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