *ll_init_sbi(void)
  70 {
  71         struct ll_sb_info *sbi = NULL;
  72         unsigned long pages;
  73         unsigned long lru_page_max;
  74         struct sysinfo si;
  75         class_uuid_t uuid;
  76         int i;
  77         ENTRY;
  78
  79         OBD_ALLOC_PTR(sbi);
  80         if (sbi == NULL)
  81                 RETURN(NULL);
  82
  83         spin_lock_init(&sbi->ll_lock);
  84         mutex_init(&sbi->ll_lco.lco_lock);
  85         spin_lock_init(&sbi->ll_pp_extent_lock);
  86         spin_lock_init(&sbi->ll_process_lock);
  87         sbi->ll_rw_stats_on = 0;
  88
  89         si_meminfo(&si);
  90         pages = si.totalram - si.totalhigh;
  91         lru_page_max = pages / 2;
  92
  93         /* initialize ll_cache data */
  94         sbi->ll_cache = cl_cache_init(lru_page_max);
  95         if (sbi->ll_cache == NULL) {
  96                 OBD_FREE(sbi, sizeof(*sbi));
  97                 RETURN(NULL);
  98         }
  99
 100         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 101                                            SBI_DEFAULT_READAHEAD_MAX);
 102         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 103         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 104                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 105         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 106         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 107
 108         ll_generate_random_uuid(uuid);
 109         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 110         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 111
 112         spin_lock(&ll_sb_lock);
 113         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 114         spin_unlock(&ll_sb_lock);
 115
 116         sbi->ll_flags |= LL_SBI_VERBOSE;
 117 #ifdef ENABLE_CHECKSUM
 118         sbi->ll_flags |= LL_SBI_CHECKSUM;
 119 #endif
 120
 121 #ifdef HAVE_LRU_RESIZE_SUPPORT
 122         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 123 #endif
 124
 125         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 126                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 127                                pp_r_hist.oh_lock);
 128                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 129                                pp_w_hist.oh_lock);
 130         }
 131
 132         /* metadata statahead is enabled by default */
 133         sbi->ll_sa_max = LL_SA_RPC_DEF;
 134         atomic_set(&sbi->ll_sa_total, 0);
 135         atomic_set(&sbi->ll_sa_wrong, 0);
 136         atomic_set(&sbi->ll_sa_running, 0);
 137         atomic_set(&sbi->ll_agl_total, 0);
 138         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 139
 140         /* root squash */
 141         sbi->ll_squash.rsi_uid = 0;
 142         sbi->ll_squash.rsi_gid = 0;
 143         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 144         init_rwsem(&sbi->ll_squash.rsi_sem);
 145
 146         RETURN(sbi);
 147 }
 148
 149 static void ll_free_sbi(struct super_block *sb)
 150 {
 151         struct ll_sb_info *sbi = ll_s2sbi(sb);
 152         ENTRY;
 153
 154         if (sbi != NULL) {
 155                 spin_lock(&ll_sb_lock);
 156                 list_del(&sbi->ll_list);
 157                 spin_unlock(&ll_sb_lock);
 158                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 159                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 160                 if (sbi->ll_cache != NULL) {
 161                         cl_cache_decref(sbi->ll_cache);
 162                         sbi->ll_cache = NULL;
 163                 }
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 170                                     struct vfsmount *mnt)
 171 {
 172         struct inode *root = NULL;
 173         struct ll_sb_info *sbi = ll_s2sbi(sb);
 174         struct obd_device *obd;
 175         struct obd_capa *oc = NULL;
 176         struct obd_statfs *osfs = NULL;
 177         struct ptlrpc_request *request = NULL;
 178         struct obd_connect_data *data = NULL;
 179         struct obd_uuid *uuid;
 180         struct md_op_data *op_data;
 181         struct lustre_md lmd;
 182         u64 valid;
 183         int size, err, checksum;
 184         ENTRY;
 185
 186         obd = class_name2obd(md);
 187         if (!obd) {
 188                 CERROR("MD %s: not setup or attached\n", md);
 189                 RETURN(-EINVAL);
 190         }
 191
 192         OBD_ALLOC_PTR(data);
 193         if (data == NULL)
 194                 RETURN(-ENOMEM);
 195
 196         OBD_ALLOC_PTR(osfs);
 197         if (osfs == NULL) {
 198                 OBD_FREE_PTR(data);
 199                 RETURN(-ENOMEM);
 200         }
 201
 202         /* indicate the features supported by this client */
 203         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 204                                   OBD_CONNECT_ATTRFID  |
 205                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 206                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 207                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 208                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 209                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 210                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 211                                   OBD_CONNECT_EINPROGRESS |
 212                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 213                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 214                                   OBD_CONNECT_MAX_EASIZE |
 215                                   OBD_CONNECT_FLOCK_DEAD |
 216                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 217                                   OBD_CONNECT_OPEN_BY_FID |
 218                                   OBD_CONNECT_DIR_STRIPE;
 219
 220         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 221                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 222
 223 #ifdef HAVE_LRU_RESIZE_SUPPORT
 224         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 225                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 226 #endif
 227 #ifdef CONFIG_FS_POSIX_ACL
 228         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 229 #endif
 230
 231         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 232                 /* flag mdc connection as lightweight, only used for test
 233                  * purpose, use with care */
 234                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 235
 236         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 237         data->ocd_version = LUSTRE_VERSION_CODE;
 238
 239         if (sb->s_flags & MS_RDONLY)
 240                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 241         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 242                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 243
 244 #ifdef HAVE_MS_FLOCK_LOCK
 245         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 246         sb->s_flags |= MS_FLOCK_LOCK;
 247 #endif
 248 #ifdef MS_HAS_NEW_AOPS
 249         sb->s_flags |= MS_HAS_NEW_AOPS;
 250 #endif
 251
 252         if (sbi->ll_flags & LL_SBI_FLOCK)
 253                 sbi->ll_fop = &ll_file_operations_flock;
 254         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 255                 sbi->ll_fop = &ll_file_operations;
 256         else
 257                 sbi->ll_fop = &ll_file_operations_noflock;
 258
 259         /* real client */
 260         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 261         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 262                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 263
 264         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 265
 266         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 267         if (err == -EBUSY) {
 268                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 269                                    "recovery, of which this client is not a "
 270                                    "part. Please wait for recovery to complete,"
 271                                    " abort, or time out.\n", md);
 272                 GOTO(out, err);
 273         } else if (err) {
 274                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 275                 GOTO(out, err);
 276         }
 277
 278         sbi->ll_md_exp->exp_connect_data = *data;
 279
 280         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 281                            LUSTRE_SEQ_METADATA);
 282         if (err) {
 283                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 284                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 285                 GOTO(out_md, err);
 286         }
 287
 288         /* For mount, we only need fs info from MDT0, and also in DNE, it
 289          * can make sure the client can be mounted as long as MDT0 is
 290          * avaible */
 291         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 292                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 293                         OBD_STATFS_FOR_MDT0);
 294         if (err)
 295                 GOTO(out_md_fid, err);
 296
 297         /* This needs to be after statfs to ensure connect has finished.
 298          * Note that "data" does NOT contain the valid connect reply.
 299          * If connecting to a 1.8 server there will be no LMV device, so
 300          * we can access the MDC export directly and exp_connect_flags will
 301          * be non-zero, but if accessing an upgraded 2.1 server it will
 302          * have the correct flags filled in.
 303          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 304         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 305         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 306             valid != CLIENT_CONNECT_MDT_REQD) {
 307                 char *buf;
 308
 309                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 310                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 311                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 312                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 313                                    "feature(s) needed for correct operation "
 314                                    "of this client (%s). Please upgrade "
 315                                    "server or downgrade client.\n",
 316                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 317                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 318                 GOTO(out_md_fid, err = -EPROTO);
 319         }
 320
 321         size = sizeof(*data);
 322         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 323                            KEY_CONN_DATA,  &size, data, NULL);
 324         if (err) {
 325                 CERROR("%s: Get connect data failed: rc = %d\n",
 326                        sbi->ll_md_exp->exp_obd->obd_name, err);
 327                 GOTO(out_md_fid, err);
 328         }
 329
 330         LASSERT(osfs->os_bsize);
 331         sb->s_blocksize = osfs->os_bsize;
 332         sb->s_blocksize_bits = log2(osfs->os_bsize);
 333         sb->s_magic = LL_SUPER_MAGIC;
 334         sb->s_maxbytes = MAX_LFS_FILESIZE;
 335         sbi->ll_namelen = osfs->os_namelen;
 336         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 337
 338         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 339             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 340                 LCONSOLE_INFO("Disabling user_xattr feature because "
 341                               "it is not supported on the server\n");
 342                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 343         }
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 346 #ifdef MS_POSIXACL
 347                 sb->s_flags |= MS_POSIXACL;
 348 #endif
 349                 sbi->ll_flags |= LL_SBI_ACL;
 350         } else {
 351                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 352 #ifdef MS_POSIXACL
 353                 sb->s_flags &= ~MS_POSIXACL;
 354 #endif
 355                 sbi->ll_flags &= ~LL_SBI_ACL;
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 359                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 360                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 361                         LCONSOLE_INFO("client is set as remote by default.\n");
 362                 }
 363         } else {
 364                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 365                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 366                         LCONSOLE_INFO("client claims to be remote, but server "
 367                                       "rejected, forced to be local.\n");
 368                 }
 369         }
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 372                 LCONSOLE_INFO("client enabled MDS capability!\n");
 373                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 374         }
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 377                 LCONSOLE_INFO("client enabled OSS capability!\n");
 378                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 379         }
 380
 381         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 382                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 383
 384         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 385                 sbi->ll_md_brw_pages = data->ocd_brw_size >> PAGE_CACHE_SHIFT;
 386         else
 387                 sbi->ll_md_brw_pages = 1;
 388
 389         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 390                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 391
 392         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 393                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 394                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 395                                       "unknown maximum xattr size.\n", dt);
 396                 } else {
 397                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 398                         sbi->ll_xattr_cache_enabled = 1;
 399                 }
 400         }
 401
 402         obd = class_name2obd(dt);
 403         if (!obd) {
 404                 CERROR("DT %s: not setup or attached\n", dt);
 405                 GOTO(out_md_fid, err = -ENODEV);
 406         }
 407
 408         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 409                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 410                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 411                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 412                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 413                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 414                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 415                                   OBD_CONNECT_MAXBYTES |
 416                                   OBD_CONNECT_EINPROGRESS |
 417                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 418                                   OBD_CONNECT_LAYOUTLOCK |
 419                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK;
 420
 421         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 422                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 423
 424         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 425                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 426                  * disabled by default, because it can still be enabled on the
 427                  * fly via /proc. As a consequence, we still need to come to an
 428                  * agreement on the supported algorithms at connect time */
 429                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 430
 431                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 432                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 433                 else
 434                         data->ocd_cksum_types = cksum_types_supported_client();
 435         }
 436
 437 #ifdef HAVE_LRU_RESIZE_SUPPORT
 438         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 439 #endif
 440         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 441                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 442
 443         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 444                "ocd_grant: %d\n", data->ocd_connect_flags,
 445                data->ocd_version, data->ocd_grant);
 446
 447         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 448         obd->obd_upcall.onu_upcall = cl_ocd_update;
 449
 450         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 451
 452         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 453                           NULL);
 454         if (err == -EBUSY) {
 455                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 456                                    "recovery, of which this client is not a "
 457                                    "part.  Please wait for recovery to "
 458                                    "complete, abort, or time out.\n", dt);
 459                 GOTO(out_md, err);
 460         } else if (err) {
 461                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 462                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 463                 GOTO(out_md, err);
 464         }
 465
 466         sbi->ll_dt_exp->exp_connect_data = *data;
 467
 468         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 469                            LUSTRE_SEQ_METADATA);
 470         if (err) {
 471                 CERROR("%s: Can't init data layer FID infrastructure, "
 472                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 473                 GOTO(out_dt, err);
 474         }
 475
 476         mutex_lock(&sbi->ll_lco.lco_lock);
 477         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 478         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 479         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 480         mutex_unlock(&sbi->ll_lco.lco_lock);
 481
 482         fid_zero(&sbi->ll_root_fid);
 483         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 484         if (err) {
 485                 CERROR("cannot mds_connect: rc = %d\n", err);
 486                 GOTO(out_lock_cn_cb, err);
 487         }
 488         if (!fid_is_sane(&sbi->ll_root_fid)) {
 489                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 490                        sbi->ll_md_exp->exp_obd->obd_name,
 491                        PFID(&sbi->ll_root_fid));
 492                 GOTO(out_lock_cn_cb, err = -EINVAL);
 493         }
 494         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 495
 496         sb->s_op = &lustre_super_operations;
 497 #if THREAD_SIZE >= 8192 /*b=17630*/
 498         sb->s_export_op = &lustre_export_operations;
 499 #endif
 500
 501         /* make root inode
 502          * XXX: move this to after cbd setup? */
 503         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 504                 OBD_MD_FLMODEASIZE;
 505         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 506                 valid |= OBD_MD_FLRMTPERM;
 507         else if (sbi->ll_flags & LL_SBI_ACL)
 508                 valid |= OBD_MD_FLACL;
 509
 510         OBD_ALLOC_PTR(op_data);
 511         if (op_data == NULL)
 512                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 513
 514         op_data->op_fid1 = sbi->ll_root_fid;
 515         op_data->op_mode = 0;
 516         op_data->op_capa1 = oc;
 517         op_data->op_valid = valid;
 518
 519         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 520         if (oc)
 521                 capa_put(oc);
 522         OBD_FREE_PTR(op_data);
 523         if (err) {
 524                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 525                        sbi->ll_md_exp->exp_obd->obd_name, err);
 526                 GOTO(out_lock_cn_cb, err);
 527         }
 528
 529         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 530                                sbi->ll_md_exp, &lmd);
 531         if (err) {
 532                 CERROR("failed to understand root inode md: rc = %d\n", err);
 533                 ptlrpc_req_finished(request);
 534                 GOTO(out_lock_cn_cb, err);
 535         }
 536
 537         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 538         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 539                                             sbi->ll_flags & LL_SBI_32BIT_API),
 540                        &lmd);
 541         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 542         ptlrpc_req_finished(request);
 543
 544         if (IS_ERR(root)) {
 545                 if (lmd.lsm)
 546                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 547 #ifdef CONFIG_FS_POSIX_ACL
 548                 if (lmd.posix_acl) {
 549                         posix_acl_release(lmd.posix_acl);
 550                         lmd.posix_acl = NULL;
 551                 }
 552 #endif
 553                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 554                 root = NULL;
 555                 CERROR("lustre_lite: bad iget4 for root\n");
 556                 GOTO(out_root, err);
 557         }
 558
 559         err = ll_close_thread_start(&sbi->ll_lcq);
 560         if (err) {
 561                 CERROR("cannot start close thread: rc %d\n", err);
 562                 GOTO(out_root, err);
 563         }
 564
 565 #ifdef CONFIG_FS_POSIX_ACL
 566         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 567                 rct_init(&sbi->ll_rct);
 568                 et_init(&sbi->ll_et);
 569         }
 570 #endif
 571
 572         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 573         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 574                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 575                                  NULL);
 576         cl_sb_init(sb);
 577
 578         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 579                                  KEY_CACHE_SET, sizeof(*sbi->ll_cache),
 580                                  sbi->ll_cache, NULL);
 581
 582         sb->s_root = d_make_root(root);
 583         if (sb->s_root == NULL) {
 584                 CERROR("%s: can't make root dentry\n",
 585                         ll_get_fsname(sb, NULL, 0));
 586                 GOTO(out_root, err = -ENOMEM);
 587         }
 588 #ifdef HAVE_DCACHE_LOCK
 589         sb->s_root->d_op = &ll_d_ops;
 590 #endif
 591
 592         sbi->ll_sdev_orig = sb->s_dev;
 593
 594         /* We set sb->s_dev equal on all lustre clients in order to support
 595          * NFS export clustering.  NFSD requires that the FSID be the same
 596          * on all clients. */
 597         /* s_dev is also used in lt_compare() to compare two fs, but that is
 598          * only a node-local comparison. */
 599         uuid = obd_get_uuid(sbi->ll_md_exp);
 600         if (uuid != NULL)
 601                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 602
 603         if (data != NULL)
 604                 OBD_FREE_PTR(data);
 605         if (osfs != NULL)
 606                 OBD_FREE_PTR(osfs);
 607         if (proc_lustre_fs_root != NULL) {
 608                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 609                                                   dt, md);
 610                 if (err < 0) {
 611                         CERROR("%s: could not register mount in lprocfs: "
 612                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 613                         err = 0;
 614                 }
 615         }
 616
 617         RETURN(err);
 618 out_root:
 619         if (root)
 620                 iput(root);
 621 out_lock_cn_cb:
 622         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 623 out_dt:
 624         obd_disconnect(sbi->ll_dt_exp);
 625         sbi->ll_dt_exp = NULL;
 626 out_md_fid:
 627         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 628 out_md:
 629         obd_disconnect(sbi->ll_md_exp);
 630         sbi->ll_md_exp = NULL;
 631 out:
 632         if (data != NULL)
 633                 OBD_FREE_PTR(data);
 634         if (osfs != NULL)
 635                 OBD_FREE_PTR(osfs);
 636         return err;
 637 }
 638
 639 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 640 {
 641         int size, rc;
 642
 643         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 644         size = sizeof(int);
 645         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 646                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 647         if (rc)
 648                 CERROR("Get max mdsize error rc %d\n", rc);
 649
 650         RETURN(rc);
 651 }
 652
 653 /**
 654  * Get the value of the default_easize parameter.
 655  *
 656  * \see client_obd::cl_default_mds_easize
 657  *
 658  * \param[in] sbi       superblock info for this filesystem
 659  * \param[out] lmmsize  pointer to storage location for value
 660  *
 661  * \retval 0            on success
 662  * \retval negative     negated errno on failure
 663  */
 664 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 665 {
 666         int size, rc;
 667
 668         size = sizeof(int);
 669         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 670                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 671         if (rc)
 672                 CERROR("Get default mdsize error rc %d\n", rc);
 673
 674         RETURN(rc);
 675 }
 676
 677 /**
 678  * Set the default_easize parameter to the given value.
 679  *
 680  * \see client_obd::cl_default_mds_easize
 681  *
 682  * \param[in] sbi       superblock info for this filesystem
 683  * \param[in] lmmsize   the size to set
 684  *
 685  * \retval 0            on success
 686  * \retval negative     negated errno on failure
 687  */
 688 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 689 {
 690         int rc;
 691
 692         if (lmmsize < sizeof(struct lov_mds_md) ||
 693             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 694                 return -EINVAL;
 695
 696         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 697                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 698                                 sizeof(int), &lmmsize, NULL);
 699
 700         RETURN(rc);
 701 }
 702
 703 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 704 {
 705         int size, rc;
 706
 707         size = sizeof(int);
 708         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 709                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 710         if (rc)
 711                 CERROR("Get max cookiesize error rc %d\n", rc);
 712
 713         RETURN(rc);
 714 }
 715
 716 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 717 {
 718         int size, rc;
 719
 720         size = sizeof(int);
 721         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 722                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 723         if (rc)
 724                 CERROR("Get default cookiesize error rc %d\n", rc);
 725
 726         RETURN(rc);
 727 }
 728
 729 static void ll_dump_inode(struct inode *inode)
 730 {
 731         struct ll_d_hlist_node *tmp;
 732         int dentry_count = 0;
 733
 734         LASSERT(inode != NULL);
 735
 736         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 737                 dentry_count++;
 738
 739         CERROR("%s: inode %p dump: dev=%s fid="DFID
 740                " mode=%o count=%u, %d dentries\n",
 741                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 742                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 743                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 744 }
 745
 746 void lustre_dump_dentry(struct dentry *dentry, int recur)
 747 {
 748         struct list_head *tmp;
 749         int subdirs = 0;
 750
 751         LASSERT(dentry != NULL);
 752
 753         list_for_each(tmp, &dentry->d_subdirs)
 754                 subdirs++;
 755
 756         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 757                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 758                dentry->d_name.len, dentry->d_name.name,
 759                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 760                dentry->d_parent, dentry->d_inode, ll_d_count(dentry),
 761                dentry->d_flags, dentry->d_fsdata, subdirs);
 762         if (dentry->d_inode != NULL)
 763                 ll_dump_inode(dentry->d_inode);
 764
 765         if (recur == 0)
 766                 return;
 767
 768         list_for_each(tmp, &dentry->d_subdirs) {
 769                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 770                 lustre_dump_dentry(d, recur - 1);
 771         }
 772 }
 773
 774 static void client_common_put_super(struct super_block *sb)
 775 {
 776         struct ll_sb_info *sbi = ll_s2sbi(sb);
 777         ENTRY;
 778
 779 #ifdef CONFIG_FS_POSIX_ACL
 780         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 781                 et_fini(&sbi->ll_et);
 782                 rct_fini(&sbi->ll_rct);
 783         }
 784 #endif
 785
 786         ll_close_thread_shutdown(sbi->ll_lcq);
 787
 788         cl_sb_fini(sb);
 789
 790         list_del(&sbi->ll_conn_chain);
 791
 792         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 793         obd_disconnect(sbi->ll_dt_exp);
 794         sbi->ll_dt_exp = NULL;
 795
 796         lprocfs_unregister_mountpoint(sbi);
 797
 798         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 799         obd_disconnect(sbi->ll_md_exp);
 800         sbi->ll_md_exp = NULL;
 801
 802         EXIT;
 803 }
 804
 805 void ll_kill_super(struct super_block *sb)
 806 {
 807         struct ll_sb_info *sbi;
 808         ENTRY;
 809
 810         /* not init sb ?*/
 811         if (!(sb->s_flags & MS_ACTIVE))
 812                 return;
 813
 814         sbi = ll_s2sbi(sb);
 815         /* we need restore s_dev from changed for clustred NFS before put_super
 816          * because new kernels have cached s_dev and change sb->s_dev in
 817          * put_super not affected real removing devices */
 818         if (sbi) {
 819                 sb->s_dev = sbi->ll_sdev_orig;
 820                 sbi->ll_umounting = 1;
 821
 822                 /* wait running statahead threads to quit */
 823                 while (atomic_read(&sbi->ll_sa_running) > 0)
 824                         schedule_timeout_and_set_state(TASK_UNINTERRUPTIBLE,
 825                                 msecs_to_jiffies(MSEC_PER_SEC >> 3));
 826         }
 827
 828         EXIT;
 829 }
 830
 831 static inline int ll_set_opt(const char *opt, char *data, int fl)
 832 {
 833         if (strncmp(opt, data, strlen(opt)) != 0)
 834                 return(0);
 835         else
 836                 return(fl);
 837 }
 838
 839 /* non-client-specific mount options are parsed in lmd_parse */
 840 static int ll_options(char *options, int *flags)
 841 {
 842         int tmp;
 843         char *s1 = options, *s2;
 844         ENTRY;
 845
 846         if (!options)
 847                 RETURN(0);
 848
 849         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 850
 851         while (*s1) {
 852                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 853                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 854                 if (tmp) {
 855                         *flags |= tmp;
 856                         goto next;
 857                 }
 858                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 859                 if (tmp) {
 860                         *flags |= tmp;
 861                         goto next;
 862                 }
 863                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 864                 if (tmp) {
 865                         *flags |= tmp;
 866                         goto next;
 867                 }
 868                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 869                 if (tmp) {
 870                         *flags &= ~tmp;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 874                 if (tmp) {
 875                         *flags |= tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 879                 if (tmp) {
 880                         *flags &= ~tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 884                 if (tmp) {
 885                         *flags |= tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 889                 if (tmp) {
 890                         *flags |= tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 894                 if (tmp) {
 895                         *flags &= ~tmp;
 896                         goto next;
 897                 }
 898
 899                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 900                 if (tmp) {
 901                         *flags |= tmp;
 902                         goto next;
 903                 }
 904                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 905                 if (tmp) {
 906                         *flags &= ~tmp;
 907                         goto next;
 908                 }
 909                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 910                 if (tmp) {
 911                         *flags |= tmp;
 912                         goto next;
 913                 }
 914                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 915                 if (tmp) {
 916                         *flags &= ~tmp;
 917                         goto next;
 918                 }
 919                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 920                 if (tmp) {
 921                         *flags |= tmp;
 922                         goto next;
 923                 }
 924                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 925                 if (tmp) {
 926                         *flags &= ~tmp;
 927                         goto next;
 928                 }
 929                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 930                 if (tmp) {
 931                         *flags |= tmp;
 932                         goto next;
 933                 }
 934                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 935                 if (tmp) {
 936                         *flags |= tmp;
 937                         goto next;
 938                 }
 939                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 940                 if (tmp) {
 941                         *flags |= tmp;
 942                         goto next;
 943                 }
 944                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 945                 if (tmp) {
 946                         *flags &= ~tmp;
 947                         goto next;
 948                 }
 949                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 950                                    s1);
 951                 RETURN(-EINVAL);
 952
 953 next:
 954                 /* Find next opt */
 955                 s2 = strchr(s1, ',');
 956                 if (s2 == NULL)
 957                         break;
 958                 s1 = s2 + 1;
 959         }
 960         RETURN(0);
 961 }
 962
 963 void ll_lli_init(struct ll_inode_info *lli)
 964 {
 965         lli->lli_inode_magic = LLI_INODE_MAGIC;
 966         lli->lli_flags = 0;
 967         lli->lli_ioepoch = 0;
 968         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 969         spin_lock_init(&lli->lli_lock);
 970         lli->lli_posix_acl = NULL;
 971         lli->lli_remote_perms = NULL;
 972         mutex_init(&lli->lli_rmtperm_mutex);
 973         /* Do not set lli_fid, it has been initialized already. */
 974         fid_zero(&lli->lli_pfid);
 975         INIT_LIST_HEAD(&lli->lli_close_list);
 976         INIT_LIST_HEAD(&lli->lli_oss_capas);
 977         atomic_set(&lli->lli_open_count, 0);
 978         lli->lli_mds_capa = NULL;
 979         lli->lli_rmtperm_time = 0;
 980         lli->lli_pending_och = NULL;
 981         lli->lli_mds_read_och = NULL;
 982         lli->lli_mds_write_och = NULL;
 983         lli->lli_mds_exec_och = NULL;
 984         lli->lli_open_fd_read_count = 0;
 985         lli->lli_open_fd_write_count = 0;
 986         lli->lli_open_fd_exec_count = 0;
 987         mutex_init(&lli->lli_och_mutex);
 988         spin_lock_init(&lli->lli_agl_lock);
 989         lli->lli_has_smd = false;
 990         spin_lock_init(&lli->lli_layout_lock);
 991         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 992         lli->lli_clob = NULL;
 993
 994         init_rwsem(&lli->lli_xattrs_list_rwsem);
 995         mutex_init(&lli->lli_xattrs_enq_lock);
 996
 997         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 998         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 999                 mutex_init(&lli->lli_readdir_mutex);
1000                 lli->lli_opendir_key = NULL;
1001                 lli->lli_sai = NULL;
1002                 spin_lock_init(&lli->lli_sa_lock);
1003                 lli->lli_opendir_pid = 0;
1004                 lli->lli_sa_enabled = 0;
1005                 lli->lli_def_stripe_offset = -1;
1006         } else {
1007                 mutex_init(&lli->lli_size_mutex);
1008                 lli->lli_symlink_name = NULL;
1009                 init_rwsem(&lli->lli_trunc_sem);
1010                 range_lock_tree_init(&lli->lli_write_tree);
1011                 init_rwsem(&lli->lli_glimpse_sem);
1012                 lli->lli_glimpse_time = 0;
1013                 INIT_LIST_HEAD(&lli->lli_agl_list);
1014                 lli->lli_agl_index = 0;
1015                 lli->lli_async_rc = 0;
1016         }
1017         mutex_init(&lli->lli_layout_mutex);
1018 }
1019
1020 static inline int ll_bdi_register(struct backing_dev_info *bdi)
1021 {
1022         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
1023
1024         bdi->name = "lustre";
1025         return bdi_register(bdi, NULL, "lustre-%d",
1026                             atomic_inc_return(&ll_bdi_num));
1027 }
1028
1029 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1030 {
1031         struct lustre_profile *lprof = NULL;
1032         struct lustre_sb_info *lsi = s2lsi(sb);
1033         struct ll_sb_info *sbi;
1034         char  *dt = NULL, *md = NULL;
1035         char  *profilenm = get_profile_name(sb);
1036         struct config_llog_instance *cfg;
1037         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1038         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1039         int    err;
1040         ENTRY;
1041
1042         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1043
1044         OBD_ALLOC_PTR(cfg);
1045         if (cfg == NULL)
1046                 RETURN(-ENOMEM);
1047
1048         try_module_get(THIS_MODULE);
1049
1050         /* client additional sb info */
1051         lsi->lsi_llsbi = sbi = ll_init_sbi();
1052         if (!sbi) {
1053                 module_put(THIS_MODULE);
1054                 OBD_FREE_PTR(cfg);
1055                 RETURN(-ENOMEM);
1056         }
1057
1058         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1059         if (err)
1060                 GOTO(out_free, err);
1061
1062         err = bdi_init(&lsi->lsi_bdi);
1063         if (err)
1064                 GOTO(out_free, err);
1065         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1066         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1067         err = ll_bdi_register(&lsi->lsi_bdi);
1068         if (err)
1069                 GOTO(out_free, err);
1070
1071         sb->s_bdi = &lsi->lsi_bdi;
1072 #ifndef HAVE_DCACHE_LOCK
1073         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1074         sb->s_d_op = &ll_d_ops;
1075 #endif
1076
1077         /* Generate a string unique to this super, in case some joker tries
1078            to mount the same fs at two mount points.
1079            Use the address of the super itself.*/
1080         cfg->cfg_instance = sb;
1081         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1082         cfg->cfg_callback = class_config_llog_handler;
1083         /* set up client obds */
1084         err = lustre_process_log(sb, profilenm, cfg);
1085         if (err < 0)
1086                 GOTO(out_free, err);
1087
1088         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1089         lprof = class_get_profile(profilenm);
1090         if (lprof == NULL) {
1091                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1092                                    " read from the MGS.  Does that filesystem "
1093                                    "exist?\n", profilenm);
1094                 GOTO(out_free, err = -EINVAL);
1095         }
1096         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1097                lprof->lp_md, lprof->lp_dt);
1098
1099         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1100         if (!dt)
1101                 GOTO(out_free, err = -ENOMEM);
1102         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1103
1104         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1105         if (!md)
1106                 GOTO(out_free, err = -ENOMEM);
1107         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1108
1109         /* connections, registrations, sb setup */
1110         err = client_common_fill_super(sb, md, dt, mnt);
1111
1112 out_free:
1113         if (md)
1114                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1115         if (dt)
1116                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1117         if (err)
1118                 ll_put_super(sb);
1119         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1120                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1121
1122         OBD_FREE_PTR(cfg);
1123         RETURN(err);
1124 } /* ll_fill_super */
1125
1126 void ll_put_super(struct super_block *sb)
1127 {
1128         struct config_llog_instance cfg, params_cfg;
1129         struct obd_device *obd;
1130         struct lustre_sb_info *lsi = s2lsi(sb);
1131         struct ll_sb_info *sbi = ll_s2sbi(sb);
1132         char *profilenm = get_profile_name(sb);
1133         long ccc_count;
1134         int next, force = 1, rc = 0;
1135         ENTRY;
1136
1137         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1138
1139         ll_print_capa_stat(sbi);
1140
1141         cfg.cfg_instance = sb;
1142         lustre_end_log(sb, profilenm, &cfg);
1143
1144         params_cfg.cfg_instance = sb;
1145         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1146
1147         if (sbi->ll_md_exp) {
1148                 obd = class_exp2obd(sbi->ll_md_exp);
1149                 if (obd)
1150                         force = obd->obd_force;
1151         }
1152
1153         /* Wait for unstable pages to be committed to stable storage */
1154         if (force == 0) {
1155                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1156                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1157                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1158                         &lwi);
1159         }
1160
1161         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1162         if (force == 0 && rc != -EINTR)
1163                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1164
1165
1166         /* We need to set force before the lov_disconnect in
1167            lustre_common_put_super, since l_d cleans up osc's as well. */
1168         if (force) {
1169                 next = 0;
1170                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1171                                                      &next)) != NULL) {
1172                         obd->obd_force = force;
1173                 }
1174         }
1175
1176         if (sbi->ll_lcq) {
1177                 /* Only if client_common_fill_super succeeded */
1178                 client_common_put_super(sb);
1179         }
1180
1181         next = 0;
1182         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1183                 class_manual_cleanup(obd);
1184         }
1185
1186         if (sbi->ll_flags & LL_SBI_VERBOSE)
1187                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1188
1189         if (profilenm)
1190                 class_del_profile(profilenm);
1191
1192         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1193                 bdi_destroy(&lsi->lsi_bdi);
1194                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1195         }
1196
1197         ll_free_sbi(sb);
1198         lsi->lsi_llsbi = NULL;
1199
1200         lustre_common_put_super(sb);
1201
1202         cl_env_cache_purge(~0);
1203
1204         module_put(THIS_MODULE);
1205
1206         EXIT;
1207 } /* client_put_super */
1208
1209 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1210 {
1211         struct inode *inode = NULL;
1212
1213         /* NOTE: we depend on atomic igrab() -bzzz */
1214         lock_res_and_lock(lock);
1215         if (lock->l_resource->lr_lvb_inode) {
1216                 struct ll_inode_info * lli;
1217                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1218                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1219                         inode = igrab(lock->l_resource->lr_lvb_inode);
1220                 } else {
1221                         inode = lock->l_resource->lr_lvb_inode;
1222                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1223                                          D_WARNING, lock, "lr_lvb_inode %p is "
1224                                          "bogus: magic %08x",
1225                                          lock->l_resource->lr_lvb_inode,
1226                                          lli->lli_inode_magic);
1227                         inode = NULL;
1228                 }
1229         }
1230         unlock_res_and_lock(lock);
1231         return inode;
1232 }
1233
1234 static void ll_dir_clear_lsm_md(struct inode *inode)
1235 {
1236         struct ll_inode_info *lli = ll_i2info(inode);
1237
1238         LASSERT(S_ISDIR(inode->i_mode));
1239
1240         if (lli->lli_lsm_md != NULL) {
1241                 lmv_free_memmd(lli->lli_lsm_md);
1242                 lli->lli_lsm_md = NULL;
1243         }
1244 }
1245
1246 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1247                                       const struct lu_fid *fid,
1248                                       struct lustre_md *md)
1249 {
1250         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1251         struct mdt_body         *body = md->body;
1252         struct inode            *inode;
1253         ino_t                   ino;
1254         ENTRY;
1255
1256         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1257         inode = iget_locked(sb, ino);
1258         if (inode == NULL) {
1259                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1260                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1261                 RETURN(ERR_PTR(-ENOENT));
1262         }
1263
1264         if (inode->i_state & I_NEW) {
1265                 struct ll_inode_info *lli = ll_i2info(inode);
1266                 struct lmv_stripe_md *lsm = md->lmv;
1267
1268                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1269                                 (body->mbo_mode & S_IFMT);
1270                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1271                          PFID(fid));
1272
1273                 LTIME_S(inode->i_mtime) = 0;
1274                 LTIME_S(inode->i_atime) = 0;
1275                 LTIME_S(inode->i_ctime) = 0;
1276                 inode->i_rdev = 0;
1277
1278                 /* initializing backing dev info. */
1279                 inode->i_mapping->backing_dev_info =
1280                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1281                 inode->i_op = &ll_dir_inode_operations;
1282                 inode->i_fop = &ll_dir_operations;
1283                 lli->lli_fid = *fid;
1284                 ll_lli_init(lli);
1285
1286                 LASSERT(lsm != NULL);
1287                 /* master object FID */
1288                 lli->lli_pfid = body->mbo_fid1;
1289                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1290                        lli, PFID(fid), PFID(&lli->lli_pfid));
1291                 unlock_new_inode(inode);
1292         }
1293
1294         RETURN(inode);
1295 }
1296
1297 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1298 {
1299         struct lu_fid *fid;
1300         struct lmv_stripe_md *lsm = md->lmv;
1301         int i;
1302
1303         LASSERT(lsm != NULL);
1304         /* XXX sigh, this lsm_root initialization should be in
1305          * LMV layer, but it needs ll_iget right now, so we
1306          * put this here right now. */
1307         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1308                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1309                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1310                 /* Unfortunately ll_iget will call ll_update_inode,
1311                  * where the initialization of slave inode is slightly
1312                  * different, so it reset lsm_md to NULL to avoid
1313                  * initializing lsm for slave inode. */
1314                 /* For migrating inode, master stripe and master object will
1315                  * be same, so we only need assign this inode */
1316                 if (lsm->lsm_md_hash_type & LMV_HASH_FLAG_MIGRATION && i == 0)
1317                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1318                 else
1319                         lsm->lsm_md_oinfo[i].lmo_root =
1320                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1321
1322                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1323                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1324
1325                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1326                         return rc;
1327                 }
1328         }
1329
1330         return 0;
1331 }
1332
1333 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1334                                 const struct lmv_stripe_md *lsm_md2)
1335 {
1336         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1337                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1338                lsm_md1->lsm_md_master_mdt_index ==
1339                                         lsm_md2->lsm_md_master_mdt_index &&
1340                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1341                lsm_md1->lsm_md_layout_version ==
1342                                         lsm_md2->lsm_md_layout_version &&
1343                strcmp(lsm_md1->lsm_md_pool_name,
1344                       lsm_md2->lsm_md_pool_name) == 0;
1345 }
1346
1347 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1348 {
1349         struct ll_inode_info *lli = ll_i2info(inode);
1350         struct lmv_stripe_md *lsm = md->lmv;
1351         int     rc;
1352         ENTRY;
1353
1354         LASSERT(S_ISDIR(inode->i_mode));
1355         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1356                PFID(ll_inode2fid(inode)));
1357
1358         /* no striped information from request. */
1359         if (lsm == NULL) {
1360                 if (lli->lli_lsm_md == NULL) {
1361                         RETURN(0);
1362                 } else if (lli->lli_lsm_md->lsm_md_hash_type &
1363                                                 LMV_HASH_FLAG_MIGRATION) {
1364                         /* migration is done, the temporay MIGRATE layout has
1365                          * been removed */
1366                         CDEBUG(D_INODE, DFID" finish migration.\n",
1367                                PFID(ll_inode2fid(inode)));
1368                         lmv_free_memmd(lli->lli_lsm_md);
1369                         lli->lli_lsm_md = NULL;
1370                         RETURN(0);
1371                 } else {
1372                         /* The lustre_md from req does not include stripeEA,
1373                          * see ll_md_setattr */
1374                         RETURN(0);
1375                 }
1376         }
1377
1378         /* set the directory layout */
1379         if (lli->lli_lsm_md == NULL) {
1380
1381                 rc = ll_init_lsm_md(inode, md);
1382                 if (rc != 0)
1383                         RETURN(rc);
1384
1385                 lli->lli_lsm_md = lsm;
1386                 /* set lsm_md to NULL, so the following free lustre_md
1387                  * will not free this lsm */
1388                 md->lmv = NULL;
1389                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1390                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1391                 RETURN(0);
1392         }
1393
1394         /* Compare the old and new stripe information */
1395         if (!lsm_md_eq(lli->lli_lsm_md, lsm)) {
1396                 struct lmv_stripe_md    *old_lsm = lli->lli_lsm_md;
1397                 int                     idx;
1398
1399                 CERROR("%s: inode "DFID"(%p)'s lmv layout mismatch (%p)/(%p)"
1400                        "magic:0x%x/0x%x stripe count: %d/%d master_mdt: %d/%d"
1401                        "hash_type:0x%x/0x%x layout: 0x%x/0x%x pool:%s/%s\n",
1402                        ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1403                        inode, lsm, old_lsm,
1404                        lsm->lsm_md_magic, old_lsm->lsm_md_magic,
1405                        lsm->lsm_md_stripe_count,
1406                        old_lsm->lsm_md_stripe_count,
1407                        lsm->lsm_md_master_mdt_index,
1408                        old_lsm->lsm_md_master_mdt_index,
1409                        lsm->lsm_md_hash_type, old_lsm->lsm_md_hash_type,
1410                        lsm->lsm_md_layout_version,
1411                        old_lsm->lsm_md_layout_version,
1412                        lsm->lsm_md_pool_name,
1413                        old_lsm->lsm_md_pool_name);
1414
1415                 for (idx = 0; idx < old_lsm->lsm_md_stripe_count; idx++) {
1416                         CERROR("%s: sub FIDs in old lsm idx %d, old: "DFID"\n",
1417                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1418                                PFID(&old_lsm->lsm_md_oinfo[idx].lmo_fid));
1419                 }
1420
1421                 for (idx = 0; idx < lsm->lsm_md_stripe_count; idx++) {
1422                         CERROR("%s: sub FIDs in new lsm idx %d, new: "DFID"\n",
1423                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1424                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1425                 }
1426
1427                 RETURN(-EIO);
1428         }
1429
1430         RETURN(0);
1431 }
1432
1433 void ll_clear_inode(struct inode *inode)
1434 {
1435         struct ll_inode_info *lli = ll_i2info(inode);
1436         struct ll_sb_info *sbi = ll_i2sbi(inode);
1437         ENTRY;
1438
1439         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1440                PFID(ll_inode2fid(inode)), inode);
1441
1442         if (S_ISDIR(inode->i_mode)) {
1443                 /* these should have been cleared in ll_file_release */
1444                 LASSERT(lli->lli_opendir_key == NULL);
1445                 LASSERT(lli->lli_sai == NULL);
1446                 LASSERT(lli->lli_opendir_pid == 0);
1447         }
1448
1449         spin_lock(&lli->lli_lock);
1450         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1451         spin_unlock(&lli->lli_lock);
1452         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1453
1454         LASSERT(!lli->lli_open_fd_write_count);
1455         LASSERT(!lli->lli_open_fd_read_count);
1456         LASSERT(!lli->lli_open_fd_exec_count);
1457
1458         if (lli->lli_mds_write_och)
1459                 ll_md_real_close(inode, FMODE_WRITE);
1460         if (lli->lli_mds_exec_och)
1461                 ll_md_real_close(inode, FMODE_EXEC);
1462         if (lli->lli_mds_read_och)
1463                 ll_md_real_close(inode, FMODE_READ);
1464
1465         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1466                 OBD_FREE(lli->lli_symlink_name,
1467                          strlen(lli->lli_symlink_name) + 1);
1468                 lli->lli_symlink_name = NULL;
1469         }
1470
1471         ll_xattr_cache_destroy(inode);
1472
1473         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1474                 LASSERT(lli->lli_posix_acl == NULL);
1475                 if (lli->lli_remote_perms) {
1476                         free_rmtperm_hash(lli->lli_remote_perms);
1477                         lli->lli_remote_perms = NULL;
1478                 }
1479         }
1480 #ifdef CONFIG_FS_POSIX_ACL
1481         else if (lli->lli_posix_acl) {
1482                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1483                 LASSERT(lli->lli_remote_perms == NULL);
1484                 posix_acl_release(lli->lli_posix_acl);
1485                 lli->lli_posix_acl = NULL;
1486         }
1487 #endif
1488         lli->lli_inode_magic = LLI_INODE_DEAD;
1489
1490         ll_clear_inode_capas(inode);
1491         if (S_ISDIR(inode->i_mode))
1492                 ll_dir_clear_lsm_md(inode);
1493         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1494                 LASSERT(list_empty(&lli->lli_agl_list));
1495
1496         /*
1497          * XXX This has to be done before lsm is freed below, because
1498          * cl_object still uses inode lsm.
1499          */
1500         cl_inode_fini(inode);
1501         lli->lli_has_smd = false;
1502
1503         EXIT;
1504 }
1505
1506 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1507                          struct md_open_data **mod)
1508 {
1509         struct lustre_md md;
1510         struct inode *inode = dentry->d_inode;
1511         struct ll_sb_info *sbi = ll_i2sbi(inode);
1512         struct ptlrpc_request *request = NULL;
1513         int rc, ia_valid;
1514         ENTRY;
1515
1516         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1517                                      LUSTRE_OPC_ANY, NULL);
1518         if (IS_ERR(op_data))
1519                 RETURN(PTR_ERR(op_data));
1520
1521         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1522                         &request, mod);
1523         if (rc) {
1524                 ptlrpc_req_finished(request);
1525                 if (rc == -ENOENT) {
1526                         clear_nlink(inode);
1527                         /* Unlinked special device node? Or just a race?
1528                          * Pretend we done everything. */
1529                         if (!S_ISREG(inode->i_mode) &&
1530                             !S_ISDIR(inode->i_mode)) {
1531                                 ia_valid = op_data->op_attr.ia_valid;
1532                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1533                                 rc = simple_setattr(dentry, &op_data->op_attr);
1534                                 op_data->op_attr.ia_valid = ia_valid;
1535                         }
1536                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1537                         CERROR("md_setattr fails: rc = %d\n", rc);
1538                 }
1539                 RETURN(rc);
1540         }
1541
1542         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1543                               sbi->ll_md_exp, &md);
1544         if (rc) {
1545                 ptlrpc_req_finished(request);
1546                 RETURN(rc);
1547         }
1548
1549         ia_valid = op_data->op_attr.ia_valid;
1550         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1551          * cache is not cleared yet. */
1552         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1553         rc = simple_setattr(dentry, &op_data->op_attr);
1554         op_data->op_attr.ia_valid = ia_valid;
1555
1556         /* Extract epoch data if obtained. */
1557         op_data->op_handle = md.body->mbo_handle;
1558         op_data->op_ioepoch = md.body->mbo_ioepoch;
1559
1560         rc = ll_update_inode(inode, &md);
1561         ptlrpc_req_finished(request);
1562
1563         RETURN(rc);
1564 }
1565
1566 /* Close IO epoch and send Size-on-MDS attribute update. */
1567 static int ll_setattr_done_writing(struct inode *inode,
1568                                    struct md_op_data *op_data,
1569                                    struct md_open_data *mod)
1570 {
1571         struct ll_inode_info *lli = ll_i2info(inode);
1572         int rc = 0;
1573         ENTRY;
1574
1575         LASSERT(op_data != NULL);
1576         if (!S_ISREG(inode->i_mode))
1577                 RETURN(0);
1578
1579         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1580                op_data->op_ioepoch, PFID(&lli->lli_fid));
1581
1582         op_data->op_flags = MF_EPOCH_CLOSE;
1583         ll_done_writing_attr(inode, op_data);
1584         ll_pack_inode2opdata(inode, op_data, NULL);
1585
1586         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1587         if (rc == -EAGAIN) {
1588                 /* MDS has instructed us to obtain Size-on-MDS attribute
1589                  * from OSTs and send setattr to back to MDS. */
1590                 rc = ll_som_update(inode, op_data);
1591         } else if (rc) {
1592                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1593                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1594                        PFID(ll_inode2fid(inode)), rc);
1595         }
1596         RETURN(rc);
1597 }
1598
1599 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1600 {
1601         struct obd_capa *capa;
1602         int rc;
1603
1604         if (attr->ia_valid & ATTR_SIZE)
1605                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1606         else
1607                 capa = ll_mdscapa_get(inode);
1608
1609         rc = cl_setattr_ost(inode, attr, capa);
1610
1611         if (attr->ia_valid & ATTR_SIZE)
1612                 ll_truncate_free_capa(capa);
1613         else
1614                 capa_put(capa);
1615
1616         return rc;
1617 }
1618
1619 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1620  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1621  * keep these values until such a time that objects are allocated for it.
1622  * We do the MDS operations first, as it is checking permissions for us.
1623  * We don't to the MDS RPC if there is nothing that we want to store there,
1624  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1625  * going to do an RPC anyways.
1626  *
1627  * If we are doing a truncate, we will send the mtime and ctime updates
1628  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1629  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1630  * at the same time.
1631  *
1632  * In case of HSMimport, we only set attr on MDS.
1633  */
1634 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1635 {
1636         struct inode *inode = dentry->d_inode;
1637         struct ll_inode_info *lli = ll_i2info(inode);
1638         struct md_op_data *op_data = NULL;
1639         struct md_open_data *mod = NULL;
1640         bool file_is_released = false;
1641         int rc = 0, rc1 = 0;
1642         ENTRY;
1643
1644         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1645                "valid %x, hsm_import %d\n",
1646                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1647                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1648                hsm_import);
1649
1650         if (attr->ia_valid & ATTR_SIZE) {
1651                 /* Check new size against VFS/VM file size limit and rlimit */
1652                 rc = inode_newsize_ok(inode, attr->ia_size);
1653                 if (rc)
1654                         RETURN(rc);
1655
1656                 /* The maximum Lustre file size is variable, based on the
1657                  * OST maximum object size and number of stripes.  This
1658                  * needs another check in addition to the VFS check above. */
1659                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1660                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1661                                PFID(&lli->lli_fid), attr->ia_size,
1662                                ll_file_maxbytes(inode));
1663                         RETURN(-EFBIG);
1664                 }
1665
1666                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1667         }
1668
1669         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1670         if (attr->ia_valid & TIMES_SET_FLAGS) {
1671                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1672                     !cfs_capable(CFS_CAP_FOWNER))
1673                         RETURN(-EPERM);
1674         }
1675
1676         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1677         if (attr->ia_valid & ATTR_CTIME) {
1678                 attr->ia_ctime = CFS_CURRENT_TIME;
1679                 attr->ia_valid |= ATTR_CTIME_SET;
1680         }
1681         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1682             (attr->ia_valid & ATTR_ATIME)) {
1683                 attr->ia_atime = CFS_CURRENT_TIME;
1684                 attr->ia_valid |= ATTR_ATIME_SET;
1685         }
1686         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1687             (attr->ia_valid & ATTR_MTIME)) {
1688                 attr->ia_mtime = CFS_CURRENT_TIME;
1689                 attr->ia_valid |= ATTR_MTIME_SET;
1690         }
1691
1692         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1693                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1694                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1695                        cfs_time_current_sec());
1696
1697         /* We always do an MDS RPC, even if we're only changing the size;
1698          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1699
1700         OBD_ALLOC_PTR(op_data);
1701         if (op_data == NULL)
1702                 RETURN(-ENOMEM);
1703
1704         if (!S_ISDIR(inode->i_mode)) {
1705                 if (attr->ia_valid & ATTR_SIZE)
1706                         inode_dio_write_done(inode);
1707                 mutex_unlock(&inode->i_mutex);
1708         }
1709
1710         /* truncate on a released file must failed with -ENODATA,
1711          * so size must not be set on MDS for released file
1712          * but other attributes must be set
1713          */
1714         if (S_ISREG(inode->i_mode)) {
1715                 struct lov_stripe_md *lsm;
1716                 __u32 gen;
1717
1718                 ll_layout_refresh(inode, &gen);
1719                 lsm = ccc_inode_lsm_get(inode);
1720                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1721                         file_is_released = true;
1722                 ccc_inode_lsm_put(inode, lsm);
1723
1724                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1725                         if (file_is_released) {
1726                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1727                                 if (rc < 0)
1728                                         GOTO(out, rc);
1729
1730                                 file_is_released = false;
1731                                 ll_layout_refresh(inode, &gen);
1732                         }
1733
1734                         /* If we are changing file size, file content is
1735                          * modified, flag it. */
1736                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1737                         spin_lock(&lli->lli_lock);
1738                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1739                         spin_unlock(&lli->lli_lock);
1740                         op_data->op_bias |= MDS_DATA_MODIFIED;
1741                 }
1742         }
1743
1744         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1745
1746         /* Open epoch for truncate. */
1747         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1748             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1749                 op_data->op_flags = MF_EPOCH_OPEN;
1750
1751         rc = ll_md_setattr(dentry, op_data, &mod);
1752         if (rc)
1753                 GOTO(out, rc);
1754
1755         /* RPC to MDT is sent, cancel data modification flag */
1756         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1757                 spin_lock(&lli->lli_lock);
1758                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1759                 spin_unlock(&lli->lli_lock);
1760         }
1761
1762         ll_ioepoch_open(lli, op_data->op_ioepoch);
1763         if (!S_ISREG(inode->i_mode) || file_is_released)
1764                 GOTO(out, rc = 0);
1765
1766         if (attr->ia_valid & (ATTR_SIZE |
1767                               ATTR_ATIME | ATTR_ATIME_SET |
1768                               ATTR_MTIME | ATTR_MTIME_SET)) {
1769                 /* For truncate and utimes sending attributes to OSTs, setting
1770                  * mtime/atime to the past will be performed under PW [0:EOF]
1771                  * extent lock (new_size:EOF for truncate).  It may seem
1772                  * excessive to send mtime/atime updates to OSTs when not
1773                  * setting times to past, but it is necessary due to possible
1774                  * time de-synchronization between MDT inode and OST objects */
1775                 if (attr->ia_valid & ATTR_SIZE)
1776                         down_write(&lli->lli_trunc_sem);
1777                 rc = ll_setattr_ost(inode, attr);
1778                 if (attr->ia_valid & ATTR_SIZE)
1779                         up_write(&lli->lli_trunc_sem);
1780         }
1781         EXIT;
1782 out:
1783         if (op_data) {
1784                 if (op_data->op_ioepoch) {
1785                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1786                         if (!rc)
1787                                 rc = rc1;
1788                 }
1789                 ll_finish_md_op_data(op_data);
1790         }
1791         if (!S_ISDIR(inode->i_mode)) {
1792                 mutex_lock(&inode->i_mutex);
1793                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1794                         inode_dio_wait(inode);
1795         }
1796
1797         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1798                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1799
1800         return rc;
1801 }
1802
1803 int ll_setattr(struct dentry *de, struct iattr *attr)
1804 {
1805         int mode = de->d_inode->i_mode;
1806
1807         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1808                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1809                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1810
1811         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1812                                (ATTR_SIZE|ATTR_MODE)) &&
1813             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1814              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1815               !(attr->ia_mode & S_ISGID))))
1816                 attr->ia_valid |= ATTR_FORCE;
1817
1818         if ((attr->ia_valid & ATTR_MODE) &&
1819             (mode & S_ISUID) &&
1820             !(attr->ia_mode & S_ISUID) &&
1821             !(attr->ia_valid & ATTR_KILL_SUID))
1822                 attr->ia_valid |= ATTR_KILL_SUID;
1823
1824         if ((attr->ia_valid & ATTR_MODE) &&
1825             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1826             !(attr->ia_mode & S_ISGID) &&
1827             !(attr->ia_valid & ATTR_KILL_SGID))
1828                 attr->ia_valid |= ATTR_KILL_SGID;
1829
1830         return ll_setattr_raw(de, attr, false);
1831 }
1832
1833 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1834                        __u64 max_age, __u32 flags)
1835 {
1836         struct ll_sb_info *sbi = ll_s2sbi(sb);
1837         struct obd_statfs obd_osfs;
1838         int rc;
1839         ENTRY;
1840
1841         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1842         if (rc) {
1843                 CERROR("md_statfs fails: rc = %d\n", rc);
1844                 RETURN(rc);
1845         }
1846
1847         osfs->os_type = sb->s_magic;
1848
1849         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1850                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1851
1852         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1853                 flags |= OBD_STATFS_NODELAY;
1854
1855         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1856         if (rc) {
1857                 CERROR("obd_statfs fails: rc = %d\n", rc);
1858                 RETURN(rc);
1859         }
1860
1861         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1862                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1863                obd_osfs.os_files);
1864
1865         osfs->os_bsize = obd_osfs.os_bsize;
1866         osfs->os_blocks = obd_osfs.os_blocks;
1867         osfs->os_bfree = obd_osfs.os_bfree;
1868         osfs->os_bavail = obd_osfs.os_bavail;
1869
1870         /* If we don't have as many objects free on the OST as inodes
1871          * on the MDS, we reduce the total number of inodes to
1872          * compensate, so that the "inodes in use" number is correct.
1873          */
1874         if (obd_osfs.os_ffree < osfs->os_ffree) {
1875                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1876                         obd_osfs.os_ffree;
1877                 osfs->os_ffree = obd_osfs.os_ffree;
1878         }
1879
1880         RETURN(rc);
1881 }
1882 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1883 {
1884         struct super_block *sb = de->d_sb;
1885         struct obd_statfs osfs;
1886         __u64 fsid = huge_encode_dev(sb->s_dev);
1887         int rc;
1888
1889         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1890         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1891
1892         /* Some amount of caching on the client is allowed */
1893         rc = ll_statfs_internal(sb, &osfs,
1894                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1895                                 0);
1896         if (rc)
1897                 return rc;
1898
1899         statfs_unpack(sfs, &osfs);
1900
1901         /* We need to downshift for all 32-bit kernels, because we can't
1902          * tell if the kernel is being called via sys_statfs64() or not.
1903          * Stop before overflowing f_bsize - in which case it is better
1904          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1905         if (sizeof(long) < 8) {
1906                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1907                         sfs->f_bsize <<= 1;
1908
1909                         osfs.os_blocks >>= 1;
1910                         osfs.os_bfree >>= 1;
1911                         osfs.os_bavail >>= 1;
1912                 }
1913         }
1914
1915         sfs->f_blocks = osfs.os_blocks;
1916         sfs->f_bfree = osfs.os_bfree;
1917         sfs->f_bavail = osfs.os_bavail;
1918         sfs->f_fsid.val[0] = (__u32)fsid;
1919         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1920         return 0;
1921 }
1922
1923 void ll_inode_size_lock(struct inode *inode)
1924 {
1925         struct ll_inode_info *lli;
1926
1927         LASSERT(!S_ISDIR(inode->i_mode));
1928
1929         lli = ll_i2info(inode);
1930         mutex_lock(&lli->lli_size_mutex);
1931 }
1932
1933 void ll_inode_size_unlock(struct inode *inode)
1934 {
1935         struct ll_inode_info *lli;
1936
1937         lli = ll_i2info(inode);
1938         mutex_unlock(&lli->lli_size_mutex);
1939 }
1940
1941 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1942 {
1943         struct ll_inode_info *lli = ll_i2info(inode);
1944         struct mdt_body *body = md->body;
1945         struct lov_stripe_md *lsm = md->lsm;
1946         struct ll_sb_info *sbi = ll_i2sbi(inode);
1947
1948         LASSERT((lsm != NULL) == ((body->mbo_valid & OBD_MD_FLEASIZE) != 0));
1949         if (lsm != NULL) {
1950                 if (!lli->lli_has_smd &&
1951                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1952                         cl_file_inode_init(inode, md);
1953
1954                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1955                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1956                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1957         }
1958
1959         if (S_ISDIR(inode->i_mode)) {
1960                 int     rc;
1961
1962                 rc = ll_update_lsm_md(inode, md);
1963                 if (rc != 0)
1964                         return rc;
1965         }
1966
1967         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1968                 if (body->mbo_valid & OBD_MD_FLRMTPERM)
1969                         ll_update_remote_perm(inode, md->remote_perm);
1970         }
1971 #ifdef CONFIG_FS_POSIX_ACL
1972         else if (body->mbo_valid & OBD_MD_FLACL) {
1973                 spin_lock(&lli->lli_lock);
1974                 if (lli->lli_posix_acl)
1975                         posix_acl_release(lli->lli_posix_acl);
1976                 lli->lli_posix_acl = md->posix_acl;
1977                 spin_unlock(&lli->lli_lock);
1978         }
1979 #endif
1980         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1981                                         sbi->ll_flags & LL_SBI_32BIT_API);
1982         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1983
1984         if (body->mbo_valid & OBD_MD_FLATIME) {
1985                 if (body->mbo_atime > LTIME_S(inode->i_atime))
1986                         LTIME_S(inode->i_atime) = body->mbo_atime;
1987                 lli->lli_atime = body->mbo_atime;
1988         }
1989
1990         if (body->mbo_valid & OBD_MD_FLMTIME) {
1991                 if (body->mbo_mtime > LTIME_S(inode->i_mtime)) {
1992                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1993                                "to "LPU64"\n", inode->i_ino,
1994                                LTIME_S(inode->i_mtime), body->mbo_mtime);
1995                         LTIME_S(inode->i_mtime) = body->mbo_mtime;
1996                 }
1997                 lli->lli_mtime = body->mbo_mtime;
1998         }
1999
2000         if (body->mbo_valid & OBD_MD_FLCTIME) {
2001                 if (body->mbo_ctime > LTIME_S(inode->i_ctime))
2002                         LTIME_S(inode->i_ctime) = body->mbo_ctime;
2003                 lli->lli_ctime = body->mbo_ctime;
2004         }
2005
2006         if (body->mbo_valid & OBD_MD_FLMODE)
2007                 inode->i_mode = (inode->i_mode & S_IFMT) |
2008                                 (body->mbo_mode & ~S_IFMT);
2009
2010         if (body->mbo_valid & OBD_MD_FLTYPE)
2011                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2012                                 (body->mbo_mode & S_IFMT);
2013
2014         LASSERT(inode->i_mode != 0);
2015         if (S_ISREG(inode->i_mode))
2016                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2017                                        LL_MAX_BLKSIZE_BITS);
2018         else
2019                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2020
2021         if (body->mbo_valid & OBD_MD_FLUID)
2022                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2023         if (body->mbo_valid & OBD_MD_FLGID)
2024                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2025         if (body->mbo_valid & OBD_MD_FLFLAGS)
2026                 inode->i_flags = ll_ext_to_inode_flags(body->mbo_flags);
2027         if (body->mbo_valid & OBD_MD_FLNLINK)
2028                 set_nlink(inode, body->mbo_nlink);
2029         if (body->mbo_valid & OBD_MD_FLRDEV)
2030                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2031
2032         if (body->mbo_valid & OBD_MD_FLID) {
2033                 /* FID shouldn't be changed! */
2034                 if (fid_is_sane(&lli->lli_fid)) {
2035                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2036                                  "Trying to change FID "DFID
2037                                  " to the "DFID", inode "DFID"(%p)\n",
2038                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2039                                  PFID(ll_inode2fid(inode)), inode);
2040                 } else {
2041                         lli->lli_fid = body->mbo_fid1;
2042                 }
2043         }
2044
2045         LASSERT(fid_seq(&lli->lli_fid) != 0);
2046
2047         if (body->mbo_valid & OBD_MD_FLSIZE) {
2048                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2049                     S_ISREG(inode->i_mode)) {
2050                         struct lustre_handle lockh;
2051                         ldlm_mode_t mode;
2052
2053                         /* As it is possible a blocking ast has been processed
2054                          * by this time, we need to check there is an UPDATE
2055                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2056                          * it. */
2057                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2058                                                &lockh, LDLM_FL_CBPENDING,
2059                                                LCK_CR | LCK_CW |
2060                                                LCK_PR | LCK_PW);
2061                         if (mode) {
2062                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2063                                                       LLIF_EPOCH_PENDING |
2064                                                       LLIF_SOM_DIRTY)) {
2065                                         CERROR("%s: inode "DFID" flags %u still"
2066                                                " has size authority! do not "
2067                                                "trust the size from MDS\n",
2068                                                sbi->ll_md_exp->exp_obd->obd_name,
2069                                                PFID(ll_inode2fid(inode)),
2070                                                lli->lli_flags);
2071                                 } else {
2072                                         /* Use old size assignment to avoid
2073                                          * deadlock bz14138 & bz14326 */
2074                                         i_size_write(inode, body->mbo_size);
2075                                         spin_lock(&lli->lli_lock);
2076                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2077                                         spin_unlock(&lli->lli_lock);
2078                                 }
2079                                 ldlm_lock_decref(&lockh, mode);
2080                         }
2081                 } else {
2082                         /* Use old size assignment to avoid
2083                          * deadlock bz14138 & bz14326 */
2084                         i_size_write(inode, body->mbo_size);
2085
2086                         CDEBUG(D_VFSTRACE,
2087                                "inode="DFID", updating i_size %llu\n",
2088                                PFID(ll_inode2fid(inode)),
2089                                (unsigned long long)body->mbo_size);
2090                 }
2091
2092                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2093                         inode->i_blocks = body->mbo_blocks;
2094         }
2095
2096         if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
2097                 LASSERT(md->mds_capa);
2098                 ll_add_capa(inode, md->mds_capa);
2099         }
2100
2101         if (body->mbo_valid & OBD_MD_FLOSSCAPA) {
2102                 LASSERT(md->oss_capa);
2103                 ll_add_capa(inode, md->oss_capa);
2104         }
2105
2106         if (body->mbo_valid & OBD_MD_TSTATE) {
2107                 if (body->mbo_t_state & MS_RESTORE)
2108                         lli->lli_flags |= LLIF_FILE_RESTORING;
2109         }
2110
2111         return 0;
2112 }
2113
2114 int ll_read_inode2(struct inode *inode, void *opaque)
2115 {
2116         struct lustre_md *md = opaque;
2117         struct ll_inode_info *lli = ll_i2info(inode);
2118         int     rc;
2119         ENTRY;
2120
2121         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2122                PFID(&lli->lli_fid), inode);
2123
2124         LASSERT(!lli->lli_has_smd);
2125
2126         /* Core attributes from the MDS first.  This is a new inode, and
2127          * the VFS doesn't zero times in the core inode so we have to do
2128          * it ourselves.  They will be overwritten by either MDS or OST
2129          * attributes - we just need to make sure they aren't newer. */
2130         LTIME_S(inode->i_mtime) = 0;
2131         LTIME_S(inode->i_atime) = 0;
2132         LTIME_S(inode->i_ctime) = 0;
2133         inode->i_rdev = 0;
2134         rc = ll_update_inode(inode, md);
2135         if (rc != 0)
2136                 RETURN(rc);
2137
2138         /* OIDEBUG(inode); */
2139
2140         /* initializing backing dev info. */
2141         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2142
2143
2144         if (S_ISREG(inode->i_mode)) {
2145                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2146                 inode->i_op = &ll_file_inode_operations;
2147                 inode->i_fop = sbi->ll_fop;
2148                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2149                 EXIT;
2150         } else if (S_ISDIR(inode->i_mode)) {
2151                 inode->i_op = &ll_dir_inode_operations;
2152                 inode->i_fop = &ll_dir_operations;
2153                 EXIT;
2154         } else if (S_ISLNK(inode->i_mode)) {
2155                 inode->i_op = &ll_fast_symlink_inode_operations;
2156                 EXIT;
2157         } else {
2158                 inode->i_op = &ll_special_inode_operations;
2159
2160                 init_special_inode(inode, inode->i_mode,
2161                                    inode->i_rdev);
2162
2163                 EXIT;
2164         }
2165
2166         return 0;
2167 }
2168
2169 void ll_delete_inode(struct inode *inode)
2170 {
2171         struct ll_inode_info *lli = ll_i2info(inode);
2172         ENTRY;
2173
2174         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2175                 /* It is last chance to write out dirty pages,
2176                  * otherwise we may lose data while umount */
2177                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2178
2179         truncate_inode_pages_final(&inode->i_data);
2180
2181         LASSERTF(inode->i_data.nrpages == 0, "inode="DFID"(%p) nrpages=%lu, "
2182                  "see https://jira.hpdd.intel.com/browse/LU-118\n",
2183                  PFID(ll_inode2fid(inode)), inode, inode->i_data.nrpages);
2184
2185 #ifdef HAVE_SBOPS_EVICT_INODE
2186         ll_clear_inode(inode);
2187 #endif
2188         clear_inode(inode);
2189
2190         EXIT;
2191 }
2192
2193 int ll_iocontrol(struct inode *inode, struct file *file,
2194                  unsigned int cmd, unsigned long arg)
2195 {
2196         struct ll_sb_info *sbi = ll_i2sbi(inode);
2197         struct ptlrpc_request *req = NULL;
2198         int rc, flags = 0;
2199         ENTRY;
2200
2201         switch(cmd) {
2202         case FSFILT_IOC_GETFLAGS: {
2203                 struct mdt_body *body;
2204                 struct md_op_data *op_data;
2205
2206                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2207                                              0, 0, LUSTRE_OPC_ANY,
2208                                              NULL);
2209                 if (IS_ERR(op_data))
2210                         RETURN(PTR_ERR(op_data));
2211
2212                 op_data->op_valid = OBD_MD_FLFLAGS;
2213                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2214                 ll_finish_md_op_data(op_data);
2215                 if (rc) {
2216                         CERROR("%s: failure inode "DFID": rc = %d\n",
2217                                sbi->ll_md_exp->exp_obd->obd_name,
2218                                PFID(ll_inode2fid(inode)), rc);
2219                         RETURN(-abs(rc));
2220                 }
2221
2222                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2223
2224                 flags = body->mbo_flags;
2225
2226                 ptlrpc_req_finished(req);
2227
2228                 RETURN(put_user(flags, (int __user *)arg));
2229         }
2230         case FSFILT_IOC_SETFLAGS: {
2231                 struct lov_stripe_md *lsm;
2232                 struct obd_info oinfo = { { { 0 } } };
2233                 struct md_op_data *op_data;
2234
2235                 if (get_user(flags, (int __user *)arg))
2236                         RETURN(-EFAULT);
2237
2238                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2239                                              LUSTRE_OPC_ANY, NULL);
2240                 if (IS_ERR(op_data))
2241                         RETURN(PTR_ERR(op_data));
2242
2243                 op_data->op_attr_flags = flags;
2244                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2245                 rc = md_setattr(sbi->ll_md_exp, op_data,
2246                                 NULL, 0, NULL, 0, &req, NULL);
2247                 ll_finish_md_op_data(op_data);
2248                 ptlrpc_req_finished(req);
2249                 if (rc)
2250                         RETURN(rc);
2251
2252                 inode->i_flags = ll_ext_to_inode_flags(flags);
2253
2254                 lsm = ccc_inode_lsm_get(inode);
2255                 if (!lsm_has_objects(lsm)) {
2256                         ccc_inode_lsm_put(inode, lsm);
2257                         RETURN(0);
2258                 }
2259
2260                 OBDO_ALLOC(oinfo.oi_oa);
2261                 if (!oinfo.oi_oa) {
2262                         ccc_inode_lsm_put(inode, lsm);
2263                         RETURN(-ENOMEM);
2264                 }
2265                 oinfo.oi_md = lsm;
2266                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2267                 oinfo.oi_oa->o_flags = flags;
2268                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2269                                        OBD_MD_FLGROUP;
2270                 oinfo.oi_capa = ll_mdscapa_get(inode);
2271                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2272                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2273                 capa_put(oinfo.oi_capa);
2274                 OBDO_FREE(oinfo.oi_oa);
2275                 ccc_inode_lsm_put(inode, lsm);
2276
2277                 if (rc && rc != -EPERM && rc != -EACCES)
2278                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2279
2280                 RETURN(rc);
2281         }
2282         default:
2283                 RETURN(-ENOSYS);
2284         }
2285
2286         RETURN(0);
2287 }
2288
2289 int ll_flush_ctx(struct inode *inode)
2290 {
2291         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2292
2293         CDEBUG(D_SEC, "flush context for user %d\n",
2294                from_kuid(&init_user_ns, current_uid()));
2295
2296         obd_set_info_async(NULL, sbi->ll_md_exp,
2297                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2298                            0, NULL, NULL);
2299         obd_set_info_async(NULL, sbi->ll_dt_exp,
2300                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2301                            0, NULL, NULL);
2302         return 0;
2303 }
2304
2305 /* umount -f client means force down, don't save state */
2306 void ll_umount_begin(struct super_block *sb)
2307 {
2308         struct ll_sb_info *sbi = ll_s2sbi(sb);
2309         struct obd_device *obd;
2310         struct obd_ioctl_data *ioc_data;
2311         ENTRY;
2312
2313         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2314                sb->s_count, atomic_read(&sb->s_active));
2315
2316         obd = class_exp2obd(sbi->ll_md_exp);
2317         if (obd == NULL) {
2318                 CERROR("Invalid MDC connection handle "LPX64"\n",
2319                        sbi->ll_md_exp->exp_handle.h_cookie);
2320                 EXIT;
2321                 return;
2322         }
2323         obd->obd_force = 1;
2324
2325         obd = class_exp2obd(sbi->ll_dt_exp);
2326         if (obd == NULL) {
2327                 CERROR("Invalid LOV connection handle "LPX64"\n",
2328                        sbi->ll_dt_exp->exp_handle.h_cookie);
2329                 EXIT;
2330                 return;
2331         }
2332         obd->obd_force = 1;
2333
2334         OBD_ALLOC_PTR(ioc_data);
2335         if (ioc_data) {
2336                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2337                               sizeof *ioc_data, ioc_data, NULL);
2338
2339                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2340                               sizeof *ioc_data, ioc_data, NULL);
2341
2342                 OBD_FREE_PTR(ioc_data);
2343         }
2344
2345         /* Really, we'd like to wait until there are no requests outstanding,
2346          * and then continue.  For now, we just invalidate the requests,
2347          * schedule() and sleep one second if needed, and hope.
2348          */
2349         schedule();
2350         EXIT;
2351 }
2352
2353 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2354 {
2355         struct ll_sb_info *sbi = ll_s2sbi(sb);
2356         char *profilenm = get_profile_name(sb);
2357         int err;
2358         __u32 read_only;
2359
2360         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2361                 read_only = *flags & MS_RDONLY;
2362                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2363                                          sizeof(KEY_READ_ONLY),
2364                                          KEY_READ_ONLY, sizeof(read_only),
2365                                          &read_only, NULL);
2366                 if (err) {
2367                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2368                                       profilenm, read_only ?
2369                                       "read-only" : "read-write", err);
2370                         return err;
2371                 }
2372
2373                 if (read_only)
2374                         sb->s_flags |= MS_RDONLY;
2375                 else
2376                         sb->s_flags &= ~MS_RDONLY;
2377
2378                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2379                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2380                                       read_only ?  "read-only" : "read-write");
2381         }
2382         return 0;
2383 }
2384
2385 /**
2386  * Cleanup the open handle that is cached on MDT-side.
2387  *
2388  * For open case, the client side open handling thread may hit error
2389  * after the MDT grant the open. Under such case, the client should
2390  * send close RPC to the MDT as cleanup; otherwise, the open handle
2391  * on the MDT will be leaked there until the client umount or evicted.
2392  *
2393  * In further, if someone unlinked the file, because the open handle
2394  * holds the reference on such file/object, then it will block the
2395  * subsequent threads that want to locate such object via FID.
2396  *
2397  * \param[in] sb        super block for this file-system
2398  * \param[in] open_req  pointer to the original open request
2399  */
2400 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2401 {
2402         struct mdt_body                 *body;
2403         struct md_op_data               *op_data;
2404         struct ptlrpc_request           *close_req = NULL;
2405         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2406         ENTRY;
2407
2408         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2409         OBD_ALLOC_PTR(op_data);
2410         if (op_data == NULL) {
2411                 CWARN("%s: cannot allocate op_data to release open handle for "
2412                       DFID"\n",
2413                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2414
2415                 RETURN_EXIT;
2416         }
2417
2418         op_data->op_fid1 = body->mbo_fid1;
2419         op_data->op_ioepoch = body->mbo_ioepoch;
2420         op_data->op_handle = body->mbo_handle;
2421         op_data->op_mod_time = cfs_time_current_sec();
2422         md_close(exp, op_data, NULL, &close_req);
2423         ptlrpc_req_finished(close_req);
2424         ll_finish_md_op_data(op_data);
2425
2426         EXIT;
2427 }
2428
2429 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2430                   struct super_block *sb, struct lookup_intent *it)
2431 {
2432         struct ll_sb_info *sbi = NULL;
2433         struct lustre_md md = { NULL };
2434         int rc;
2435         ENTRY;
2436
2437         LASSERT(*inode || sb);
2438         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2439         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2440                               sbi->ll_md_exp, &md);
2441         if (rc != 0)
2442                 GOTO(cleanup, rc);
2443
2444         if (*inode) {
2445                 rc = ll_update_inode(*inode, &md);
2446                 if (rc != 0)
2447                         GOTO(out, rc);
2448         } else {
2449                 LASSERT(sb != NULL);
2450
2451                 /*
2452                  * At this point server returns to client's same fid as client
2453                  * generated for creating. So using ->fid1 is okay here.
2454                  */
2455                 LASSERT(fid_is_sane(&md.body->mbo_fid1));
2456
2457                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2458                                              sbi->ll_flags & LL_SBI_32BIT_API),
2459                                  &md);
2460                 if (IS_ERR(*inode)) {
2461 #ifdef CONFIG_FS_POSIX_ACL
2462                         if (md.posix_acl) {
2463                                 posix_acl_release(md.posix_acl);
2464                                 md.posix_acl = NULL;
2465                         }
2466 #endif
2467                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2468                         *inode = NULL;
2469                         CERROR("new_inode -fatal: rc %d\n", rc);
2470                         GOTO(out, rc);
2471                 }
2472         }
2473
2474         /* Handling piggyback layout lock.
2475          * Layout lock can be piggybacked by getattr and open request.
2476          * The lsm can be applied to inode only if it comes with a layout lock
2477          * otherwise correct layout may be overwritten, for example:
2478          * 1. proc1: mdt returns a lsm but not granting layout
2479          * 2. layout was changed by another client
2480          * 3. proc2: refresh layout and layout lock granted
2481          * 4. proc1: to apply a stale layout */
2482         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2483                 struct lustre_handle lockh;
2484                 struct ldlm_lock *lock;
2485
2486                 lockh.cookie = it->d.lustre.it_lock_handle;
2487                 lock = ldlm_handle2lock(&lockh);
2488                 LASSERT(lock != NULL);
2489                 if (ldlm_has_layout(lock)) {
2490                         struct cl_object_conf conf;
2491
2492                         memset(&conf, 0, sizeof(conf));
2493                         conf.coc_opc = OBJECT_CONF_SET;
2494                         conf.coc_inode = *inode;
2495                         conf.coc_lock = lock;
2496                         conf.u.coc_md = &md;
2497                         (void)ll_layout_conf(*inode, &conf);
2498                 }
2499                 LDLM_LOCK_PUT(lock);
2500         }
2501
2502         GOTO(out, rc = 0);
2503
2504 out:
2505         if (md.lsm != NULL)
2506                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2507         md_free_lustre_md(sbi->ll_md_exp, &md);
2508
2509 cleanup:
2510         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2511                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2512
2513         return rc;
2514 }
2515
2516 int ll_obd_statfs(struct inode *inode, void __user *arg)
2517 {
2518         struct ll_sb_info *sbi = NULL;
2519         struct obd_export *exp;
2520         char *buf = NULL;
2521         struct obd_ioctl_data *data = NULL;
2522         __u32 type;
2523         __u32 __user flags;     /* not user, but obd_iocontrol is abused */
2524         int len = 0, rc;
2525
2526         if (!inode || !(sbi = ll_i2sbi(inode)))
2527                 GOTO(out_statfs, rc = -EINVAL);
2528
2529         rc = obd_ioctl_getdata(&buf, &len, arg);
2530         if (rc)
2531                 GOTO(out_statfs, rc);
2532
2533         data = (void*)buf;
2534         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2535             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2536                 GOTO(out_statfs, rc = -EINVAL);
2537
2538         if (data->ioc_inllen1 != sizeof(__u32) ||
2539             data->ioc_inllen2 != sizeof(__u32) ||
2540             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2541             data->ioc_plen2 != sizeof(struct obd_uuid))
2542                 GOTO(out_statfs, rc = -EINVAL);
2543
2544         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2545         if (type & LL_STATFS_LMV)
2546                 exp = sbi->ll_md_exp;
2547         else if (type & LL_STATFS_LOV)
2548                 exp = sbi->ll_dt_exp;
2549         else
2550                 GOTO(out_statfs, rc = -ENODEV);
2551
2552         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2553         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2554         if (rc)
2555                 GOTO(out_statfs, rc);
2556 out_statfs:
2557         if (buf)
2558                 obd_ioctl_freedata(buf, len);
2559         return rc;
2560 }
2561
2562 int ll_process_config(struct lustre_cfg *lcfg)
2563 {
2564         struct super_block *sb;
2565         unsigned long x;
2566         int rc = 0;
2567         char *ptr;
2568
2569         /* The instance name contains the sb: lustre-client-aacfe000 */
2570         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2571         if (!ptr || !*(++ptr))
2572                 return -EINVAL;
2573         if (sscanf(ptr, "%lx", &x) != 1)
2574                 return -EINVAL;
2575         sb = (struct super_block *)x;
2576         /* This better be a real Lustre superblock! */
2577         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2578
2579         /* Note we have not called client_common_fill_super yet, so
2580            proc fns must be able to handle that! */
2581         rc = class_process_proc_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2582                                       lcfg, sb);
2583         if (rc > 0)
2584                 rc = 0;
2585         return rc;
2586 }
2587
2588 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2589 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2590                                        struct inode *i1, struct inode *i2,
2591                                        const char *name, size_t namelen,
2592                                        __u32 mode, __u32 opc, void *data)
2593 {
2594         LASSERT(i1 != NULL);
2595
2596         if (name == NULL) {
2597                 /* Do not reuse namelen for something else. */
2598                 if (namelen != 0)
2599                         return ERR_PTR(-EINVAL);
2600         } else {
2601                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2602                         return ERR_PTR(-ENAMETOOLONG);
2603
2604                 if (!lu_name_is_valid_2(name, namelen))
2605                         return ERR_PTR(-EINVAL);
2606         }
2607
2608         if (op_data == NULL)
2609                 OBD_ALLOC_PTR(op_data);
2610
2611         if (op_data == NULL)
2612                 return ERR_PTR(-ENOMEM);
2613
2614         ll_i2gids(op_data->op_suppgids, i1, i2);
2615         op_data->op_fid1 = *ll_inode2fid(i1);
2616         op_data->op_capa1 = ll_mdscapa_get(i1);
2617         op_data->op_default_stripe_offset = -1;
2618         if (S_ISDIR(i1->i_mode)) {
2619                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2620                 op_data->op_default_stripe_offset =
2621                            ll_i2info(i1)->lli_def_stripe_offset;
2622         }
2623
2624         if (i2) {
2625                 op_data->op_fid2 = *ll_inode2fid(i2);
2626                 op_data->op_capa2 = ll_mdscapa_get(i2);
2627                 if (S_ISDIR(i2->i_mode))
2628                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2629         } else {
2630                 fid_zero(&op_data->op_fid2);
2631                 op_data->op_capa2 = NULL;
2632         }
2633
2634         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2635                 op_data->op_cli_flags |= CLI_HASH64;
2636
2637         if (ll_need_32bit_api(ll_i2sbi(i1)))
2638                 op_data->op_cli_flags |= CLI_API32;
2639
2640         op_data->op_name = name;
2641         op_data->op_namelen = namelen;
2642         op_data->op_mode = mode;
2643         op_data->op_mod_time = cfs_time_current_sec();
2644         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2645         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2646         op_data->op_cap = cfs_curproc_cap_pack();
2647         op_data->op_bias = 0;
2648         op_data->op_cli_flags = 0;
2649         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2650              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2651                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2652         } else {
2653                 op_data->op_mds = 0;
2654         }
2655         op_data->op_data = data;
2656
2657         /* When called by ll_setattr_raw, file is i1. */
2658         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2659                 op_data->op_bias |= MDS_DATA_MODIFIED;
2660
2661         return op_data;
2662 }
2663
2664 void ll_finish_md_op_data(struct md_op_data *op_data)
2665 {
2666         capa_put(op_data->op_capa1);
2667         capa_put(op_data->op_capa2);
2668         OBD_FREE_PTR(op_data);
2669 }
2670
2671 #ifdef HAVE_SUPEROPS_USE_DENTRY
2672 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2673 #else
2674 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2675 #endif
2676 {
2677         struct ll_sb_info *sbi;
2678
2679 #ifdef HAVE_SUPEROPS_USE_DENTRY
2680         LASSERT((seq != NULL) && (dentry != NULL));
2681         sbi = ll_s2sbi(dentry->d_sb);
2682 #else
2683         LASSERT((seq != NULL) && (vfs != NULL));
2684         sbi = ll_s2sbi(vfs->mnt_sb);
2685 #endif
2686
2687         if (sbi->ll_flags & LL_SBI_NOLCK)
2688                 seq_puts(seq, ",nolock");
2689
2690         if (sbi->ll_flags & LL_SBI_FLOCK)
2691                 seq_puts(seq, ",flock");
2692
2693         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2694                 seq_puts(seq, ",localflock");
2695
2696         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2697                 seq_puts(seq, ",user_xattr");
2698
2699         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2700                 seq_puts(seq, ",lazystatfs");
2701
2702         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2703                 seq_puts(seq, ",user_fid2path");
2704
2705         RETURN(0);
2706 }
2707
2708 /**
2709  * Get obd name by cmd, and copy out to user space
2710  */
2711 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2712 {
2713         struct ll_sb_info *sbi = ll_i2sbi(inode);
2714         struct obd_device *obd;
2715         ENTRY;
2716
2717         if (cmd == OBD_IOC_GETDTNAME)
2718                 obd = class_exp2obd(sbi->ll_dt_exp);
2719         else if (cmd == OBD_IOC_GETMDNAME)
2720                 obd = class_exp2obd(sbi->ll_md_exp);
2721         else
2722                 RETURN(-EINVAL);
2723
2724         if (!obd)
2725                 RETURN(-ENOENT);
2726
2727         if (copy_to_user((void __user *)arg, obd->obd_name,
2728                          strlen(obd->obd_name) + 1))
2729                 RETURN(-EFAULT);
2730
2731         RETURN(0);
2732 }
2733
2734 /**
2735  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2736  * fsname will be returned in this buffer; otherwise, a static buffer will be
2737  * used to store the fsname and returned to caller.
2738  */
2739 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2740 {
2741         static char fsname_static[MTI_NAME_MAXLEN];
2742         struct lustre_sb_info *lsi = s2lsi(sb);
2743         char *ptr;
2744         int len;
2745
2746         if (buf == NULL) {
2747                 /* this means the caller wants to use static buffer
2748                  * and it doesn't care about race. Usually this is
2749                  * in error reporting path */
2750                 buf = fsname_static;
2751                 buflen = sizeof(fsname_static);
2752         }
2753
2754         len = strlen(lsi->lsi_lmd->lmd_profile);
2755         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2756         if (ptr && (strcmp(ptr, "-client") == 0))
2757                 len -= 7;
2758
2759         if (unlikely(len >= buflen))
2760                 len = buflen - 1;
2761         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2762         buf[len] = '\0';
2763
2764         return buf;
2765 }
2766
2767 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2768 {
2769         char *path = NULL;
2770
2771         struct path p;
2772
2773         p.dentry = dentry;
2774         p.mnt = current->fs->root.mnt;
2775         path_get(&p);
2776         path = d_path(&p, buf, bufsize);
2777         path_put(&p);
2778         return path;
2779 }
2780
2781 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2782 {
2783         char *buf, *path = NULL;
2784         struct dentry *dentry = NULL;
2785         struct vvp_object *obj = cl_inode2vvp(page->mapping->host);
2786
2787         /* this can be called inside spin lock so use GFP_ATOMIC. */
2788         buf = (char *)__get_free_page(GFP_ATOMIC);
2789         if (buf != NULL) {
2790                 dentry = d_find_alias(page->mapping->host);
2791                 if (dentry != NULL)
2792                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2793         }
2794
2795         CDEBUG(D_WARNING,
2796                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2797                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2798                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2799                PFID(&obj->vob_header.coh_lu.loh_fid),
2800                (path && !IS_ERR(path)) ? path : "", ioret);
2801
2802         if (dentry != NULL)
2803                 dput(dentry);
2804
2805         if (buf != NULL)
2806                 free_page((unsigned long)buf);
2807 }
2808
2809 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2810                         struct lov_user_md **kbuf)
2811 {
2812         struct lov_user_md      lum;
2813         ssize_t                 lum_size;
2814         ENTRY;
2815
2816         if (copy_from_user(&lum, md, sizeof(lum)))
2817                 RETURN(-EFAULT);
2818
2819         lum_size = ll_lov_user_md_size(&lum);
2820         if (lum_size < 0)
2821                 RETURN(lum_size);
2822
2823         OBD_ALLOC(*kbuf, lum_size);
2824         if (*kbuf == NULL)
2825                 RETURN(-ENOMEM);
2826
2827         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2828                 OBD_FREE(*kbuf, lum_size);
2829                 RETURN(-EFAULT);
2830         }
2831
2832         RETURN(lum_size);
2833 }
2834
2835 /*
2836  * Compute llite root squash state after a change of root squash
2837  * configuration setting or add/remove of a lnet nid
2838  */
2839 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2840 {
2841         struct root_squash_info *squash = &sbi->ll_squash;
2842         int i;
2843         bool matched;
2844         lnet_process_id_t id;
2845
2846         /* Update norootsquash flag */
2847         down_write(&squash->rsi_sem);
2848         if (list_empty(&squash->rsi_nosquash_nids))
2849                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2850         else {
2851                 /* Do not apply root squash as soon as one of our NIDs is
2852                  * in the nosquash_nids list */
2853                 matched = false;
2854                 i = 0;
2855                 while (LNetGetId(i++, &id) != -ENOENT) {
2856                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2857                                 continue;
2858                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2859                                 matched = true;
2860                                 break;
2861                         }
2862                 }
2863                 if (matched)
2864                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2865                 else
2866                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2867         }
2868         up_write(&squash->rsi_sem);
2869 }
2870
2871 /**
2872  * Parse linkea content to extract information about a given hardlink
2873  *
2874  * \param[in]   ldata      - Initialized linkea data
2875  * \param[in]   linkno     - Link identifier
2876  * \param[out]  parent_fid - The entry's parent FID
2877  * \param[out]  ln         - Entry name destination buffer
2878  *
2879  * \retval 0 on success
2880  * \retval Appropriate negative error code on failure
2881  */
2882 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2883                             struct lu_fid *parent_fid, struct lu_name *ln)
2884 {
2885         unsigned int    idx;
2886         int             rc;
2887         ENTRY;
2888
2889         rc = linkea_init(ldata);
2890         if (rc < 0)
2891                 RETURN(rc);
2892
2893         if (linkno >= ldata->ld_leh->leh_reccount)
2894                 /* beyond last link */
2895                 RETURN(-ENODATA);
2896
2897         linkea_first_entry(ldata);
2898         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2899                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2900                                     parent_fid);
2901                 if (idx == linkno)
2902                         break;
2903
2904                 linkea_next_entry(ldata);
2905         }
2906
2907         if (idx < linkno)
2908                 RETURN(-ENODATA);
2909
2910         RETURN(0);
2911 }
2912
2913 /**
2914  * Get parent FID and name of an identified link. Operation is performed for
2915  * a given link number, letting the caller iterate over linkno to list one or
2916  * all links of an entry.
2917  *
2918  * \param[in]     file - File descriptor against which to perform the operation
2919  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2920  *                       on and the available size. It is eventually filled with
2921  *                       the requested information or left untouched on error
2922  *
2923  * \retval - 0 on success
2924  * \retval - Appropriate negative error code on failure
2925  */
2926 int ll_getparent(struct file *file, struct getparent __user *arg)
2927 {
2928         struct dentry           *dentry = file->f_dentry;
2929         struct inode            *inode = file->f_dentry->d_inode;
2930         struct linkea_data      *ldata;
2931         struct lu_buf            buf = LU_BUF_NULL;
2932         struct lu_name           ln;
2933         struct lu_fid            parent_fid;
2934         __u32                    linkno;
2935         __u32                    name_size;
2936         int                      rc;
2937
2938         ENTRY;
2939
2940         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2941             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2942                 RETURN(-EPERM);
2943
2944         if (get_user(name_size, &arg->gp_name_size))
2945                 RETURN(-EFAULT);
2946
2947         if (get_user(linkno, &arg->gp_linkno))
2948                 RETURN(-EFAULT);
2949
2950         if (name_size > PATH_MAX)
2951                 RETURN(-EINVAL);
2952
2953         OBD_ALLOC(ldata, sizeof(*ldata));
2954         if (ldata == NULL)
2955                 RETURN(-ENOMEM);
2956
2957         rc = linkea_data_new(ldata, &buf);
2958         if (rc < 0)
2959                 GOTO(ldata_free, rc);
2960
2961         rc = ll_getxattr(dentry, XATTR_NAME_LINK, buf.lb_buf, buf.lb_len);
2962         if (rc < 0)
2963                 GOTO(lb_free, rc);
2964
2965         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2966         if (rc < 0)
2967                 GOTO(lb_free, rc);
2968
2969         if (ln.ln_namelen >= name_size)
2970                 GOTO(lb_free, rc = -EOVERFLOW);
2971
2972         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2973                 GOTO(lb_free, rc = -EFAULT);
2974
2975         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2976                 GOTO(lb_free, rc = -EFAULT);
2977
2978         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2979                 GOTO(lb_free, rc = -EFAULT);
2980
2981 lb_free:
2982         lu_buf_free(&buf);
2983 ldata_free:
2984         OBD_FREE(ldata, sizeof(*ldata));
2985
2986         RETURN(rc);
2987 }