drivers/staging/lustre/lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.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, 2012, 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/mm.h>
  47
  48 #include "../include/lustre_lite.h"
  49 #include "../include/lustre_ha.h"
  50 #include "../include/lustre_dlm.h"
  51 #include "../include/lprocfs_status.h"
  52 #include "../include/lustre_disk.h"
  53 #include "../include/lustre_param.h"
  54 #include "../include/lustre_log.h"
  55 #include "../include/cl_object.h"
  56 #include "../include/obd_cksum.h"
  57 #include "llite_internal.h"
  58
  59 struct kmem_cache *ll_file_data_slab;
  60 struct proc_dir_entry *proc_lustre_fs_root;
  61
  62 static LIST_HEAD(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
  78         sbi = kzalloc(sizeof(*sbi), GFP_NOFS);
  79         if (!sbi)
  80                 return NULL;
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         if (pages >> (20 - PAGE_CACHE_SHIFT) < 512) {
  91                 lru_page_max = pages / 2;
  92         } else {
  93                 lru_page_max = (pages / 4) * 3;
  94         }
  95
  96         /* initialize lru data */
  97         atomic_set(&sbi->ll_cache.ccc_users, 0);
  98         sbi->ll_cache.ccc_lru_max = lru_page_max;
  99         atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 100         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 101         INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 102
 103         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 104                                            SBI_DEFAULT_READAHEAD_MAX);
 105         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 106         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 107                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 108         INIT_LIST_HEAD(&sbi->ll_conn_chain);
 109         INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 110
 111         ll_generate_random_uuid(uuid);
 112         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 113         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 114
 115         spin_lock(&ll_sb_lock);
 116         list_add_tail(&sbi->ll_list, &ll_super_blocks);
 117         spin_unlock(&ll_sb_lock);
 118
 119         sbi->ll_flags |= LL_SBI_VERBOSE;
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121
 122         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 123
 124         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 125                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 126                                pp_r_hist.oh_lock);
 127                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 128                                pp_w_hist.oh_lock);
 129         }
 130
 131         /* metadata statahead is enabled by default */
 132         sbi->ll_sa_max = LL_SA_RPC_DEF;
 133         atomic_set(&sbi->ll_sa_total, 0);
 134         atomic_set(&sbi->ll_sa_wrong, 0);
 135         atomic_set(&sbi->ll_agl_total, 0);
 136         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 137
 138         return sbi;
 139 }
 140
 141 static void ll_free_sbi(struct super_block *sb)
 142 {
 143         struct ll_sb_info *sbi = ll_s2sbi(sb);
 144
 145         if (sbi != NULL) {
 146                 spin_lock(&ll_sb_lock);
 147                 list_del(&sbi->ll_list);
 148                 spin_unlock(&ll_sb_lock);
 149                 OBD_FREE(sbi, sizeof(*sbi));
 150         }
 151 }
 152
 153 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 154                                     struct vfsmount *mnt)
 155 {
 156         struct inode *root = NULL;
 157         struct ll_sb_info *sbi = ll_s2sbi(sb);
 158         struct obd_device *obd;
 159         struct obd_capa *oc = NULL;
 160         struct obd_statfs *osfs = NULL;
 161         struct ptlrpc_request *request = NULL;
 162         struct obd_connect_data *data = NULL;
 163         struct obd_uuid *uuid;
 164         struct md_op_data *op_data;
 165         struct lustre_md lmd;
 166         u64 valid;
 167         int size, err, checksum;
 168
 169         obd = class_name2obd(md);
 170         if (!obd) {
 171                 CERROR("MD %s: not setup or attached\n", md);
 172                 return -EINVAL;
 173         }
 174
 175         data = kzalloc(sizeof(*data), GFP_NOFS);
 176         if (!data)
 177                 return -ENOMEM;
 178
 179         osfs = kzalloc(sizeof(*osfs), GFP_NOFS);
 180         if (!osfs) {
 181                 OBD_FREE_PTR(data);
 182                 return -ENOMEM;
 183         }
 184
 185         if (proc_lustre_fs_root) {
 186                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 187                                                   dt, md);
 188                 if (err < 0)
 189                         CERROR("could not register mount in /proc/fs/lustre\n");
 190         }
 191
 192         /* indicate the features supported by this client */
 193         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 194                                   OBD_CONNECT_ATTRFID  |
 195                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 196                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 197                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 198                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 199                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 200                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 201                                   OBD_CONNECT_EINPROGRESS |
 202                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 203                                   OBD_CONNECT_LAYOUTLOCK |
 204                                   OBD_CONNECT_PINGLESS |
 205                                   OBD_CONNECT_MAX_EASIZE |
 206                                   OBD_CONNECT_FLOCK_DEAD |
 207                                   OBD_CONNECT_DISP_STRIPE;
 208
 209         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 210                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 211
 212         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 213                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 214 #ifdef CONFIG_FS_POSIX_ACL
 215         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 216 #endif
 217
 218         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 219                 /* flag mdc connection as lightweight, only used for test
 220                  * purpose, use with care */
 221                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 222
 223         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 224         data->ocd_version = LUSTRE_VERSION_CODE;
 225
 226         if (sb->s_flags & MS_RDONLY)
 227                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 228         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 229                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 230
 231 #ifdef HAVE_MS_FLOCK_LOCK
 232         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 233         sb->s_flags |= MS_FLOCK_LOCK;
 234 #endif
 235 #ifdef MS_HAS_NEW_AOPS
 236         sb->s_flags |= MS_HAS_NEW_AOPS;
 237 #endif
 238
 239         if (sbi->ll_flags & LL_SBI_FLOCK)
 240                 sbi->ll_fop = &ll_file_operations_flock;
 241         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 242                 sbi->ll_fop = &ll_file_operations;
 243         else
 244                 sbi->ll_fop = &ll_file_operations_noflock;
 245
 246         /* real client */
 247         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 248         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 249                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 250
 251         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 252
 253         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid,
 254                           data, NULL);
 255         if (err == -EBUSY) {
 256                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing recovery, of which this client is not a part. Please wait for recovery to complete, abort, or time out.\n",
 257                                    md);
 258                 goto out;
 259         } else if (err) {
 260                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 261                 goto out;
 262         }
 263
 264         sbi->ll_md_exp->exp_connect_data = *data;
 265
 266         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 267                            LUSTRE_SEQ_METADATA);
 268         if (err) {
 269                 CERROR("%s: Can't init metadata layer FID infrastructure, rc = %d\n",
 270                        sbi->ll_md_exp->exp_obd->obd_name, err);
 271                 goto out_md;
 272         }
 273
 274         /* For mount, we only need fs info from MDT0, and also in DNE, it
 275          * can make sure the client can be mounted as long as MDT0 is
 276          * available */
 277         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 278                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 279                         OBD_STATFS_FOR_MDT0);
 280         if (err)
 281                 goto out_md_fid;
 282
 283         /* This needs to be after statfs to ensure connect has finished.
 284          * Note that "data" does NOT contain the valid connect reply.
 285          * If connecting to a 1.8 server there will be no LMV device, so
 286          * we can access the MDC export directly and exp_connect_flags will
 287          * be non-zero, but if accessing an upgraded 2.1 server it will
 288          * have the correct flags filled in.
 289          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 290         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 291         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 292             valid != CLIENT_CONNECT_MDT_REQD) {
 293                 char *buf;
 294
 295                 buf = kzalloc(PAGE_CACHE_SIZE, GFP_KERNEL);
 296                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 297                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 298                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support feature(s) needed for correct operation of this client (%s). Please upgrade server or downgrade client.\n",
 299                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 300                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 301                 err = -EPROTO;
 302                 goto out_md_fid;
 303         }
 304
 305         size = sizeof(*data);
 306         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 307                            KEY_CONN_DATA,  &size, data, NULL);
 308         if (err) {
 309                 CERROR("%s: Get connect data failed: rc = %d\n",
 310                        sbi->ll_md_exp->exp_obd->obd_name, err);
 311                 goto out_md_fid;
 312         }
 313
 314         LASSERT(osfs->os_bsize);
 315         sb->s_blocksize = osfs->os_bsize;
 316         sb->s_blocksize_bits = log2(osfs->os_bsize);
 317         sb->s_magic = LL_SUPER_MAGIC;
 318         sb->s_maxbytes = MAX_LFS_FILESIZE;
 319         sbi->ll_namelen = osfs->os_namelen;
 320         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 321
 322         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 323             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 324                 LCONSOLE_INFO("Disabling user_xattr feature because it is not supported on the server\n");
 325                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 326         }
 327
 328         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 329 #ifdef MS_POSIXACL
 330                 sb->s_flags |= MS_POSIXACL;
 331 #endif
 332                 sbi->ll_flags |= LL_SBI_ACL;
 333         } else {
 334                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 335 #ifdef MS_POSIXACL
 336                 sb->s_flags &= ~MS_POSIXACL;
 337 #endif
 338                 sbi->ll_flags &= ~LL_SBI_ACL;
 339         }
 340
 341         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 342                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 343                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 344                         LCONSOLE_INFO("client is set as remote by default.\n");
 345                 }
 346         } else {
 347                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 348                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 349                         LCONSOLE_INFO("client claims to be remote, but server rejected, forced to be local.\n");
 350                 }
 351         }
 352
 353         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 354                 LCONSOLE_INFO("client enabled MDS capability!\n");
 355                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 359                 LCONSOLE_INFO("client enabled OSS capability!\n");
 360                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 361         }
 362
 363         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 364                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 365
 366         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 367                 sbi->ll_md_brw_size = data->ocd_brw_size;
 368         else
 369                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 372                 LCONSOLE_INFO("Layout lock feature supported.\n");
 373                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 374         }
 375
 376         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 377                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 378                         LCONSOLE_INFO(
 379                                 "%s: disabling xattr cache due to unknown maximum xattr size.\n",
 380                                 dt);
 381                 } else {
 382                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 383                         sbi->ll_xattr_cache_enabled = 1;
 384                 }
 385         }
 386
 387         obd = class_name2obd(dt);
 388         if (!obd) {
 389                 CERROR("DT %s: not setup or attached\n", dt);
 390                 err = -ENODEV;
 391                 goto out_md_fid;
 392         }
 393
 394         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 395                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 396                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 397                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 398                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 399                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 400                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 401                                   OBD_CONNECT_MAXBYTES |
 402                                   OBD_CONNECT_EINPROGRESS |
 403                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 404                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 405
 406         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 407                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 408
 409         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 410                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 411                  * disabled by default, because it can still be enabled on the
 412                  * fly via /proc. As a consequence, we still need to come to an
 413                  * agreement on the supported algorithms at connect time */
 414                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 415
 416                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 417                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 418                 else
 419                         data->ocd_cksum_types = cksum_types_supported_client();
 420         }
 421
 422         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 423         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 424                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 425
 426         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d ocd_grant: %d\n",
 427                data->ocd_connect_flags,
 428                data->ocd_version, data->ocd_grant);
 429
 430         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 431         obd->obd_upcall.onu_upcall = cl_ocd_update;
 432
 433         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 434
 435         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 436                           NULL);
 437         if (err == -EBUSY) {
 438                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing recovery, of which this client is not a part.  Please wait for recovery to complete, abort, or time out.\n",
 439                                    dt);
 440                 goto out_md;
 441         } else if (err) {
 442                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 443                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 444                 goto out_md;
 445         }
 446
 447         sbi->ll_dt_exp->exp_connect_data = *data;
 448
 449         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 450                            LUSTRE_SEQ_METADATA);
 451         if (err) {
 452                 CERROR("%s: Can't init data layer FID infrastructure, rc = %d\n",
 453                        sbi->ll_dt_exp->exp_obd->obd_name, err);
 454                 goto out_dt;
 455         }
 456
 457         mutex_lock(&sbi->ll_lco.lco_lock);
 458         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 459         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 460         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 461         mutex_unlock(&sbi->ll_lco.lco_lock);
 462
 463         fid_zero(&sbi->ll_root_fid);
 464         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 465         if (err) {
 466                 CERROR("cannot mds_connect: rc = %d\n", err);
 467                 goto out_lock_cn_cb;
 468         }
 469         if (!fid_is_sane(&sbi->ll_root_fid)) {
 470                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 471                        sbi->ll_md_exp->exp_obd->obd_name,
 472                        PFID(&sbi->ll_root_fid));
 473                 err = -EINVAL;
 474                 goto out_lock_cn_cb;
 475         }
 476         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 477
 478         sb->s_op = &lustre_super_operations;
 479 #if THREAD_SIZE >= 8192 /*b=17630*/
 480         sb->s_export_op = &lustre_export_operations;
 481 #endif
 482
 483         /* make root inode
 484          * XXX: move this to after cbd setup? */
 485         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 486         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 487                 valid |= OBD_MD_FLRMTPERM;
 488         else if (sbi->ll_flags & LL_SBI_ACL)
 489                 valid |= OBD_MD_FLACL;
 490
 491         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
 492         if (!op_data) {
 493                 err = -ENOMEM;
 494                 goto out_lock_cn_cb;
 495         }
 496
 497         op_data->op_fid1 = sbi->ll_root_fid;
 498         op_data->op_mode = 0;
 499         op_data->op_capa1 = oc;
 500         op_data->op_valid = valid;
 501
 502         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 503         if (oc)
 504                 capa_put(oc);
 505         OBD_FREE_PTR(op_data);
 506         if (err) {
 507                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 508                        sbi->ll_md_exp->exp_obd->obd_name, err);
 509                 goto out_lock_cn_cb;
 510         }
 511
 512         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 513                                sbi->ll_md_exp, &lmd);
 514         if (err) {
 515                 CERROR("failed to understand root inode md: rc = %d\n", err);
 516                 ptlrpc_req_finished(request);
 517                 goto out_lock_cn_cb;
 518         }
 519
 520         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 521         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 522                                             sbi->ll_flags & LL_SBI_32BIT_API),
 523                        &lmd);
 524         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 525         ptlrpc_req_finished(request);
 526
 527         if (root == NULL || IS_ERR(root)) {
 528                 if (lmd.lsm)
 529                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 530 #ifdef CONFIG_FS_POSIX_ACL
 531                 if (lmd.posix_acl) {
 532                         posix_acl_release(lmd.posix_acl);
 533                         lmd.posix_acl = NULL;
 534                 }
 535 #endif
 536                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 537                 root = NULL;
 538                 CERROR("lustre_lite: bad iget4 for root\n");
 539                 goto out_root;
 540         }
 541
 542         err = ll_close_thread_start(&sbi->ll_lcq);
 543         if (err) {
 544                 CERROR("cannot start close thread: rc %d\n", err);
 545                 goto out_root;
 546         }
 547
 548 #ifdef CONFIG_FS_POSIX_ACL
 549         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 550                 rct_init(&sbi->ll_rct);
 551                 et_init(&sbi->ll_et);
 552         }
 553 #endif
 554
 555         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 556         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 557                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 558                                  NULL);
 559         cl_sb_init(sb);
 560
 561         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 562                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 563                                  &sbi->ll_cache, NULL);
 564
 565         sb->s_root = d_make_root(root);
 566         if (sb->s_root == NULL) {
 567                 CERROR("%s: can't make root dentry\n",
 568                         ll_get_fsname(sb, NULL, 0));
 569                 err = -ENOMEM;
 570                 goto out_lock_cn_cb;
 571         }
 572
 573         sbi->ll_sdev_orig = sb->s_dev;
 574
 575         /* We set sb->s_dev equal on all lustre clients in order to support
 576          * NFS export clustering.  NFSD requires that the FSID be the same
 577          * on all clients. */
 578         /* s_dev is also used in lt_compare() to compare two fs, but that is
 579          * only a node-local comparison. */
 580         uuid = obd_get_uuid(sbi->ll_md_exp);
 581         if (uuid != NULL) {
 582                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 583                 get_uuid2fsid(uuid->uuid, strlen(uuid->uuid), &sbi->ll_fsid);
 584         }
 585
 586         if (data != NULL)
 587                 OBD_FREE_PTR(data);
 588         if (osfs != NULL)
 589                 OBD_FREE_PTR(osfs);
 590
 591         return err;
 592 out_root:
 593         iput(root);
 594 out_lock_cn_cb:
 595         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 596 out_dt:
 597         obd_disconnect(sbi->ll_dt_exp);
 598         sbi->ll_dt_exp = NULL;
 599         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 600         obd_zombie_barrier();
 601 out_md_fid:
 602         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 603 out_md:
 604         obd_disconnect(sbi->ll_md_exp);
 605         sbi->ll_md_exp = NULL;
 606 out:
 607         if (data != NULL)
 608                 OBD_FREE_PTR(data);
 609         if (osfs != NULL)
 610                 OBD_FREE_PTR(osfs);
 611         lprocfs_unregister_mountpoint(sbi);
 612         return err;
 613 }
 614
 615 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 616 {
 617         int size, rc;
 618
 619         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 620         size = sizeof(int);
 621         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 622                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 623         if (rc)
 624                 CERROR("Get max mdsize error rc %d \n", rc);
 625
 626         return rc;
 627 }
 628
 629 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 630 {
 631         int size, rc;
 632
 633         size = sizeof(int);
 634         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 635                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 636         if (rc)
 637                 CERROR("Get default mdsize error rc %d\n", rc);
 638
 639         return rc;
 640 }
 641
 642 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 643 {
 644         int size, rc;
 645
 646         size = sizeof(int);
 647         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 648                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 649         if (rc)
 650                 CERROR("Get max cookiesize error rc %d\n", rc);
 651
 652         return rc;
 653 }
 654
 655 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 656 {
 657         int size, rc;
 658
 659         size = sizeof(int);
 660         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 661                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 662         if (rc)
 663                 CERROR("Get default cookiesize error rc %d\n", rc);
 664
 665         return rc;
 666 }
 667
 668 static void ll_dump_inode(struct inode *inode)
 669 {
 670         struct ll_d_hlist_node *tmp;
 671         int dentry_count = 0;
 672
 673         LASSERT(inode != NULL);
 674
 675         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 676                 dentry_count++;
 677
 678         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 679                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 680                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 681 }
 682
 683 void lustre_dump_dentry(struct dentry *dentry, int recur)
 684 {
 685         struct list_head *tmp;
 686         int subdirs = 0;
 687
 688         LASSERT(dentry != NULL);
 689
 690         list_for_each(tmp, &dentry->d_subdirs)
 691                 subdirs++;
 692
 693         CERROR("dentry %p dump: name=%pd parent=%pd (%p), inode=%p, count=%u, flags=0x%x, fsdata=%p, %d subdirs\n",
 694                dentry, dentry, dentry->d_parent, dentry->d_parent,
 695                dentry->d_inode, d_count(dentry),
 696                dentry->d_flags, dentry->d_fsdata, subdirs);
 697         if (dentry->d_inode != NULL)
 698                 ll_dump_inode(dentry->d_inode);
 699
 700         if (recur == 0)
 701                 return;
 702
 703         list_for_each(tmp, &dentry->d_subdirs) {
 704                 struct dentry *d = list_entry(tmp, struct dentry, d_child);
 705
 706                 lustre_dump_dentry(d, recur - 1);
 707         }
 708 }
 709
 710 static void client_common_put_super(struct super_block *sb)
 711 {
 712         struct ll_sb_info *sbi = ll_s2sbi(sb);
 713
 714 #ifdef CONFIG_FS_POSIX_ACL
 715         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 716                 et_fini(&sbi->ll_et);
 717                 rct_fini(&sbi->ll_rct);
 718         }
 719 #endif
 720
 721         ll_close_thread_shutdown(sbi->ll_lcq);
 722
 723         cl_sb_fini(sb);
 724
 725         list_del(&sbi->ll_conn_chain);
 726
 727         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 728         obd_disconnect(sbi->ll_dt_exp);
 729         sbi->ll_dt_exp = NULL;
 730         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 731          * see LU-2543. */
 732         obd_zombie_barrier();
 733
 734         lprocfs_unregister_mountpoint(sbi);
 735
 736         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 737         obd_disconnect(sbi->ll_md_exp);
 738         sbi->ll_md_exp = NULL;
 739 }
 740
 741 void ll_kill_super(struct super_block *sb)
 742 {
 743         struct ll_sb_info *sbi;
 744
 745         /* not init sb ?*/
 746         if (!(sb->s_flags & MS_ACTIVE))
 747                 return;
 748
 749         sbi = ll_s2sbi(sb);
 750         /* we need to restore s_dev from changed for clustered NFS before
 751          * put_super because new kernels have cached s_dev and change sb->s_dev
 752          * in put_super not affected real removing devices */
 753         if (sbi) {
 754                 sb->s_dev = sbi->ll_sdev_orig;
 755                 sbi->ll_umounting = 1;
 756         }
 757 }
 758
 759 static inline int ll_set_opt(const char *opt, char *data, int fl)
 760 {
 761         if (strncmp(opt, data, strlen(opt)) != 0)
 762                 return 0;
 763         else
 764                 return fl;
 765 }
 766
 767 /* non-client-specific mount options are parsed in lmd_parse */
 768 static int ll_options(char *options, int *flags)
 769 {
 770         int tmp;
 771         char *s1 = options, *s2;
 772
 773         if (!options)
 774                 return 0;
 775
 776         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 777
 778         while (*s1) {
 779                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 780                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 781                 if (tmp) {
 782                         *flags |= tmp;
 783                         goto next;
 784                 }
 785                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 786                 if (tmp) {
 787                         *flags |= tmp;
 788                         goto next;
 789                 }
 790                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 791                 if (tmp) {
 792                         *flags |= tmp;
 793                         goto next;
 794                 }
 795                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 796                 if (tmp) {
 797                         *flags &= ~tmp;
 798                         goto next;
 799                 }
 800                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 801                 if (tmp) {
 802                         *flags |= tmp;
 803                         goto next;
 804                 }
 805                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 806                 if (tmp) {
 807                         *flags &= ~tmp;
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 811                 if (tmp) {
 812                         *flags |= tmp;
 813                         goto next;
 814                 }
 815                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 816                 if (tmp) {
 817                         *flags |= tmp;
 818                         goto next;
 819                 }
 820                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 821                 if (tmp) {
 822                         *flags &= ~tmp;
 823                         goto next;
 824                 }
 825
 826                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 827                 if (tmp) {
 828                         *flags |= tmp;
 829                         goto next;
 830                 }
 831                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 832                 if (tmp) {
 833                         *flags &= ~tmp;
 834                         goto next;
 835                 }
 836                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 837                 if (tmp) {
 838                         *flags |= tmp;
 839                         goto next;
 840                 }
 841                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 842                 if (tmp) {
 843                         *flags &= ~tmp;
 844                         goto next;
 845                 }
 846                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 847                 if (tmp) {
 848                         *flags |= tmp;
 849                         goto next;
 850                 }
 851                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 852                 if (tmp) {
 853                         *flags &= ~tmp;
 854                         goto next;
 855                 }
 856                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 862                 if (tmp) {
 863                         *flags |= tmp;
 864                         goto next;
 865                 }
 866                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 867                 if (tmp) {
 868                         *flags |= tmp;
 869                         goto next;
 870                 }
 871                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 872                 if (tmp) {
 873                         *flags &= ~tmp;
 874                         goto next;
 875                 }
 876                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 877                                    s1);
 878                 return -EINVAL;
 879
 880 next:
 881                 /* Find next opt */
 882                 s2 = strchr(s1, ',');
 883                 if (s2 == NULL)
 884                         break;
 885                 s1 = s2 + 1;
 886         }
 887         return 0;
 888 }
 889
 890 void ll_lli_init(struct ll_inode_info *lli)
 891 {
 892         lli->lli_inode_magic = LLI_INODE_MAGIC;
 893         lli->lli_flags = 0;
 894         lli->lli_ioepoch = 0;
 895         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 896         spin_lock_init(&lli->lli_lock);
 897         lli->lli_posix_acl = NULL;
 898         lli->lli_remote_perms = NULL;
 899         mutex_init(&lli->lli_rmtperm_mutex);
 900         /* Do not set lli_fid, it has been initialized already. */
 901         fid_zero(&lli->lli_pfid);
 902         INIT_LIST_HEAD(&lli->lli_close_list);
 903         INIT_LIST_HEAD(&lli->lli_oss_capas);
 904         atomic_set(&lli->lli_open_count, 0);
 905         lli->lli_mds_capa = NULL;
 906         lli->lli_rmtperm_time = 0;
 907         lli->lli_pending_och = NULL;
 908         lli->lli_mds_read_och = NULL;
 909         lli->lli_mds_write_och = NULL;
 910         lli->lli_mds_exec_och = NULL;
 911         lli->lli_open_fd_read_count = 0;
 912         lli->lli_open_fd_write_count = 0;
 913         lli->lli_open_fd_exec_count = 0;
 914         mutex_init(&lli->lli_och_mutex);
 915         spin_lock_init(&lli->lli_agl_lock);
 916         lli->lli_has_smd = false;
 917         spin_lock_init(&lli->lli_layout_lock);
 918         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 919         lli->lli_clob = NULL;
 920
 921         init_rwsem(&lli->lli_xattrs_list_rwsem);
 922         mutex_init(&lli->lli_xattrs_enq_lock);
 923
 924         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 925         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 926                 mutex_init(&lli->lli_readdir_mutex);
 927                 lli->lli_opendir_key = NULL;
 928                 lli->lli_sai = NULL;
 929                 spin_lock_init(&lli->lli_sa_lock);
 930                 lli->lli_opendir_pid = 0;
 931         } else {
 932                 mutex_init(&lli->lli_size_mutex);
 933                 lli->lli_symlink_name = NULL;
 934                 init_rwsem(&lli->lli_trunc_sem);
 935                 mutex_init(&lli->lli_write_mutex);
 936                 init_rwsem(&lli->lli_glimpse_sem);
 937                 lli->lli_glimpse_time = 0;
 938                 INIT_LIST_HEAD(&lli->lli_agl_list);
 939                 lli->lli_agl_index = 0;
 940                 lli->lli_async_rc = 0;
 941         }
 942         mutex_init(&lli->lli_layout_mutex);
 943 }
 944
 945 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 946 {
 947         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 948
 949         bdi->name = "lustre";
 950         return bdi_register(bdi, NULL, "lustre-%d",
 951                             atomic_inc_return(&ll_bdi_num));
 952 }
 953
 954 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 955 {
 956         struct lustre_profile *lprof = NULL;
 957         struct lustre_sb_info *lsi = s2lsi(sb);
 958         struct ll_sb_info *sbi;
 959         char  *dt = NULL, *md = NULL;
 960         char  *profilenm = get_profile_name(sb);
 961         struct config_llog_instance *cfg;
 962         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 963         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 964         int    err;
 965
 966         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 967
 968         cfg = kzalloc(sizeof(*cfg), GFP_NOFS);
 969         if (!cfg)
 970                 return -ENOMEM;
 971
 972         try_module_get(THIS_MODULE);
 973
 974         /* client additional sb info */
 975         lsi->lsi_llsbi = sbi = ll_init_sbi();
 976         if (!sbi) {
 977                 module_put(THIS_MODULE);
 978                 OBD_FREE_PTR(cfg);
 979                 return -ENOMEM;
 980         }
 981
 982         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 983         if (err)
 984                 goto out_free;
 985
 986         err = bdi_init(&lsi->lsi_bdi);
 987         if (err)
 988                 goto out_free;
 989         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
 990         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
 991         err = ll_bdi_register(&lsi->lsi_bdi);
 992         if (err)
 993                 goto out_free;
 994
 995         sb->s_bdi = &lsi->lsi_bdi;
 996         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 997         sb->s_d_op = &ll_d_ops;
 998
 999         /* Generate a string unique to this super, in case some joker tries
1000            to mount the same fs at two mount points.
1001            Use the address of the super itself.*/
1002         cfg->cfg_instance = sb;
1003         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1004         cfg->cfg_callback = class_config_llog_handler;
1005         /* set up client obds */
1006         err = lustre_process_log(sb, profilenm, cfg);
1007         if (err < 0) {
1008                 CERROR("Unable to process log: %d\n", err);
1009                 goto out_free;
1010         }
1011
1012         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1013         lprof = class_get_profile(profilenm);
1014         if (lprof == NULL) {
1015                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be read from the MGS.  Does that filesystem exist?\n",
1016                                    profilenm);
1017                 err = -EINVAL;
1018                 goto out_free;
1019         }
1020         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1021                lprof->lp_md, lprof->lp_dt);
1022
1023         dt = kzalloc(strlen(lprof->lp_dt) + instlen + 2, GFP_NOFS);
1024         if (!dt) {
1025                 err = -ENOMEM;
1026                 goto out_free;
1027         }
1028         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1029
1030         md = kzalloc(strlen(lprof->lp_md) + instlen + 2, GFP_NOFS);
1031         if (!md) {
1032                 err = -ENOMEM;
1033                 goto out_free;
1034         }
1035         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1036
1037         /* connections, registrations, sb setup */
1038         err = client_common_fill_super(sb, md, dt, mnt);
1039
1040 out_free:
1041         if (md)
1042                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1043         if (dt)
1044                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1045         if (err)
1046                 ll_put_super(sb);
1047         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1048                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1049
1050         OBD_FREE_PTR(cfg);
1051         return err;
1052 } /* ll_fill_super */
1053
1054 void ll_put_super(struct super_block *sb)
1055 {
1056         struct config_llog_instance cfg, params_cfg;
1057         struct obd_device *obd;
1058         struct lustre_sb_info *lsi = s2lsi(sb);
1059         struct ll_sb_info *sbi = ll_s2sbi(sb);
1060         char *profilenm = get_profile_name(sb);
1061         int next, force = 1;
1062
1063         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1064
1065         ll_print_capa_stat(sbi);
1066
1067         cfg.cfg_instance = sb;
1068         lustre_end_log(sb, profilenm, &cfg);
1069
1070         params_cfg.cfg_instance = sb;
1071         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1072
1073         if (sbi->ll_md_exp) {
1074                 obd = class_exp2obd(sbi->ll_md_exp);
1075                 if (obd)
1076                         force = obd->obd_force;
1077         }
1078
1079         /* We need to set force before the lov_disconnect in
1080            lustre_common_put_super, since l_d cleans up osc's as well. */
1081         if (force) {
1082                 next = 0;
1083                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1084                                                      &next)) != NULL) {
1085                         obd->obd_force = force;
1086                 }
1087         }
1088
1089         if (sbi->ll_lcq) {
1090                 /* Only if client_common_fill_super succeeded */
1091                 client_common_put_super(sb);
1092         }
1093
1094         next = 0;
1095         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1096                 class_manual_cleanup(obd);
1097
1098         if (sbi->ll_flags & LL_SBI_VERBOSE)
1099                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1100
1101         if (profilenm)
1102                 class_del_profile(profilenm);
1103
1104         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1105                 bdi_destroy(&lsi->lsi_bdi);
1106                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1107         }
1108
1109         ll_free_sbi(sb);
1110         lsi->lsi_llsbi = NULL;
1111
1112         lustre_common_put_super(sb);
1113
1114         module_put(THIS_MODULE);
1115 } /* client_put_super */
1116
1117 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1118 {
1119         struct inode *inode = NULL;
1120
1121         /* NOTE: we depend on atomic igrab() -bzzz */
1122         lock_res_and_lock(lock);
1123         if (lock->l_resource->lr_lvb_inode) {
1124                 struct ll_inode_info *lli;
1125
1126                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1127                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1128                         inode = igrab(lock->l_resource->lr_lvb_inode);
1129                 } else {
1130                         inode = lock->l_resource->lr_lvb_inode;
1131                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1132                                          D_WARNING, lock, "lr_lvb_inode %p is bogus: magic %08x",
1133                                          lock->l_resource->lr_lvb_inode,
1134                                          lli->lli_inode_magic);
1135                         inode = NULL;
1136                 }
1137         }
1138         unlock_res_and_lock(lock);
1139         return inode;
1140 }
1141
1142 void ll_clear_inode(struct inode *inode)
1143 {
1144         struct ll_inode_info *lli = ll_i2info(inode);
1145         struct ll_sb_info *sbi = ll_i2sbi(inode);
1146
1147         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1148                inode->i_generation, inode);
1149
1150         if (S_ISDIR(inode->i_mode)) {
1151                 /* these should have been cleared in ll_file_release */
1152                 LASSERT(lli->lli_opendir_key == NULL);
1153                 LASSERT(lli->lli_sai == NULL);
1154                 LASSERT(lli->lli_opendir_pid == 0);
1155         }
1156
1157         spin_lock(&lli->lli_lock);
1158         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1159         spin_unlock(&lli->lli_lock);
1160         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1161
1162         LASSERT(!lli->lli_open_fd_write_count);
1163         LASSERT(!lli->lli_open_fd_read_count);
1164         LASSERT(!lli->lli_open_fd_exec_count);
1165
1166         if (lli->lli_mds_write_och)
1167                 ll_md_real_close(inode, FMODE_WRITE);
1168         if (lli->lli_mds_exec_och)
1169                 ll_md_real_close(inode, FMODE_EXEC);
1170         if (lli->lli_mds_read_och)
1171                 ll_md_real_close(inode, FMODE_READ);
1172
1173         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1174                 OBD_FREE(lli->lli_symlink_name,
1175                          strlen(lli->lli_symlink_name) + 1);
1176                 lli->lli_symlink_name = NULL;
1177         }
1178
1179         ll_xattr_cache_destroy(inode);
1180
1181         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1182                 LASSERT(lli->lli_posix_acl == NULL);
1183                 if (lli->lli_remote_perms) {
1184                         free_rmtperm_hash(lli->lli_remote_perms);
1185                         lli->lli_remote_perms = NULL;
1186                 }
1187         }
1188 #ifdef CONFIG_FS_POSIX_ACL
1189         else if (lli->lli_posix_acl) {
1190                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1191                 LASSERT(lli->lli_remote_perms == NULL);
1192                 posix_acl_release(lli->lli_posix_acl);
1193                 lli->lli_posix_acl = NULL;
1194         }
1195 #endif
1196         lli->lli_inode_magic = LLI_INODE_DEAD;
1197
1198         ll_clear_inode_capas(inode);
1199         if (!S_ISDIR(inode->i_mode))
1200                 LASSERT(list_empty(&lli->lli_agl_list));
1201
1202         /*
1203          * XXX This has to be done before lsm is freed below, because
1204          * cl_object still uses inode lsm.
1205          */
1206         cl_inode_fini(inode);
1207         lli->lli_has_smd = false;
1208 }
1209
1210 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1211                   struct md_open_data **mod)
1212 {
1213         struct lustre_md md;
1214         struct inode *inode = dentry->d_inode;
1215         struct ll_sb_info *sbi = ll_i2sbi(inode);
1216         struct ptlrpc_request *request = NULL;
1217         int rc, ia_valid;
1218
1219         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1220                                      LUSTRE_OPC_ANY, NULL);
1221         if (IS_ERR(op_data))
1222                 return PTR_ERR(op_data);
1223
1224         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1225                         &request, mod);
1226         if (rc) {
1227                 ptlrpc_req_finished(request);
1228                 if (rc == -ENOENT) {
1229                         clear_nlink(inode);
1230                         /* Unlinked special device node? Or just a race?
1231                          * Pretend we done everything. */
1232                         if (!S_ISREG(inode->i_mode) &&
1233                             !S_ISDIR(inode->i_mode)) {
1234                                 ia_valid = op_data->op_attr.ia_valid;
1235                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1236                                 rc = simple_setattr(dentry, &op_data->op_attr);
1237                                 op_data->op_attr.ia_valid = ia_valid;
1238                         }
1239                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1240                         CERROR("md_setattr fails: rc = %d\n", rc);
1241                 }
1242                 return rc;
1243         }
1244
1245         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1246                               sbi->ll_md_exp, &md);
1247         if (rc) {
1248                 ptlrpc_req_finished(request);
1249                 return rc;
1250         }
1251
1252         ia_valid = op_data->op_attr.ia_valid;
1253         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1254          * cache is not cleared yet. */
1255         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1256         rc = simple_setattr(dentry, &op_data->op_attr);
1257         op_data->op_attr.ia_valid = ia_valid;
1258
1259         /* Extract epoch data if obtained. */
1260         op_data->op_handle = md.body->handle;
1261         op_data->op_ioepoch = md.body->ioepoch;
1262
1263         ll_update_inode(inode, &md);
1264         ptlrpc_req_finished(request);
1265
1266         return rc;
1267 }
1268
1269 /* Close IO epoch and send Size-on-MDS attribute update. */
1270 static int ll_setattr_done_writing(struct inode *inode,
1271                                    struct md_op_data *op_data,
1272                                    struct md_open_data *mod)
1273 {
1274         struct ll_inode_info *lli = ll_i2info(inode);
1275         int rc = 0;
1276
1277         LASSERT(op_data != NULL);
1278         if (!S_ISREG(inode->i_mode))
1279                 return 0;
1280
1281         CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
1282                op_data->op_ioepoch, PFID(&lli->lli_fid));
1283
1284         op_data->op_flags = MF_EPOCH_CLOSE;
1285         ll_done_writing_attr(inode, op_data);
1286         ll_pack_inode2opdata(inode, op_data, NULL);
1287
1288         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1289         if (rc == -EAGAIN) {
1290                 /* MDS has instructed us to obtain Size-on-MDS attribute
1291                  * from OSTs and send setattr to back to MDS. */
1292                 rc = ll_som_update(inode, op_data);
1293         } else if (rc) {
1294                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1295                        inode->i_ino, rc);
1296         }
1297         return rc;
1298 }
1299
1300 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1301 {
1302         struct obd_capa *capa;
1303         int rc;
1304
1305         if (attr->ia_valid & ATTR_SIZE)
1306                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1307         else
1308                 capa = ll_mdscapa_get(inode);
1309
1310         rc = cl_setattr_ost(inode, attr, capa);
1311
1312         if (attr->ia_valid & ATTR_SIZE)
1313                 ll_truncate_free_capa(capa);
1314         else
1315                 capa_put(capa);
1316
1317         return rc;
1318 }
1319
1320
1321 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1322  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1323  * keep these values until such a time that objects are allocated for it.
1324  * We do the MDS operations first, as it is checking permissions for us.
1325  * We don't to the MDS RPC if there is nothing that we want to store there,
1326  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1327  * going to do an RPC anyways.
1328  *
1329  * If we are doing a truncate, we will send the mtime and ctime updates
1330  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1331  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1332  * at the same time.
1333  *
1334  * In case of HSMimport, we only set attr on MDS.
1335  */
1336 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1337 {
1338         struct inode *inode = dentry->d_inode;
1339         struct ll_inode_info *lli = ll_i2info(inode);
1340         struct md_op_data *op_data = NULL;
1341         struct md_open_data *mod = NULL;
1342         bool file_is_released = false;
1343         int rc = 0, rc1 = 0;
1344
1345         CDEBUG(D_VFSTRACE,
1346                 "%s: setattr inode %p/fid:"DFID
1347                 " from %llu to %llu, valid %x, hsm_import %d\n",
1348                 ll_get_fsname(inode->i_sb, NULL, 0), inode,
1349                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1350                 attr->ia_valid, hsm_import);
1351
1352         if (attr->ia_valid & ATTR_SIZE) {
1353                 /* Check new size against VFS/VM file size limit and rlimit */
1354                 rc = inode_newsize_ok(inode, attr->ia_size);
1355                 if (rc)
1356                         return rc;
1357
1358                 /* The maximum Lustre file size is variable, based on the
1359                  * OST maximum object size and number of stripes.  This
1360                  * needs another check in addition to the VFS check above. */
1361                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1362                         CDEBUG(D_INODE, "file "DFID" too large %llu > %llu\n",
1363                                PFID(&lli->lli_fid), attr->ia_size,
1364                                ll_file_maxbytes(inode));
1365                         return -EFBIG;
1366                 }
1367
1368                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1369         }
1370
1371         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1372         if (attr->ia_valid & TIMES_SET_FLAGS) {
1373                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1374                     !capable(CFS_CAP_FOWNER))
1375                         return -EPERM;
1376         }
1377
1378         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1379         if (attr->ia_valid & ATTR_CTIME) {
1380                 attr->ia_ctime = CURRENT_TIME;
1381                 attr->ia_valid |= ATTR_CTIME_SET;
1382         }
1383         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1384             (attr->ia_valid & ATTR_ATIME)) {
1385                 attr->ia_atime = CURRENT_TIME;
1386                 attr->ia_valid |= ATTR_ATIME_SET;
1387         }
1388         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1389             (attr->ia_valid & ATTR_MTIME)) {
1390                 attr->ia_mtime = CURRENT_TIME;
1391                 attr->ia_valid |= ATTR_MTIME_SET;
1392         }
1393
1394         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1395                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1396                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1397                        get_seconds());
1398
1399         /* If we are changing file size, file content is modified, flag it. */
1400         if (attr->ia_valid & ATTR_SIZE) {
1401                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1402                 spin_lock(&lli->lli_lock);
1403                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1404                 spin_unlock(&lli->lli_lock);
1405         }
1406
1407         /* We always do an MDS RPC, even if we're only changing the size;
1408          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1409
1410         op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
1411         if (!op_data)
1412                 return -ENOMEM;
1413
1414         if (!S_ISDIR(inode->i_mode)) {
1415                 if (attr->ia_valid & ATTR_SIZE)
1416                         inode_dio_write_done(inode);
1417                 mutex_unlock(&inode->i_mutex);
1418         }
1419
1420         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1421
1422         /* Open epoch for truncate. */
1423         if (exp_connect_som(ll_i2mdexp(inode)) &&
1424             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1425                 op_data->op_flags = MF_EPOCH_OPEN;
1426
1427         /* truncate on a released file must failed with -ENODATA,
1428          * so size must not be set on MDS for released file
1429          * but other attributes must be set
1430          */
1431         if (S_ISREG(inode->i_mode)) {
1432                 struct lov_stripe_md *lsm;
1433                 __u32 gen;
1434
1435                 ll_layout_refresh(inode, &gen);
1436                 lsm = ccc_inode_lsm_get(inode);
1437                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1438                         file_is_released = true;
1439                 ccc_inode_lsm_put(inode, lsm);
1440         }
1441
1442         /* if not in HSM import mode, clear size attr for released file
1443          * we clear the attribute send to MDT in op_data, not the original
1444          * received from caller in attr which is used later to
1445          * decide return code */
1446         if (file_is_released && (attr->ia_valid & ATTR_SIZE) && !hsm_import)
1447                 op_data->op_attr.ia_valid &= ~ATTR_SIZE;
1448
1449         rc = ll_md_setattr(dentry, op_data, &mod);
1450         if (rc)
1451                 goto out;
1452
1453         /* truncate failed (only when non HSM import), others succeed */
1454         if (file_is_released) {
1455                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1456                         rc = -ENODATA;
1457                 else
1458                         rc = 0;
1459                 goto out;
1460         }
1461
1462         /* RPC to MDT is sent, cancel data modification flag */
1463         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1464                 spin_lock(&lli->lli_lock);
1465                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1466                 spin_unlock(&lli->lli_lock);
1467         }
1468
1469         ll_ioepoch_open(lli, op_data->op_ioepoch);
1470         if (!S_ISREG(inode->i_mode)) {
1471                 rc = 0;
1472                 goto out;
1473         }
1474
1475         if (attr->ia_valid & (ATTR_SIZE |
1476                               ATTR_ATIME | ATTR_ATIME_SET |
1477                               ATTR_MTIME | ATTR_MTIME_SET))
1478                 /* For truncate and utimes sending attributes to OSTs, setting
1479                  * mtime/atime to the past will be performed under PW [0:EOF]
1480                  * extent lock (new_size:EOF for truncate).  It may seem
1481                  * excessive to send mtime/atime updates to OSTs when not
1482                  * setting times to past, but it is necessary due to possible
1483                  * time de-synchronization between MDT inode and OST objects */
1484                 if (attr->ia_valid & ATTR_SIZE)
1485                         down_write(&lli->lli_trunc_sem);
1486                 rc = ll_setattr_ost(inode, attr);
1487                 if (attr->ia_valid & ATTR_SIZE)
1488                         up_write(&lli->lli_trunc_sem);
1489 out:
1490         if (op_data) {
1491                 if (op_data->op_ioepoch) {
1492                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1493                         if (!rc)
1494                                 rc = rc1;
1495                 }
1496                 ll_finish_md_op_data(op_data);
1497         }
1498         if (!S_ISDIR(inode->i_mode)) {
1499                 mutex_lock(&inode->i_mutex);
1500                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1501                         inode_dio_wait(inode);
1502         }
1503
1504         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1505                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1506
1507         return rc;
1508 }
1509
1510 int ll_setattr(struct dentry *de, struct iattr *attr)
1511 {
1512         int mode = de->d_inode->i_mode;
1513
1514         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1515                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1516                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1517
1518         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1519                                (ATTR_SIZE|ATTR_MODE)) &&
1520             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1521              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1522               !(attr->ia_mode & S_ISGID))))
1523                 attr->ia_valid |= ATTR_FORCE;
1524
1525         if ((attr->ia_valid & ATTR_MODE) &&
1526             (mode & S_ISUID) &&
1527             !(attr->ia_mode & S_ISUID) &&
1528             !(attr->ia_valid & ATTR_KILL_SUID))
1529                 attr->ia_valid |= ATTR_KILL_SUID;
1530
1531         if ((attr->ia_valid & ATTR_MODE) &&
1532             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1533             !(attr->ia_mode & S_ISGID) &&
1534             !(attr->ia_valid & ATTR_KILL_SGID))
1535                 attr->ia_valid |= ATTR_KILL_SGID;
1536
1537         return ll_setattr_raw(de, attr, false);
1538 }
1539
1540 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1541                        __u64 max_age, __u32 flags)
1542 {
1543         struct ll_sb_info *sbi = ll_s2sbi(sb);
1544         struct obd_statfs obd_osfs;
1545         int rc;
1546
1547         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1548         if (rc) {
1549                 CERROR("md_statfs fails: rc = %d\n", rc);
1550                 return rc;
1551         }
1552
1553         osfs->os_type = sb->s_magic;
1554
1555         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1556                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,
1557                osfs->os_files);
1558
1559         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1560                 flags |= OBD_STATFS_NODELAY;
1561
1562         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1563         if (rc) {
1564                 CERROR("obd_statfs fails: rc = %d\n", rc);
1565                 return rc;
1566         }
1567
1568         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1569                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1570                obd_osfs.os_files);
1571
1572         osfs->os_bsize = obd_osfs.os_bsize;
1573         osfs->os_blocks = obd_osfs.os_blocks;
1574         osfs->os_bfree = obd_osfs.os_bfree;
1575         osfs->os_bavail = obd_osfs.os_bavail;
1576
1577         /* If we don't have as many objects free on the OST as inodes
1578          * on the MDS, we reduce the total number of inodes to
1579          * compensate, so that the "inodes in use" number is correct.
1580          */
1581         if (obd_osfs.os_ffree < osfs->os_ffree) {
1582                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1583                         obd_osfs.os_ffree;
1584                 osfs->os_ffree = obd_osfs.os_ffree;
1585         }
1586
1587         return rc;
1588 }
1589 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1590 {
1591         struct super_block *sb = de->d_sb;
1592         struct obd_statfs osfs;
1593         int rc;
1594
1595         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1596         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1597
1598         /* Some amount of caching on the client is allowed */
1599         rc = ll_statfs_internal(sb, &osfs,
1600                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1601                                 0);
1602         if (rc)
1603                 return rc;
1604
1605         statfs_unpack(sfs, &osfs);
1606
1607         /* We need to downshift for all 32-bit kernels, because we can't
1608          * tell if the kernel is being called via sys_statfs64() or not.
1609          * Stop before overflowing f_bsize - in which case it is better
1610          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1611         if (sizeof(long) < 8) {
1612                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1613                         sfs->f_bsize <<= 1;
1614
1615                         osfs.os_blocks >>= 1;
1616                         osfs.os_bfree >>= 1;
1617                         osfs.os_bavail >>= 1;
1618                 }
1619         }
1620
1621         sfs->f_blocks = osfs.os_blocks;
1622         sfs->f_bfree = osfs.os_bfree;
1623         sfs->f_bavail = osfs.os_bavail;
1624         sfs->f_fsid = ll_s2sbi(sb)->ll_fsid;
1625         return 0;
1626 }
1627
1628 void ll_inode_size_lock(struct inode *inode)
1629 {
1630         struct ll_inode_info *lli;
1631
1632         LASSERT(!S_ISDIR(inode->i_mode));
1633
1634         lli = ll_i2info(inode);
1635         mutex_lock(&lli->lli_size_mutex);
1636 }
1637
1638 void ll_inode_size_unlock(struct inode *inode)
1639 {
1640         struct ll_inode_info *lli;
1641
1642         lli = ll_i2info(inode);
1643         mutex_unlock(&lli->lli_size_mutex);
1644 }
1645
1646 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1647 {
1648         struct ll_inode_info *lli = ll_i2info(inode);
1649         struct mdt_body *body = md->body;
1650         struct lov_stripe_md *lsm = md->lsm;
1651         struct ll_sb_info *sbi = ll_i2sbi(inode);
1652
1653         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1654         if (lsm != NULL) {
1655                 if (!lli->lli_has_smd &&
1656                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1657                         cl_file_inode_init(inode, md);
1658
1659                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1660                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1661                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1662         }
1663
1664         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1665                 if (body->valid & OBD_MD_FLRMTPERM)
1666                         ll_update_remote_perm(inode, md->remote_perm);
1667         }
1668 #ifdef CONFIG_FS_POSIX_ACL
1669         else if (body->valid & OBD_MD_FLACL) {
1670                 spin_lock(&lli->lli_lock);
1671                 if (lli->lli_posix_acl)
1672                         posix_acl_release(lli->lli_posix_acl);
1673                 lli->lli_posix_acl = md->posix_acl;
1674                 spin_unlock(&lli->lli_lock);
1675         }
1676 #endif
1677         inode->i_ino = cl_fid_build_ino(&body->fid1,
1678                                         sbi->ll_flags & LL_SBI_32BIT_API);
1679         inode->i_generation = cl_fid_build_gen(&body->fid1);
1680
1681         if (body->valid & OBD_MD_FLATIME) {
1682                 if (body->atime > LTIME_S(inode->i_atime))
1683                         LTIME_S(inode->i_atime) = body->atime;
1684                 lli->lli_lvb.lvb_atime = body->atime;
1685         }
1686         if (body->valid & OBD_MD_FLMTIME) {
1687                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1688                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu to %llu\n",
1689                                inode->i_ino, LTIME_S(inode->i_mtime),
1690                                body->mtime);
1691                         LTIME_S(inode->i_mtime) = body->mtime;
1692                 }
1693                 lli->lli_lvb.lvb_mtime = body->mtime;
1694         }
1695         if (body->valid & OBD_MD_FLCTIME) {
1696                 if (body->ctime > LTIME_S(inode->i_ctime))
1697                         LTIME_S(inode->i_ctime) = body->ctime;
1698                 lli->lli_lvb.lvb_ctime = body->ctime;
1699         }
1700         if (body->valid & OBD_MD_FLMODE)
1701                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1702         if (body->valid & OBD_MD_FLTYPE)
1703                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1704         LASSERT(inode->i_mode != 0);
1705         if (S_ISREG(inode->i_mode))
1706                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1707                                        LL_MAX_BLKSIZE_BITS);
1708         else
1709                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1710         if (body->valid & OBD_MD_FLUID)
1711                 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1712         if (body->valid & OBD_MD_FLGID)
1713                 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1714         if (body->valid & OBD_MD_FLFLAGS)
1715                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1716         if (body->valid & OBD_MD_FLNLINK)
1717                 set_nlink(inode, body->nlink);
1718         if (body->valid & OBD_MD_FLRDEV)
1719                 inode->i_rdev = old_decode_dev(body->rdev);
1720
1721         if (body->valid & OBD_MD_FLID) {
1722                 /* FID shouldn't be changed! */
1723                 if (fid_is_sane(&lli->lli_fid)) {
1724                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1725                                  "Trying to change FID "DFID
1726                                  " to the "DFID", inode %lu/%u(%p)\n",
1727                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1728                                  inode->i_ino, inode->i_generation, inode);
1729                 } else
1730                         lli->lli_fid = body->fid1;
1731         }
1732
1733         LASSERT(fid_seq(&lli->lli_fid) != 0);
1734
1735         if (body->valid & OBD_MD_FLSIZE) {
1736                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1737                     S_ISREG(inode->i_mode)) {
1738                         struct lustre_handle lockh;
1739                         ldlm_mode_t mode;
1740
1741                         /* As it is possible a blocking ast has been processed
1742                          * by this time, we need to check there is an UPDATE
1743                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1744                          * it. */
1745                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1746                                                &lockh, LDLM_FL_CBPENDING,
1747                                                LCK_CR | LCK_CW |
1748                                                LCK_PR | LCK_PW);
1749                         if (mode) {
1750                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1751                                                       LLIF_EPOCH_PENDING |
1752                                                       LLIF_SOM_DIRTY)) {
1753                                         CERROR("ino %lu flags %u still has size authority! do not trust the size got from MDS\n",
1754                                                inode->i_ino, lli->lli_flags);
1755                                 } else {
1756                                         /* Use old size assignment to avoid
1757                                          * deadlock bz14138 & bz14326 */
1758                                         i_size_write(inode, body->size);
1759                                         spin_lock(&lli->lli_lock);
1760                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1761                                         spin_unlock(&lli->lli_lock);
1762                                 }
1763                                 ldlm_lock_decref(&lockh, mode);
1764                         }
1765                 } else {
1766                         /* Use old size assignment to avoid
1767                          * deadlock bz14138 & bz14326 */
1768                         i_size_write(inode, body->size);
1769
1770                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1771                                inode->i_ino, (unsigned long long)body->size);
1772                 }
1773
1774                 if (body->valid & OBD_MD_FLBLOCKS)
1775                         inode->i_blocks = body->blocks;
1776         }
1777
1778         if (body->valid & OBD_MD_FLMDSCAPA) {
1779                 LASSERT(md->mds_capa);
1780                 ll_add_capa(inode, md->mds_capa);
1781         }
1782         if (body->valid & OBD_MD_FLOSSCAPA) {
1783                 LASSERT(md->oss_capa);
1784                 ll_add_capa(inode, md->oss_capa);
1785         }
1786
1787         if (body->valid & OBD_MD_TSTATE) {
1788                 if (body->t_state & MS_RESTORE)
1789                         lli->lli_flags |= LLIF_FILE_RESTORING;
1790         }
1791 }
1792
1793 void ll_read_inode2(struct inode *inode, void *opaque)
1794 {
1795         struct lustre_md *md = opaque;
1796         struct ll_inode_info *lli = ll_i2info(inode);
1797
1798         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1799                PFID(&lli->lli_fid), inode);
1800
1801         LASSERT(!lli->lli_has_smd);
1802
1803         /* Core attributes from the MDS first.  This is a new inode, and
1804          * the VFS doesn't zero times in the core inode so we have to do
1805          * it ourselves.  They will be overwritten by either MDS or OST
1806          * attributes - we just need to make sure they aren't newer. */
1807         LTIME_S(inode->i_mtime) = 0;
1808         LTIME_S(inode->i_atime) = 0;
1809         LTIME_S(inode->i_ctime) = 0;
1810         inode->i_rdev = 0;
1811         ll_update_inode(inode, md);
1812
1813         /* OIDEBUG(inode); */
1814
1815         /* initializing backing dev info. */
1816         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1817
1818
1819         if (S_ISREG(inode->i_mode)) {
1820                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1821
1822                 inode->i_op = &ll_file_inode_operations;
1823                 inode->i_fop = sbi->ll_fop;
1824                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1825         } else if (S_ISDIR(inode->i_mode)) {
1826                 inode->i_op = &ll_dir_inode_operations;
1827                 inode->i_fop = &ll_dir_operations;
1828         } else if (S_ISLNK(inode->i_mode)) {
1829                 inode->i_op = &ll_fast_symlink_inode_operations;
1830         } else {
1831                 inode->i_op = &ll_special_inode_operations;
1832
1833                 init_special_inode(inode, inode->i_mode,
1834                                    inode->i_rdev);
1835         }
1836 }
1837
1838 void ll_delete_inode(struct inode *inode)
1839 {
1840         struct cl_inode_info *lli = cl_i2info(inode);
1841
1842         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1843                 /* discard all dirty pages before truncating them, required by
1844                  * osc_extent implementation at LU-1030. */
1845                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
1846                                    CL_FSYNC_DISCARD, 1);
1847
1848         truncate_inode_pages_final(&inode->i_data);
1849
1850         /* Workaround for LU-118 */
1851         if (inode->i_data.nrpages) {
1852                 spin_lock_irq(&inode->i_data.tree_lock);
1853                 spin_unlock_irq(&inode->i_data.tree_lock);
1854                 LASSERTF(inode->i_data.nrpages == 0,
1855                          "inode=%lu/%u(%p) nrpages=%lu, see http://jira.whamcloud.com/browse/LU-118\n",
1856                          inode->i_ino, inode->i_generation, inode,
1857                          inode->i_data.nrpages);
1858         }
1859         /* Workaround end */
1860
1861         ll_clear_inode(inode);
1862         clear_inode(inode);
1863 }
1864
1865 int ll_iocontrol(struct inode *inode, struct file *file,
1866                  unsigned int cmd, unsigned long arg)
1867 {
1868         struct ll_sb_info *sbi = ll_i2sbi(inode);
1869         struct ptlrpc_request *req = NULL;
1870         int rc, flags = 0;
1871
1872         switch (cmd) {
1873         case FSFILT_IOC_GETFLAGS: {
1874                 struct mdt_body *body;
1875                 struct md_op_data *op_data;
1876
1877                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1878                                              0, 0, LUSTRE_OPC_ANY,
1879                                              NULL);
1880                 if (IS_ERR(op_data))
1881                         return PTR_ERR(op_data);
1882
1883                 op_data->op_valid = OBD_MD_FLFLAGS;
1884                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1885                 ll_finish_md_op_data(op_data);
1886                 if (rc) {
1887                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1888                         return -abs(rc);
1889                 }
1890
1891                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1892
1893                 flags = body->flags;
1894
1895                 ptlrpc_req_finished(req);
1896
1897                 return put_user(flags, (int *)arg);
1898         }
1899         case FSFILT_IOC_SETFLAGS: {
1900                 struct lov_stripe_md *lsm;
1901                 struct obd_info oinfo = { { { 0 } } };
1902                 struct md_op_data *op_data;
1903
1904                 if (get_user(flags, (int *)arg))
1905                         return -EFAULT;
1906
1907                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1908                                              LUSTRE_OPC_ANY, NULL);
1909                 if (IS_ERR(op_data))
1910                         return PTR_ERR(op_data);
1911
1912                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1913                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1914                 rc = md_setattr(sbi->ll_md_exp, op_data,
1915                                 NULL, 0, NULL, 0, &req, NULL);
1916                 ll_finish_md_op_data(op_data);
1917                 ptlrpc_req_finished(req);
1918                 if (rc)
1919                         return rc;
1920
1921                 inode->i_flags = ll_ext_to_inode_flags(flags);
1922
1923                 lsm = ccc_inode_lsm_get(inode);
1924                 if (!lsm_has_objects(lsm)) {
1925                         ccc_inode_lsm_put(inode, lsm);
1926                         return 0;
1927                 }
1928
1929                 OBDO_ALLOC(oinfo.oi_oa);
1930                 if (!oinfo.oi_oa) {
1931                         ccc_inode_lsm_put(inode, lsm);
1932                         return -ENOMEM;
1933                 }
1934                 oinfo.oi_md = lsm;
1935                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
1936                 oinfo.oi_oa->o_flags = flags;
1937                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1938                                        OBD_MD_FLGROUP;
1939                 oinfo.oi_capa = ll_mdscapa_get(inode);
1940                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
1941                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
1942                 capa_put(oinfo.oi_capa);
1943                 OBDO_FREE(oinfo.oi_oa);
1944                 ccc_inode_lsm_put(inode, lsm);
1945
1946                 if (rc && rc != -EPERM && rc != -EACCES)
1947                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
1948
1949                 return rc;
1950         }
1951         default:
1952                 return -ENOSYS;
1953         }
1954
1955         return 0;
1956 }
1957
1958 int ll_flush_ctx(struct inode *inode)
1959 {
1960         struct ll_sb_info  *sbi = ll_i2sbi(inode);
1961
1962         CDEBUG(D_SEC, "flush context for user %d\n",
1963                       from_kuid(&init_user_ns, current_uid()));
1964
1965         obd_set_info_async(NULL, sbi->ll_md_exp,
1966                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1967                            0, NULL, NULL);
1968         obd_set_info_async(NULL, sbi->ll_dt_exp,
1969                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
1970                            0, NULL, NULL);
1971         return 0;
1972 }
1973
1974 /* umount -f client means force down, don't save state */
1975 void ll_umount_begin(struct super_block *sb)
1976 {
1977         struct ll_sb_info *sbi = ll_s2sbi(sb);
1978         struct obd_device *obd;
1979         struct obd_ioctl_data *ioc_data;
1980
1981         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
1982                sb->s_count, atomic_read(&sb->s_active));
1983
1984         obd = class_exp2obd(sbi->ll_md_exp);
1985         if (obd == NULL) {
1986                 CERROR("Invalid MDC connection handle %#llx\n",
1987                        sbi->ll_md_exp->exp_handle.h_cookie);
1988                 return;
1989         }
1990         obd->obd_force = 1;
1991
1992         obd = class_exp2obd(sbi->ll_dt_exp);
1993         if (obd == NULL) {
1994                 CERROR("Invalid LOV connection handle %#llx\n",
1995                        sbi->ll_dt_exp->exp_handle.h_cookie);
1996                 return;
1997         }
1998         obd->obd_force = 1;
1999
2000         ioc_data = kzalloc(sizeof(*ioc_data), GFP_NOFS);
2001         if (ioc_data) {
2002                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2003                               sizeof(*ioc_data), ioc_data, NULL);
2004
2005                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2006                               sizeof(*ioc_data), ioc_data, NULL);
2007
2008                 OBD_FREE_PTR(ioc_data);
2009         }
2010
2011         /* Really, we'd like to wait until there are no requests outstanding,
2012          * and then continue.  For now, we just invalidate the requests,
2013          * schedule() and sleep one second if needed, and hope.
2014          */
2015         schedule();
2016 }
2017
2018 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2019 {
2020         struct ll_sb_info *sbi = ll_s2sbi(sb);
2021         char *profilenm = get_profile_name(sb);
2022         int err;
2023         __u32 read_only;
2024
2025         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2026                 read_only = *flags & MS_RDONLY;
2027                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2028                                          sizeof(KEY_READ_ONLY),
2029                                          KEY_READ_ONLY, sizeof(read_only),
2030                                          &read_only, NULL);
2031                 if (err) {
2032                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2033                                       profilenm, read_only ?
2034                                       "read-only" : "read-write", err);
2035                         return err;
2036                 }
2037
2038                 if (read_only)
2039                         sb->s_flags |= MS_RDONLY;
2040                 else
2041                         sb->s_flags &= ~MS_RDONLY;
2042
2043                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2044                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2045                                       read_only ?  "read-only" : "read-write");
2046         }
2047         return 0;
2048 }
2049
2050 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2051                   struct super_block *sb, struct lookup_intent *it)
2052 {
2053         struct ll_sb_info *sbi = NULL;
2054         struct lustre_md md;
2055         int rc;
2056
2057         LASSERT(*inode || sb);
2058         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2059         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2060                               sbi->ll_md_exp, &md);
2061         if (rc)
2062                 return rc;
2063
2064         if (*inode) {
2065                 ll_update_inode(*inode, &md);
2066         } else {
2067                 LASSERT(sb != NULL);
2068
2069                 /*
2070                  * At this point server returns to client's same fid as client
2071                  * generated for creating. So using ->fid1 is okay here.
2072                  */
2073                 LASSERT(fid_is_sane(&md.body->fid1));
2074
2075                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2076                                              sbi->ll_flags & LL_SBI_32BIT_API),
2077                                  &md);
2078                 if (*inode == NULL || IS_ERR(*inode)) {
2079 #ifdef CONFIG_FS_POSIX_ACL
2080                         if (md.posix_acl) {
2081                                 posix_acl_release(md.posix_acl);
2082                                 md.posix_acl = NULL;
2083                         }
2084 #endif
2085                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2086                         *inode = NULL;
2087                         CERROR("new_inode -fatal: rc %d\n", rc);
2088                         goto out;
2089                 }
2090         }
2091
2092         /* Handling piggyback layout lock.
2093          * Layout lock can be piggybacked by getattr and open request.
2094          * The lsm can be applied to inode only if it comes with a layout lock
2095          * otherwise correct layout may be overwritten, for example:
2096          * 1. proc1: mdt returns a lsm but not granting layout
2097          * 2. layout was changed by another client
2098          * 3. proc2: refresh layout and layout lock granted
2099          * 4. proc1: to apply a stale layout */
2100         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2101                 struct lustre_handle lockh;
2102                 struct ldlm_lock *lock;
2103
2104                 lockh.cookie = it->d.lustre.it_lock_handle;
2105                 lock = ldlm_handle2lock(&lockh);
2106                 LASSERT(lock != NULL);
2107                 if (ldlm_has_layout(lock)) {
2108                         struct cl_object_conf conf;
2109
2110                         memset(&conf, 0, sizeof(conf));
2111                         conf.coc_opc = OBJECT_CONF_SET;
2112                         conf.coc_inode = *inode;
2113                         conf.coc_lock = lock;
2114                         conf.u.coc_md = &md;
2115                         (void)ll_layout_conf(*inode, &conf);
2116                 }
2117                 LDLM_LOCK_PUT(lock);
2118         }
2119
2120 out:
2121         if (md.lsm != NULL)
2122                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2123         md_free_lustre_md(sbi->ll_md_exp, &md);
2124         return rc;
2125 }
2126
2127 int ll_obd_statfs(struct inode *inode, void *arg)
2128 {
2129         struct ll_sb_info *sbi = NULL;
2130         struct obd_export *exp;
2131         char *buf = NULL;
2132         struct obd_ioctl_data *data = NULL;
2133         __u32 type;
2134         __u32 flags;
2135         int len = 0, rc;
2136
2137         if (!inode) {
2138                 rc = -EINVAL;
2139                 goto out_statfs;
2140         }
2141
2142         sbi = ll_i2sbi(inode);
2143         if (!sbi) {
2144                 rc = -EINVAL;
2145                 goto out_statfs;
2146         }
2147
2148         rc = obd_ioctl_getdata(&buf, &len, arg);
2149         if (rc)
2150                 goto out_statfs;
2151
2152         data = (void *)buf;
2153         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2154             !data->ioc_pbuf1 || !data->ioc_pbuf2) {
2155                 rc = -EINVAL;
2156                 goto out_statfs;
2157         }
2158
2159         if (data->ioc_inllen1 != sizeof(__u32) ||
2160             data->ioc_inllen2 != sizeof(__u32) ||
2161             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2162             data->ioc_plen2 != sizeof(struct obd_uuid)) {
2163                 rc = -EINVAL;
2164                 goto out_statfs;
2165         }
2166
2167         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2168         if (type & LL_STATFS_LMV)
2169                 exp = sbi->ll_md_exp;
2170         else if (type & LL_STATFS_LOV)
2171                 exp = sbi->ll_dt_exp;
2172         else {
2173                 rc = -ENODEV;
2174                 goto out_statfs;
2175         }
2176
2177         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2178         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2179         if (rc)
2180                 goto out_statfs;
2181 out_statfs:
2182         if (buf)
2183                 obd_ioctl_freedata(buf, len);
2184         return rc;
2185 }
2186
2187 int ll_process_config(struct lustre_cfg *lcfg)
2188 {
2189         char *ptr;
2190         void *sb;
2191         struct lprocfs_static_vars lvars;
2192         unsigned long x;
2193         int rc = 0;
2194
2195         lprocfs_llite_init_vars(&lvars);
2196
2197         /* The instance name contains the sb: lustre-client-aacfe000 */
2198         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2199         if (!ptr || !*(++ptr))
2200                 return -EINVAL;
2201         if (sscanf(ptr, "%lx", &x) != 1)
2202                 return -EINVAL;
2203         sb = (void *)x;
2204         /* This better be a real Lustre superblock! */
2205         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2206
2207         /* Note we have not called client_common_fill_super yet, so
2208            proc fns must be able to handle that! */
2209         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2210                                       lcfg, sb);
2211         if (rc > 0)
2212                 rc = 0;
2213         return rc;
2214 }
2215
2216 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2217 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2218                                        struct inode *i1, struct inode *i2,
2219                                        const char *name, int namelen,
2220                                        int mode, __u32 opc, void *data)
2221 {
2222         LASSERT(i1 != NULL);
2223
2224         if (namelen > ll_i2sbi(i1)->ll_namelen)
2225                 return ERR_PTR(-ENAMETOOLONG);
2226
2227         if (op_data == NULL)
2228                 op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
2229
2230         if (op_data == NULL)
2231                 return ERR_PTR(-ENOMEM);
2232
2233         ll_i2gids(op_data->op_suppgids, i1, i2);
2234         op_data->op_fid1 = *ll_inode2fid(i1);
2235         op_data->op_capa1 = ll_mdscapa_get(i1);
2236
2237         if (i2) {
2238                 op_data->op_fid2 = *ll_inode2fid(i2);
2239                 op_data->op_capa2 = ll_mdscapa_get(i2);
2240         } else {
2241                 fid_zero(&op_data->op_fid2);
2242                 op_data->op_capa2 = NULL;
2243         }
2244
2245         op_data->op_name = name;
2246         op_data->op_namelen = namelen;
2247         op_data->op_mode = mode;
2248         op_data->op_mod_time = get_seconds();
2249         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2250         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2251         op_data->op_cap = cfs_curproc_cap_pack();
2252         op_data->op_bias = 0;
2253         op_data->op_cli_flags = 0;
2254         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2255              filename_is_volatile(name, namelen, NULL))
2256                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2257         op_data->op_opc = opc;
2258         op_data->op_mds = 0;
2259         op_data->op_data = data;
2260
2261         /* If the file is being opened after mknod() (normally due to NFS)
2262          * try to use the default stripe data from parent directory for
2263          * allocating OST objects.  Try to pass the parent FID to MDS. */
2264         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2265             !ll_i2info(i2)->lli_has_smd) {
2266                 struct ll_inode_info *lli = ll_i2info(i2);
2267
2268                 spin_lock(&lli->lli_lock);
2269                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2270                         op_data->op_fid1 = lli->lli_pfid;
2271                 spin_unlock(&lli->lli_lock);
2272                 /** We ignore parent's capability temporary. */
2273         }
2274
2275         /* When called by ll_setattr_raw, file is i1. */
2276         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2277                 op_data->op_bias |= MDS_DATA_MODIFIED;
2278
2279         return op_data;
2280 }
2281
2282 void ll_finish_md_op_data(struct md_op_data *op_data)
2283 {
2284         capa_put(op_data->op_capa1);
2285         capa_put(op_data->op_capa2);
2286         OBD_FREE_PTR(op_data);
2287 }
2288
2289 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2290 {
2291         struct ll_sb_info *sbi;
2292
2293         LASSERT((seq != NULL) && (dentry != NULL));
2294         sbi = ll_s2sbi(dentry->d_sb);
2295
2296         if (sbi->ll_flags & LL_SBI_NOLCK)
2297                 seq_puts(seq, ",nolock");
2298
2299         if (sbi->ll_flags & LL_SBI_FLOCK)
2300                 seq_puts(seq, ",flock");
2301
2302         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2303                 seq_puts(seq, ",localflock");
2304
2305         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2306                 seq_puts(seq, ",user_xattr");
2307
2308         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2309                 seq_puts(seq, ",lazystatfs");
2310
2311         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2312                 seq_puts(seq, ",user_fid2path");
2313
2314         return 0;
2315 }
2316
2317 /**
2318  * Get obd name by cmd, and copy out to user space
2319  */
2320 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2321 {
2322         struct ll_sb_info *sbi = ll_i2sbi(inode);
2323         struct obd_device *obd;
2324
2325         if (cmd == OBD_IOC_GETDTNAME)
2326                 obd = class_exp2obd(sbi->ll_dt_exp);
2327         else if (cmd == OBD_IOC_GETMDNAME)
2328                 obd = class_exp2obd(sbi->ll_md_exp);
2329         else
2330                 return -EINVAL;
2331
2332         if (!obd)
2333                 return -ENOENT;
2334
2335         if (copy_to_user((void *)arg, obd->obd_name,
2336                              strlen(obd->obd_name) + 1))
2337                 return -EFAULT;
2338
2339         return 0;
2340 }
2341
2342 /**
2343  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2344  * fsname will be returned in this buffer; otherwise, a static buffer will be
2345  * used to store the fsname and returned to caller.
2346  */
2347 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2348 {
2349         static char fsname_static[MTI_NAME_MAXLEN];
2350         struct lustre_sb_info *lsi = s2lsi(sb);
2351         char *ptr;
2352         int len;
2353
2354         if (buf == NULL) {
2355                 /* this means the caller wants to use static buffer
2356                  * and it doesn't care about race. Usually this is
2357                  * in error reporting path */
2358                 buf = fsname_static;
2359                 buflen = sizeof(fsname_static);
2360         }
2361
2362         len = strlen(lsi->lsi_lmd->lmd_profile);
2363         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2364         if (ptr && (strcmp(ptr, "-client") == 0))
2365                 len -= 7;
2366
2367         if (unlikely(len >= buflen))
2368                 len = buflen - 1;
2369         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2370         buf[len] = '\0';
2371
2372         return buf;
2373 }
2374
2375 static char *ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2376 {
2377         char *path = NULL;
2378
2379         struct path p;
2380
2381         p.dentry = dentry;
2382         p.mnt = current->fs->root.mnt;
2383         path_get(&p);
2384         path = d_path(&p, buf, bufsize);
2385         path_put(&p);
2386
2387         return path;
2388 }
2389
2390 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2391 {
2392         char *buf, *path = NULL;
2393         struct dentry *dentry = NULL;
2394         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2395
2396         /* this can be called inside spin lock so use GFP_ATOMIC. */
2397         buf = (char *)__get_free_page(GFP_ATOMIC);
2398         if (buf != NULL) {
2399                 dentry = d_find_alias(page->mapping->host);
2400                 if (dentry != NULL)
2401                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2402         }
2403
2404         CDEBUG(D_WARNING,
2405                "%s: dirty page discard: %s/fid: " DFID "/%s may get corrupted (rc %d)\n",
2406                ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2407                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2408                PFID(&obj->cob_header.coh_lu.loh_fid),
2409                (path && !IS_ERR(path)) ? path : "", ioret);
2410
2411         if (dentry != NULL)
2412                 dput(dentry);
2413
2414         if (buf != NULL)
2415                 free_page((unsigned long)buf);
2416 }