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/types.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47
  48 #include <lustre_lite.h>
  49 #include <lustre_ha.h>
  50 #include <lustre_dlm.h>
  51 #include <lprocfs_status.h>
  52 #include <lustre_disk.h>
  53 #include <lustre_param.h>
  54 #include <lustre_log.h>
  55 #include <cl_object.h>
  56 #include <obd_cksum.h>
  57 #include "llite_internal.h"
  58
  59 cfs_mem_cache_t *ll_file_data_slab;
  60
  61 CFS_LIST_HEAD(ll_super_blocks);
  62 DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef MS_HAS_NEW_AOPS
  65 extern struct address_space_operations ll_aops;
  66 #else
  67 extern struct address_space_operations_ext ll_aops;
  68 #endif
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         class_uuid_t uuid;
  81         int i;
  82         ENTRY;
  83
  84         OBD_ALLOC(sbi, sizeof(*sbi));
  85         if (!sbi)
  86                 RETURN(NULL);
  87
  88         spin_lock_init(&sbi->ll_lock);
  89         mutex_init(&sbi->ll_lco.lco_lock);
  90         spin_lock_init(&sbi->ll_pp_extent_lock);
  91         spin_lock_init(&sbi->ll_process_lock);
  92         sbi->ll_rw_stats_on = 0;
  93
  94         si_meminfo(&si);
  95         pages = si.totalram - si.totalhigh;
  96         if (pages >> (20 - CFS_PAGE_SHIFT) < 512) {
  97 #ifdef HAVE_BGL_SUPPORT
  98                 lru_page_max = pages / 4;
  99 #else
 100                 lru_page_max = pages / 2;
 101 #endif
 102         } else {
 103                 lru_page_max = (pages / 4) * 3;
 104         }
 105
 106         /* initialize lru data */
 107         cfs_atomic_set(&sbi->ll_cache.ccc_users, 0);
 108         sbi->ll_cache.ccc_lru_max = lru_page_max;
 109         cfs_atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
 110         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
 111         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
 112
 113         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 114                                            SBI_DEFAULT_READAHEAD_MAX);
 115         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 116         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 117                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 118         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 119         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 120
 121         ll_generate_random_uuid(uuid);
 122         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 123         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 124
 125         spin_lock(&ll_sb_lock);
 126         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 127         spin_unlock(&ll_sb_lock);
 128
 129         sbi->ll_flags |= LL_SBI_VERBOSE;
 130 #ifdef ENABLE_CHECKSUM
 131         sbi->ll_flags |= LL_SBI_CHECKSUM;
 132 #endif
 133
 134 #ifdef HAVE_LRU_RESIZE_SUPPORT
 135         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 136 #endif
 137
 138         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 139                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 140                                pp_r_hist.oh_lock);
 141                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 142                                pp_w_hist.oh_lock);
 143         }
 144
 145         /* metadata statahead is enabled by default */
 146         sbi->ll_sa_max = LL_SA_RPC_DEF;
 147         cfs_atomic_set(&sbi->ll_sa_total, 0);
 148         cfs_atomic_set(&sbi->ll_sa_wrong, 0);
 149         cfs_atomic_set(&sbi->ll_agl_total, 0);
 150         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 151
 152         RETURN(sbi);
 153 }
 154
 155 void ll_free_sbi(struct super_block *sb)
 156 {
 157         struct ll_sb_info *sbi = ll_s2sbi(sb);
 158         ENTRY;
 159
 160         if (sbi != NULL) {
 161                 spin_lock(&ll_sb_lock);
 162                 cfs_list_del(&sbi->ll_list);
 163                 spin_unlock(&ll_sb_lock);
 164                 OBD_FREE(sbi, sizeof(*sbi));
 165         }
 166         EXIT;
 167 }
 168
 169 static struct dentry_operations ll_d_root_ops = {
 170         .d_compare = ll_dcompare,
 171         .d_revalidate = ll_revalidate_nd,
 172 };
 173
 174 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 175                                     struct vfsmount *mnt)
 176 {
 177         struct inode *root = 0;
 178         struct ll_sb_info *sbi = ll_s2sbi(sb);
 179         struct obd_device *obd;
 180         struct obd_capa *oc = NULL;
 181         struct obd_statfs *osfs = NULL;
 182         struct ptlrpc_request *request = NULL;
 183         struct obd_connect_data *data = NULL;
 184         struct obd_uuid *uuid;
 185         struct md_op_data *op_data;
 186         struct lustre_md lmd;
 187         obd_valid valid;
 188         int size, err, checksum;
 189         ENTRY;
 190
 191         obd = class_name2obd(md);
 192         if (!obd) {
 193                 CERROR("MD %s: not setup or attached\n", md);
 194                 RETURN(-EINVAL);
 195         }
 196
 197         OBD_ALLOC_PTR(data);
 198         if (data == NULL)
 199                 RETURN(-ENOMEM);
 200
 201         OBD_ALLOC_PTR(osfs);
 202         if (osfs == NULL) {
 203                 OBD_FREE_PTR(data);
 204                 RETURN(-ENOMEM);
 205         }
 206
 207         if (proc_lustre_fs_root) {
 208                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 209                                                   dt, md);
 210                 if (err < 0)
 211                         CERROR("could not register mount in /proc/fs/lustre\n");
 212         }
 213
 214         /* indicate the features supported by this client */
 215         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 216                                   OBD_CONNECT_ATTRFID  |
 217                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 218                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 219                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 220                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 221                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 222                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 223                                   OBD_CONNECT_EINPROGRESS |
 224                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE;
 225
 226         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 227                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 228
 229 #ifdef HAVE_LRU_RESIZE_SUPPORT
 230         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 231                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 232 #endif
 233 #ifdef CONFIG_FS_POSIX_ACL
 234         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 235 #endif
 236
 237         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 238                 /* flag mdc connection as lightweight, only used for test
 239                  * purpose, use with care */
 240                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 241
 242         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 243         data->ocd_version = LUSTRE_VERSION_CODE;
 244
 245         if (sb->s_flags & MS_RDONLY)
 246                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 247         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 248                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 249
 250 #ifdef HAVE_MS_FLOCK_LOCK
 251         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 252         sb->s_flags |= MS_FLOCK_LOCK;
 253 #endif
 254 #ifdef MS_HAS_NEW_AOPS
 255         sb->s_flags |= MS_HAS_NEW_AOPS;
 256 #endif
 257
 258         if (sbi->ll_flags & LL_SBI_FLOCK)
 259                 sbi->ll_fop = &ll_file_operations_flock;
 260         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 261                 sbi->ll_fop = &ll_file_operations;
 262         else
 263                 sbi->ll_fop = &ll_file_operations_noflock;
 264
 265         /* real client */
 266         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 267         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 268                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 269
 270         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 271
 272         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 273         if (err == -EBUSY) {
 274                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 275                                    "recovery, of which this client is not a "
 276                                    "part. Please wait for recovery to complete,"
 277                                    " abort, or time out.\n", md);
 278                 GOTO(out, err);
 279         } else if (err) {
 280                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 281                 GOTO(out, err);
 282         }
 283
 284         err = obd_fid_init(sbi->ll_md_exp);
 285         if (err) {
 286                 CERROR("Can't init metadata layer FID infrastructure, "
 287                        "rc %d\n", err);
 288                 GOTO(out_md, err);
 289         }
 290
 291         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 292                          cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS), 0);
 293         if (err)
 294                 GOTO(out_md_fid, err);
 295
 296         /* This needs to be after statfs to ensure connect has finished.
 297          * Note that "data" does NOT contain the valid connect reply.
 298          * If connecting to a 1.8 server there will be no LMV device, so
 299          * we can access the MDC export directly and exp_connect_flags will
 300          * be non-zero, but if accessing an upgraded 2.1 server it will
 301          * have the correct flags filled in.
 302          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 303         valid = sbi->ll_md_exp->exp_connect_flags & CLIENT_CONNECT_MDT_REQD;
 304         if (sbi->ll_md_exp->exp_connect_flags != 0 &&
 305             valid != CLIENT_CONNECT_MDT_REQD) {
 306                 char *buf;
 307
 308                 OBD_ALLOC_WAIT(buf, CFS_PAGE_SIZE);
 309                 obd_connect_flags2str(buf, CFS_PAGE_SIZE,
 310                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 311                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 312                                    "feature(s) needed for correct operation "
 313                                    "of this client (%s). Please upgrade "
 314                                    "server or downgrade client.\n",
 315                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 316                 OBD_FREE(buf, CFS_PAGE_SIZE);
 317                 GOTO(out_md, err = -EPROTO);
 318         }
 319
 320         size = sizeof(*data);
 321         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 322                            KEY_CONN_DATA,  &size, data, NULL);
 323         if (err) {
 324                 CERROR("Get connect data failed: %d \n", err);
 325                 GOTO(out_md, err);
 326         }
 327
 328         LASSERT(osfs->os_bsize);
 329         sb->s_blocksize = osfs->os_bsize;
 330         sb->s_blocksize_bits = log2(osfs->os_bsize);
 331         sb->s_magic = LL_SUPER_MAGIC;
 332         sb->s_maxbytes = MAX_LFS_FILESIZE;
 333         sbi->ll_namelen = osfs->os_namelen;
 334         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 335
 336         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 337             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 338                 LCONSOLE_INFO("Disabling user_xattr feature because "
 339                               "it is not supported on the server\n");
 340                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 341         }
 342
 343         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 344 #ifdef MS_POSIXACL
 345                 sb->s_flags |= MS_POSIXACL;
 346 #endif
 347                 sbi->ll_flags |= LL_SBI_ACL;
 348         } else {
 349                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 350 #ifdef MS_POSIXACL
 351                 sb->s_flags &= ~MS_POSIXACL;
 352 #endif
 353                 sbi->ll_flags &= ~LL_SBI_ACL;
 354         }
 355
 356         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 357                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 358                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 359                         LCONSOLE_INFO("client is set as remote by default.\n");
 360                 }
 361         } else {
 362                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 363                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 364                         LCONSOLE_INFO("client claims to be remote, but server "
 365                                       "rejected, forced to be local.\n");
 366                 }
 367         }
 368
 369         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 370                 LCONSOLE_INFO("client enabled MDS capability!\n");
 371                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 372         }
 373
 374         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 375                 LCONSOLE_INFO("client enabled OSS capability!\n");
 376                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 377         }
 378
 379         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 380                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 381
 382         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 383                 sbi->ll_md_brw_size = data->ocd_brw_size;
 384         else
 385                 sbi->ll_md_brw_size = CFS_PAGE_SIZE;
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK) {
 388                 LCONSOLE_INFO("Layout lock feature supported.\n");
 389                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 390         }
 391
 392         obd = class_name2obd(dt);
 393         if (!obd) {
 394                 CERROR("DT %s: not setup or attached\n", dt);
 395                 GOTO(out_md_fid, err = -ENODEV);
 396         }
 397
 398         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 399                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 400                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 401                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 402                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 403                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 404                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 405                                   OBD_CONNECT_MAXBYTES |
 406                                   OBD_CONNECT_EINPROGRESS |
 407                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE;
 408
 409         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 410                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 411
 412         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 413                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 414                  * disabled by default, because it can still be enabled on the
 415                  * fly via /proc. As a consequence, we still need to come to an
 416                  * agreement on the supported algorithms at connect time */
 417                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 418
 419                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 420                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 421                 else
 422                         data->ocd_cksum_types = cksum_types_supported_client();
 423         }
 424
 425 #ifdef HAVE_LRU_RESIZE_SUPPORT
 426         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 427 #endif
 428         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 429                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 430
 431         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 432                "ocd_grant: %d\n", data->ocd_connect_flags,
 433                data->ocd_version, data->ocd_grant);
 434
 435         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 436         obd->obd_upcall.onu_upcall = cl_ocd_update;
 437
 438         data->ocd_brw_size = PTLRPC_MAX_BRW_SIZE;
 439
 440         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 441         if (err == -EBUSY) {
 442                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 443                                    "recovery, of which this client is not a "
 444                                    "part.  Please wait for recovery to "
 445                                    "complete, abort, or time out.\n", dt);
 446                 GOTO(out_md_fid, err);
 447         } else if (err) {
 448                 CERROR("Cannot connect to %s: rc = %d\n", dt, err);
 449                 GOTO(out_md_fid, err);
 450         }
 451
 452         err = obd_fid_init(sbi->ll_dt_exp);
 453         if (err) {
 454                 CERROR("Can't init data layer FID infrastructure, "
 455                        "rc %d\n", err);
 456                 GOTO(out_dt, err);
 457         }
 458
 459         mutex_lock(&sbi->ll_lco.lco_lock);
 460         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 461         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 462         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 463         mutex_unlock(&sbi->ll_lco.lco_lock);
 464
 465         fid_zero(&sbi->ll_root_fid);
 466         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 467         if (err) {
 468                 CERROR("cannot mds_connect: rc = %d\n", err);
 469                 GOTO(out_lock_cn_cb, err);
 470         }
 471         if (!fid_is_sane(&sbi->ll_root_fid)) {
 472                 CERROR("Invalid root fid during mount\n");
 473                 GOTO(out_lock_cn_cb, err = -EINVAL);
 474         }
 475         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 476
 477         sb->s_op = &lustre_super_operations;
 478 #if THREAD_SIZE >= 8192 /*b=17630*/
 479         sb->s_export_op = &lustre_export_operations;
 480 #endif
 481
 482         /* make root inode
 483          * XXX: move this to after cbd setup? */
 484         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA;
 485         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 486                 valid |= OBD_MD_FLRMTPERM;
 487         else if (sbi->ll_flags & LL_SBI_ACL)
 488                 valid |= OBD_MD_FLACL;
 489
 490         OBD_ALLOC_PTR(op_data);
 491         if (op_data == NULL)
 492                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 493
 494         op_data->op_fid1 = sbi->ll_root_fid;
 495         op_data->op_mode = 0;
 496         op_data->op_capa1 = oc;
 497         op_data->op_valid = valid;
 498
 499         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 500         if (oc)
 501                 capa_put(oc);
 502         OBD_FREE_PTR(op_data);
 503         if (err) {
 504                 CERROR("md_getattr failed for root: rc = %d\n", err);
 505                 GOTO(out_lock_cn_cb, err);
 506         }
 507         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 508                                sbi->ll_md_exp, &lmd);
 509         if (err) {
 510                 CERROR("failed to understand root inode md: rc = %d\n", err);
 511                 ptlrpc_req_finished (request);
 512                 GOTO(out_lock_cn_cb, err);
 513         }
 514
 515         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 516         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid, 0), &lmd);
 517         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 518         ptlrpc_req_finished(request);
 519
 520         if (root == NULL || IS_ERR(root)) {
 521                 if (lmd.lsm)
 522                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 523 #ifdef CONFIG_FS_POSIX_ACL
 524                 if (lmd.posix_acl) {
 525                         posix_acl_release(lmd.posix_acl);
 526                         lmd.posix_acl = NULL;
 527                 }
 528 #endif
 529                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 530                 root = NULL;
 531                 CERROR("lustre_lite: bad iget4 for root\n");
 532                 GOTO(out_root, err);
 533         }
 534
 535         err = ll_close_thread_start(&sbi->ll_lcq);
 536         if (err) {
 537                 CERROR("cannot start close thread: rc %d\n", err);
 538                 GOTO(out_root, err);
 539         }
 540
 541 #ifdef CONFIG_FS_POSIX_ACL
 542         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 543                 rct_init(&sbi->ll_rct);
 544                 et_init(&sbi->ll_et);
 545         }
 546 #endif
 547
 548         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 549         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 550                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 551                                  NULL);
 552         cl_sb_init(sb);
 553
 554         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 555                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 556                                  &sbi->ll_cache, NULL);
 557
 558         sb->s_root = d_make_root(root);
 559         if (sb->s_root == NULL) {
 560                 CERROR("%s: can't make root dentry\n",
 561                         ll_get_fsname(sb, NULL, 0));
 562                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 563         }
 564
 565 #ifdef HAVE_DCACHE_LOCK
 566         sb->s_root->d_op = &ll_d_root_ops;
 567 #else
 568         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
 569         d_set_d_op(sb->s_root, &ll_d_root_ops);
 570         sb->s_d_op = &ll_d_ops;
 571 #endif
 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
 584         if (data != NULL)
 585                 OBD_FREE_PTR(data);
 586         if (osfs != NULL)
 587                 OBD_FREE_PTR(osfs);
 588
 589         RETURN(err);
 590 out_root:
 591         if (root)
 592                 iput(root);
 593 out_lock_cn_cb:
 594         obd_fid_fini(sbi->ll_dt_exp);
 595 out_dt:
 596         obd_disconnect(sbi->ll_dt_exp);
 597         sbi->ll_dt_exp = NULL;
 598         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 599         obd_zombie_barrier();
 600 out_md_fid:
 601         obd_fid_fini(sbi->ll_md_exp);
 602 out_md:
 603         obd_disconnect(sbi->ll_md_exp);
 604         sbi->ll_md_exp = NULL;
 605 out:
 606         if (data != NULL)
 607                 OBD_FREE_PTR(data);
 608         if (osfs != NULL)
 609                 OBD_FREE_PTR(osfs);
 610         lprocfs_unregister_mountpoint(sbi);
 611         return err;
 612 }
 613
 614 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 615 {
 616         int size, rc;
 617
 618         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 619         size = sizeof(int);
 620         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 621                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 622         if (rc)
 623                 CERROR("Get max mdsize error rc %d \n", rc);
 624
 625         RETURN(rc);
 626 }
 627
 628 void ll_dump_inode(struct inode *inode)
 629 {
 630         struct ll_d_hlist_node *tmp;
 631         int dentry_count = 0;
 632
 633         LASSERT(inode != NULL);
 634
 635         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 636                 dentry_count++;
 637
 638         CERROR("inode %p dump: dev=%s ino=%lu mode=%o count=%u, %d dentries\n",
 639                inode, ll_i2mdexp(inode)->exp_obd->obd_name, inode->i_ino,
 640                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 641 }
 642
 643 void lustre_dump_dentry(struct dentry *dentry, int recur)
 644 {
 645         struct list_head *tmp;
 646         int subdirs = 0;
 647
 648         LASSERT(dentry != NULL);
 649
 650         list_for_each(tmp, &dentry->d_subdirs)
 651                 subdirs++;
 652
 653         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 654                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 655                dentry->d_name.len, dentry->d_name.name,
 656                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 657                dentry->d_parent, dentry->d_inode, d_refcount(dentry),
 658                dentry->d_flags, dentry->d_fsdata, subdirs);
 659         if (dentry->d_inode != NULL)
 660                 ll_dump_inode(dentry->d_inode);
 661
 662         if (recur == 0)
 663                 return;
 664
 665         list_for_each(tmp, &dentry->d_subdirs) {
 666                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 667                 lustre_dump_dentry(d, recur - 1);
 668         }
 669 }
 670
 671 void client_common_put_super(struct super_block *sb)
 672 {
 673         struct ll_sb_info *sbi = ll_s2sbi(sb);
 674         ENTRY;
 675
 676 #ifdef CONFIG_FS_POSIX_ACL
 677         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 678                 et_fini(&sbi->ll_et);
 679                 rct_fini(&sbi->ll_rct);
 680         }
 681 #endif
 682
 683         ll_close_thread_shutdown(sbi->ll_lcq);
 684
 685         cl_sb_fini(sb);
 686
 687         cfs_list_del(&sbi->ll_conn_chain);
 688
 689         obd_fid_fini(sbi->ll_dt_exp);
 690         obd_disconnect(sbi->ll_dt_exp);
 691         sbi->ll_dt_exp = NULL;
 692         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 693          * see LU-2543. */
 694         obd_zombie_barrier();
 695
 696         lprocfs_unregister_mountpoint(sbi);
 697
 698         obd_fid_fini(sbi->ll_md_exp);
 699         obd_disconnect(sbi->ll_md_exp);
 700         sbi->ll_md_exp = NULL;
 701
 702         EXIT;
 703 }
 704
 705 void ll_kill_super(struct super_block *sb)
 706 {
 707         struct ll_sb_info *sbi;
 708
 709         ENTRY;
 710
 711         /* not init sb ?*/
 712         if (!(sb->s_flags & MS_ACTIVE))
 713                 return;
 714
 715         sbi = ll_s2sbi(sb);
 716         /* we need restore s_dev from changed for clustred NFS before put_super
 717          * because new kernels have cached s_dev and change sb->s_dev in
 718          * put_super not affected real removing devices */
 719         if (sbi)
 720                 sb->s_dev = sbi->ll_sdev_orig;
 721         EXIT;
 722 }
 723
 724 char *ll_read_opt(const char *opt, char *data)
 725 {
 726         char *value;
 727         char *retval;
 728         ENTRY;
 729
 730         CDEBUG(D_SUPER, "option: %s, data %s\n", opt, data);
 731         if (strncmp(opt, data, strlen(opt)))
 732                 RETURN(NULL);
 733         if ((value = strchr(data, '=')) == NULL)
 734                 RETURN(NULL);
 735
 736         value++;
 737         OBD_ALLOC(retval, strlen(value) + 1);
 738         if (!retval) {
 739                 CERROR("out of memory!\n");
 740                 RETURN(NULL);
 741         }
 742
 743         memcpy(retval, value, strlen(value)+1);
 744         CDEBUG(D_SUPER, "Assigned option: %s, value %s\n", opt, retval);
 745         RETURN(retval);
 746 }
 747
 748 static inline int ll_set_opt(const char *opt, char *data, int fl)
 749 {
 750         if (strncmp(opt, data, strlen(opt)) != 0)
 751                 return(0);
 752         else
 753                 return(fl);
 754 }
 755
 756 /* non-client-specific mount options are parsed in lmd_parse */
 757 static int ll_options(char *options, int *flags)
 758 {
 759         int tmp;
 760         char *s1 = options, *s2;
 761         ENTRY;
 762
 763         if (!options)
 764                 RETURN(0);
 765
 766         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 767
 768         while (*s1) {
 769                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 770                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 771                 if (tmp) {
 772                         *flags |= tmp;
 773                         goto next;
 774                 }
 775                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 776                 if (tmp) {
 777                         *flags |= tmp;
 778                         goto next;
 779                 }
 780                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 781                 if (tmp) {
 782                         *flags |= tmp;
 783                         goto next;
 784                 }
 785                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 786                 if (tmp) {
 787                         *flags &= ~tmp;
 788                         goto next;
 789                 }
 790                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 791                 if (tmp) {
 792                         *flags |= tmp;
 793                         goto next;
 794                 }
 795                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 796                 if (tmp) {
 797                         *flags &= ~tmp;
 798                         goto next;
 799                 }
 800 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 5, 50, 0)
 801                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 802                 if (tmp) {
 803                         /* Ignore deprecated mount option.  The client will
 804                          * always try to mount with ACL support, whether this
 805                          * is used depends on whether server supports it. */
 806                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 807                                                   "mount option 'acl'.\n");
 808                         goto next;
 809                 }
 810                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 811                 if (tmp) {
 812                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 813                                                   "mount option 'noacl'.\n");
 814                         goto next;
 815                 }
 816 #else
 817 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 818 #endif
 819                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 820                 if (tmp) {
 821                         *flags |= tmp;
 822                         goto next;
 823                 }
 824                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 825                 if (tmp) {
 826                         *flags |= tmp;
 827                         goto next;
 828                 }
 829                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 830                 if (tmp) {
 831                         *flags &= ~tmp;
 832                         goto next;
 833                 }
 834
 835                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 836                 if (tmp) {
 837                         *flags |= tmp;
 838                         goto next;
 839                 }
 840                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 841                 if (tmp) {
 842                         *flags &= ~tmp;
 843                         goto next;
 844                 }
 845                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 846                 if (tmp) {
 847                         *flags |= tmp;
 848                         goto next;
 849                 }
 850                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 851                 if (tmp) {
 852                         *flags &= ~tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 856                 if (tmp) {
 857                         *flags |= tmp;
 858                         goto next;
 859                 }
 860                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 861                 if (tmp) {
 862                         *flags &= ~tmp;
 863                         goto next;
 864                 }
 865                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 866                 if (tmp) {
 867                         *flags |= tmp;
 868                         goto next;
 869                 }
 870                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 871                 if (tmp) {
 872                         *flags |= tmp;
 873                         goto next;
 874                 }
 875                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 876                 if (tmp) {
 877                         *flags |= tmp;
 878                         goto next;
 879                 }
 880                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 881                 if (tmp) {
 882                         *flags &= ~tmp;
 883                         goto next;
 884                 }
 885                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 886                                    s1);
 887                 RETURN(-EINVAL);
 888
 889 next:
 890                 /* Find next opt */
 891                 s2 = strchr(s1, ',');
 892                 if (s2 == NULL)
 893                         break;
 894                 s1 = s2 + 1;
 895         }
 896         RETURN(0);
 897 }
 898
 899 void ll_lli_init(struct ll_inode_info *lli)
 900 {
 901         lli->lli_inode_magic = LLI_INODE_MAGIC;
 902         lli->lli_flags = 0;
 903         lli->lli_ioepoch = 0;
 904         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 905         spin_lock_init(&lli->lli_lock);
 906         lli->lli_posix_acl = NULL;
 907         lli->lli_remote_perms = NULL;
 908         mutex_init(&lli->lli_rmtperm_mutex);
 909         /* Do not set lli_fid, it has been initialized already. */
 910         fid_zero(&lli->lli_pfid);
 911         CFS_INIT_LIST_HEAD(&lli->lli_close_list);
 912         CFS_INIT_LIST_HEAD(&lli->lli_oss_capas);
 913         cfs_atomic_set(&lli->lli_open_count, 0);
 914         lli->lli_mds_capa = NULL;
 915         lli->lli_rmtperm_time = 0;
 916         lli->lli_pending_och = NULL;
 917         lli->lli_mds_read_och = NULL;
 918         lli->lli_mds_write_och = NULL;
 919         lli->lli_mds_exec_och = NULL;
 920         lli->lli_open_fd_read_count = 0;
 921         lli->lli_open_fd_write_count = 0;
 922         lli->lli_open_fd_exec_count = 0;
 923         mutex_init(&lli->lli_och_mutex);
 924         spin_lock_init(&lli->lli_agl_lock);
 925         lli->lli_has_smd = false;
 926         lli->lli_clob = NULL;
 927
 928         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 929         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 930                 mutex_init(&lli->lli_readdir_mutex);
 931                 lli->lli_opendir_key = NULL;
 932                 lli->lli_sai = NULL;
 933                 lli->lli_def_acl = NULL;
 934                 spin_lock_init(&lli->lli_sa_lock);
 935                 lli->lli_opendir_pid = 0;
 936         } else {
 937                 sema_init(&lli->lli_size_sem, 1);
 938                 lli->lli_size_sem_owner = NULL;
 939                 lli->lli_symlink_name = NULL;
 940                 init_rwsem(&lli->lli_trunc_sem);
 941                 mutex_init(&lli->lli_write_mutex);
 942                 init_rwsem(&lli->lli_glimpse_sem);
 943                 lli->lli_glimpse_time = 0;
 944                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 945                 lli->lli_agl_index = 0;
 946                 lli->lli_async_rc = 0;
 947         }
 948         mutex_init(&lli->lli_layout_mutex);
 949 }
 950
 951 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 952 {
 953 #ifdef HAVE_BDI_REGISTER
 954         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 955
 956 #ifdef HAVE_BDI_NAME
 957         bdi->name = "lustre";
 958 #endif
 959         return bdi_register(bdi, NULL, "lustre-%d",
 960                             atomic_inc_return(&ll_bdi_num));
 961 #else
 962         return 0;
 963 #endif
 964 }
 965
 966 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 967 {
 968         struct lustre_profile *lprof = NULL;
 969         struct lustre_sb_info *lsi = s2lsi(sb);
 970         struct ll_sb_info *sbi;
 971         char  *dt = NULL, *md = NULL;
 972         char  *profilenm = get_profile_name(sb);
 973         struct config_llog_instance *cfg;
 974         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
 975         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
 976         int    err;
 977         ENTRY;
 978
 979         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
 980
 981         OBD_ALLOC_PTR(cfg);
 982         if (cfg == NULL)
 983                 RETURN(-ENOMEM);
 984
 985         cfs_module_get();
 986
 987         /* client additional sb info */
 988         lsi->lsi_llsbi = sbi = ll_init_sbi();
 989         if (!sbi) {
 990                 cfs_module_put(THIS_MODULE);
 991                 OBD_FREE_PTR(cfg);
 992                 RETURN(-ENOMEM);
 993         }
 994
 995         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
 996         if (err)
 997                 GOTO(out_free, err);
 998
 999         err = ll_bdi_init(&lsi->lsi_bdi);
1000         if (err)
1001                 GOTO(out_free, err);
1002         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1003         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1004         err = ll_bdi_register(&lsi->lsi_bdi);
1005         if (err)
1006                 GOTO(out_free, err);
1007
1008 #ifdef HAVE_SB_BDI
1009         sb->s_bdi = &lsi->lsi_bdi;
1010 #endif
1011
1012         /* Generate a string unique to this super, in case some joker tries
1013            to mount the same fs at two mount points.
1014            Use the address of the super itself.*/
1015         cfg->cfg_instance = sb;
1016         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1017         cfg->cfg_callback = class_config_llog_handler;
1018         /* set up client obds */
1019         err = lustre_process_log(sb, profilenm, cfg);
1020         if (err < 0) {
1021                 CERROR("Unable to process log: %d\n", err);
1022                 GOTO(out_free, err);
1023         }
1024
1025         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1026         lprof = class_get_profile(profilenm);
1027         if (lprof == NULL) {
1028                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1029                                    " read from the MGS.  Does that filesystem "
1030                                    "exist?\n", profilenm);
1031                 GOTO(out_free, err = -EINVAL);
1032         }
1033         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1034                lprof->lp_md, lprof->lp_dt);
1035
1036         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1037         if (!dt)
1038                 GOTO(out_free, err = -ENOMEM);
1039         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1040
1041         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1042         if (!md)
1043                 GOTO(out_free, err = -ENOMEM);
1044         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1045
1046         /* connections, registrations, sb setup */
1047         err = client_common_fill_super(sb, md, dt, mnt);
1048
1049 out_free:
1050         if (md)
1051                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1052         if (dt)
1053                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1054         if (err)
1055                 ll_put_super(sb);
1056         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1057                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1058
1059         OBD_FREE_PTR(cfg);
1060         RETURN(err);
1061 } /* ll_fill_super */
1062
1063
1064 void lu_context_keys_dump(void);
1065
1066 void ll_put_super(struct super_block *sb)
1067 {
1068         struct config_llog_instance cfg;
1069         struct obd_device *obd;
1070         struct lustre_sb_info *lsi = s2lsi(sb);
1071         struct ll_sb_info *sbi = ll_s2sbi(sb);
1072         char *profilenm = get_profile_name(sb);
1073         int force = 1, next;
1074         ENTRY;
1075
1076         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1077
1078         ll_print_capa_stat(sbi);
1079
1080         cfg.cfg_instance = sb;
1081         lustre_end_log(sb, profilenm, &cfg);
1082
1083         if (sbi->ll_md_exp) {
1084                 obd = class_exp2obd(sbi->ll_md_exp);
1085                 if (obd)
1086                         force = obd->obd_force;
1087         }
1088
1089         /* We need to set force before the lov_disconnect in
1090            lustre_common_put_super, since l_d cleans up osc's as well. */
1091         if (force) {
1092                 next = 0;
1093                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1094                                                      &next)) != NULL) {
1095                         obd->obd_force = force;
1096                 }
1097         }
1098
1099         if (sbi->ll_lcq) {
1100                 /* Only if client_common_fill_super succeeded */
1101                 client_common_put_super(sb);
1102         }
1103
1104         next = 0;
1105         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1106                 class_manual_cleanup(obd);
1107         }
1108
1109         if (sbi->ll_flags & LL_SBI_VERBOSE)
1110                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1111
1112         if (profilenm)
1113                 class_del_profile(profilenm);
1114
1115         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1116                 ll_bdi_destroy(&lsi->lsi_bdi);
1117                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1118         }
1119
1120         ll_free_sbi(sb);
1121         lsi->lsi_llsbi = NULL;
1122
1123         lustre_common_put_super(sb);
1124
1125         cl_env_cache_purge(~0);
1126
1127         cfs_module_put(THIS_MODULE);
1128
1129         EXIT;
1130 } /* client_put_super */
1131
1132 struct inode *ll_inode_from_lock(struct ldlm_lock *lock)
1133 {
1134         struct inode *inode = NULL;
1135         /* NOTE: we depend on atomic igrab() -bzzz */
1136         lock_res_and_lock(lock);
1137         if (lock->l_ast_data) {
1138                 struct ll_inode_info *lli = ll_i2info(lock->l_ast_data);
1139                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1140                         inode = igrab(lock->l_ast_data);
1141                 } else {
1142                         inode = lock->l_ast_data;
1143                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1144                                          D_WARNING, lock, "l_ast_data %p is "
1145                                          "bogus: magic %08x", lock->l_ast_data,
1146                                          lli->lli_inode_magic);
1147                         inode = NULL;
1148                 }
1149         }
1150         unlock_res_and_lock(lock);
1151         return inode;
1152 }
1153
1154 static int null_if_equal(struct ldlm_lock *lock, void *data)
1155 {
1156         if (data == lock->l_ast_data) {
1157                 lock->l_ast_data = NULL;
1158
1159                 if (lock->l_req_mode != lock->l_granted_mode)
1160                         LDLM_ERROR(lock,"clearing inode with ungranted lock");
1161         }
1162
1163         return LDLM_ITER_CONTINUE;
1164 }
1165
1166 void ll_clear_inode(struct inode *inode)
1167 {
1168         struct ll_inode_info *lli = ll_i2info(inode);
1169         struct ll_sb_info *sbi = ll_i2sbi(inode);
1170         ENTRY;
1171
1172         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
1173                inode->i_generation, inode);
1174
1175         if (S_ISDIR(inode->i_mode)) {
1176                 /* these should have been cleared in ll_file_release */
1177                 LASSERT(lli->lli_opendir_key == NULL);
1178                 LASSERT(lli->lli_sai == NULL);
1179                 LASSERT(lli->lli_opendir_pid == 0);
1180         }
1181
1182         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1183         md_change_cbdata(sbi->ll_md_exp, ll_inode2fid(inode),
1184                          null_if_equal, inode);
1185
1186         LASSERT(!lli->lli_open_fd_write_count);
1187         LASSERT(!lli->lli_open_fd_read_count);
1188         LASSERT(!lli->lli_open_fd_exec_count);
1189
1190         if (lli->lli_mds_write_och)
1191                 ll_md_real_close(inode, FMODE_WRITE);
1192         if (lli->lli_mds_exec_och)
1193                 ll_md_real_close(inode, FMODE_EXEC);
1194         if (lli->lli_mds_read_och)
1195                 ll_md_real_close(inode, FMODE_READ);
1196
1197         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1198                 OBD_FREE(lli->lli_symlink_name,
1199                          strlen(lli->lli_symlink_name) + 1);
1200                 lli->lli_symlink_name = NULL;
1201         }
1202
1203         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1204                 LASSERT(lli->lli_posix_acl == NULL);
1205                 if (lli->lli_remote_perms) {
1206                         free_rmtperm_hash(lli->lli_remote_perms);
1207                         lli->lli_remote_perms = NULL;
1208                 }
1209         }
1210 #ifdef CONFIG_FS_POSIX_ACL
1211         else if (lli->lli_posix_acl) {
1212                 LASSERT(cfs_atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1213                 LASSERT(lli->lli_remote_perms == NULL);
1214                 posix_acl_release(lli->lli_posix_acl);
1215                 lli->lli_posix_acl = NULL;
1216         }
1217 #endif
1218         lli->lli_inode_magic = LLI_INODE_DEAD;
1219
1220         ll_clear_inode_capas(inode);
1221         if (!S_ISDIR(inode->i_mode))
1222                 LASSERT(cfs_list_empty(&lli->lli_agl_list));
1223
1224         /*
1225          * XXX This has to be done before lsm is freed below, because
1226          * cl_object still uses inode lsm.
1227          */
1228         cl_inode_fini(inode);
1229         lli->lli_has_smd = false;
1230
1231         EXIT;
1232 }
1233
1234 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1235                   struct md_open_data **mod)
1236 {
1237         struct lustre_md md;
1238         struct inode *inode = dentry->d_inode;
1239         struct ll_sb_info *sbi = ll_i2sbi(inode);
1240         struct ptlrpc_request *request = NULL;
1241         int rc, ia_valid;
1242         ENTRY;
1243
1244         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1245                                      LUSTRE_OPC_ANY, NULL);
1246         if (IS_ERR(op_data))
1247                 RETURN(PTR_ERR(op_data));
1248
1249         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1250                         &request, mod);
1251         if (rc) {
1252                 ptlrpc_req_finished(request);
1253                 if (rc == -ENOENT) {
1254                         clear_nlink(inode);
1255                         /* Unlinked special device node? Or just a race?
1256                          * Pretend we done everything. */
1257                         if (!S_ISREG(inode->i_mode) &&
1258                             !S_ISDIR(inode->i_mode)) {
1259                                 ia_valid = op_data->op_attr.ia_valid;
1260                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1261                                 rc = simple_setattr(dentry, &op_data->op_attr);
1262                                 op_data->op_attr.ia_valid = ia_valid;
1263                         }
1264                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1265                         CERROR("md_setattr fails: rc = %d\n", rc);
1266                 }
1267                 RETURN(rc);
1268         }
1269
1270         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1271                               sbi->ll_md_exp, &md);
1272         if (rc) {
1273                 ptlrpc_req_finished(request);
1274                 RETURN(rc);
1275         }
1276
1277         /* We want to adjust timestamps.
1278          * If there is at least some data in file, we cleared ATTR_SIZE
1279          * to avoid update size, otherwise it is important to do.(SOM case)
1280          * (bug 6196) */
1281         ia_valid = op_data->op_attr.ia_valid;
1282         /* Since we set ATTR_*_SET flags above, and already done permission
1283          * check, So don't let inode_change_ok() check it again.  */
1284         op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1285         rc = simple_setattr(dentry, &op_data->op_attr);
1286         op_data->op_attr.ia_valid = ia_valid;
1287
1288         /* Extract epoch data if obtained. */
1289         op_data->op_handle = md.body->handle;
1290         op_data->op_ioepoch = md.body->ioepoch;
1291
1292         ll_update_inode(inode, &md);
1293         ptlrpc_req_finished(request);
1294
1295         RETURN(rc);
1296 }
1297
1298 /* Close IO epoch and send Size-on-MDS attribute update. */
1299 static int ll_setattr_done_writing(struct inode *inode,
1300                                    struct md_op_data *op_data,
1301                                    struct md_open_data *mod)
1302 {
1303         struct ll_inode_info *lli = ll_i2info(inode);
1304         int rc = 0;
1305         ENTRY;
1306
1307         LASSERT(op_data != NULL);
1308         if (!S_ISREG(inode->i_mode))
1309                 RETURN(0);
1310
1311         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1312                op_data->op_ioepoch, PFID(&lli->lli_fid));
1313
1314         op_data->op_flags = MF_EPOCH_CLOSE;
1315         ll_done_writing_attr(inode, op_data);
1316         ll_pack_inode2opdata(inode, op_data, NULL);
1317
1318         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1319         if (rc == -EAGAIN) {
1320                 /* MDS has instructed us to obtain Size-on-MDS attribute
1321                  * from OSTs and send setattr to back to MDS. */
1322                 rc = ll_som_update(inode, op_data);
1323         } else if (rc) {
1324                 CERROR("inode %lu mdc truncate failed: rc = %d\n",
1325                        inode->i_ino, rc);
1326         }
1327         RETURN(rc);
1328 }
1329
1330 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1331 {
1332         struct obd_capa *capa;
1333         int rc;
1334
1335         if (attr->ia_valid & ATTR_SIZE)
1336                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1337         else
1338                 capa = ll_mdscapa_get(inode);
1339
1340         rc = cl_setattr_ost(inode, attr, capa);
1341
1342         if (attr->ia_valid & ATTR_SIZE)
1343                 ll_truncate_free_capa(capa);
1344         else
1345                 capa_put(capa);
1346
1347         return rc;
1348 }
1349
1350 #ifndef HAVE_VFS_INODE_NEWSIZE_OK
1351 /**
1352  * inode_newsize_ok - may this inode be truncated to a given size
1353  * @inode:      the inode to be truncated
1354  * @offset:     the new size to assign to the inode
1355  * @Returns:    0 on success, -ve errno on failure
1356  *
1357  * inode_newsize_ok will check filesystem limits and ulimits to check that the
1358  * new inode size is within limits. inode_newsize_ok will also send SIGXFSZ
1359  * when necessary. Caller must not proceed with inode size change if failure is
1360  * returned. @inode must be a file (not directory), with appropriate
1361  * permissions to allow truncate (inode_newsize_ok does NOT check these
1362  * conditions).
1363  *
1364  * inode_newsize_ok must be called with i_mutex held.
1365  */
1366 int inode_newsize_ok(const struct inode *inode, loff_t offset)
1367 {
1368         if (inode->i_size < offset) {
1369                 unsigned long limit;
1370
1371                 limit = rlimit(RLIMIT_FSIZE);
1372                 if (limit != RLIM_INFINITY && offset > limit)
1373                         goto out_sig;
1374                 if (offset > inode->i_sb->s_maxbytes)
1375                         goto out_big;
1376         } else {
1377                 /*
1378                  * truncation of in-use swapfiles is disallowed - it would
1379                  * cause subsequent swapout to scribble on the now-freed
1380                  * blocks.
1381                  */
1382                 if (IS_SWAPFILE(inode))
1383                         return -ETXTBSY;
1384         }
1385
1386         return 0;
1387 out_sig:
1388         send_sig(SIGXFSZ, current, 0);
1389 out_big:
1390         return -EFBIG;
1391 }
1392 #endif
1393
1394 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1395  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1396  * keep these values until such a time that objects are allocated for it.
1397  * We do the MDS operations first, as it is checking permissions for us.
1398  * We don't to the MDS RPC if there is nothing that we want to store there,
1399  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1400  * going to do an RPC anyways.
1401  *
1402  * If we are doing a truncate, we will send the mtime and ctime updates
1403  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1404  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1405  * at the same time.
1406  */
1407 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr)
1408 {
1409         struct lov_stripe_md *lsm;
1410         struct inode *inode = dentry->d_inode;
1411         struct ll_inode_info *lli = ll_i2info(inode);
1412         struct md_op_data *op_data = NULL;
1413         struct md_open_data *mod = NULL;
1414         int ia_valid = attr->ia_valid;
1415         int rc = 0, rc1 = 0;
1416         ENTRY;
1417
1418         CDEBUG(D_VFSTRACE, "%s: setattr inode %p/fid:"DFID" from %llu to %llu, "
1419                 "valid %x\n", ll_get_fsname(inode->i_sb, NULL, 0), inode,
1420                 PFID(&lli->lli_fid), i_size_read(inode), attr->ia_size,
1421                 attr->ia_valid);
1422
1423         if (ia_valid & ATTR_SIZE) {
1424                 /* Check new size against VFS/VM file size limit and rlimit */
1425                 rc = inode_newsize_ok(inode, attr->ia_size);
1426                 if (rc)
1427                         RETURN(rc);
1428
1429                 /* The maximum Lustre file size is variable, based on the
1430                  * OST maximum object size and number of stripes.  This
1431                  * needs another check in addition to the VFS check above. */
1432                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1433                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1434                                PFID(&lli->lli_fid), attr->ia_size,
1435                                ll_file_maxbytes(inode));
1436                         RETURN(-EFBIG);
1437                 }
1438
1439                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1440         }
1441
1442         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1443         if (ia_valid & TIMES_SET_FLAGS) {
1444                 if (cfs_curproc_fsuid() != inode->i_uid &&
1445                     !cfs_capable(CFS_CAP_FOWNER))
1446                         RETURN(-EPERM);
1447         }
1448
1449         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1450         if (attr->ia_valid & ATTR_CTIME) {
1451                 attr->ia_ctime = CFS_CURRENT_TIME;
1452                 attr->ia_valid |= ATTR_CTIME_SET;
1453         }
1454         if (!(ia_valid & ATTR_ATIME_SET) && (attr->ia_valid & ATTR_ATIME)) {
1455                 attr->ia_atime = CFS_CURRENT_TIME;
1456                 attr->ia_valid |= ATTR_ATIME_SET;
1457         }
1458         if (!(ia_valid & ATTR_MTIME_SET) && (attr->ia_valid & ATTR_MTIME)) {
1459                 attr->ia_mtime = CFS_CURRENT_TIME;
1460                 attr->ia_valid |= ATTR_MTIME_SET;
1461         }
1462
1463         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1464                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1465                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1466                        cfs_time_current_sec());
1467
1468         /* If we are changing file size, file content is modified, flag it. */
1469         if (attr->ia_valid & ATTR_SIZE) {
1470                 spin_lock(&lli->lli_lock);
1471                 lli->lli_flags |= LLIF_DATA_MODIFIED;
1472                 spin_unlock(&lli->lli_lock);
1473         }
1474
1475         /* We always do an MDS RPC, even if we're only changing the size;
1476          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1477
1478         OBD_ALLOC_PTR(op_data);
1479         if (op_data == NULL)
1480                 RETURN(-ENOMEM);
1481
1482         if (!S_ISDIR(inode->i_mode)) {
1483                 if (ia_valid & ATTR_SIZE)
1484                         inode_dio_write_done(inode);
1485                 mutex_unlock(&inode->i_mutex);
1486                 down_write(&lli->lli_trunc_sem);
1487                 mutex_lock(&inode->i_mutex);
1488                 if (ia_valid & ATTR_SIZE)
1489                         inode_dio_wait(inode);
1490         }
1491
1492         /* We need a steady stripe configuration for setattr to avoid
1493          * confusion. */
1494         lsm = ccc_inode_lsm_get(inode);
1495
1496         /* NB: ATTR_SIZE will only be set after this point if the size
1497          * resides on the MDS, ie, this file has no objects. */
1498         if (lsm != NULL)
1499                 attr->ia_valid &= ~ATTR_SIZE;
1500         /* can't call ll_setattr_ost() while holding a refcount of lsm */
1501         ccc_inode_lsm_put(inode, lsm);
1502
1503         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1504
1505         /* Open epoch for truncate. */
1506         if (exp_connect_som(ll_i2mdexp(inode)) &&
1507             (ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1508                 op_data->op_flags = MF_EPOCH_OPEN;
1509
1510         rc = ll_md_setattr(dentry, op_data, &mod);
1511         if (rc)
1512                 GOTO(out, rc);
1513
1514         /* RPC to MDT is sent, cancel data modification flag */
1515         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1516                 spin_lock(&lli->lli_lock);
1517                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1518                 spin_unlock(&lli->lli_lock);
1519         }
1520
1521         ll_ioepoch_open(lli, op_data->op_ioepoch);
1522         if (!S_ISREG(inode->i_mode))
1523                 GOTO(out, rc = 0);
1524
1525         if (ia_valid & ATTR_SIZE)
1526                 attr->ia_valid |= ATTR_SIZE;
1527         if (ia_valid & (ATTR_SIZE |
1528                         ATTR_ATIME | ATTR_ATIME_SET |
1529                         ATTR_MTIME | ATTR_MTIME_SET))
1530                 /* For truncate and utimes sending attributes to OSTs, setting
1531                  * mtime/atime to the past will be performed under PW [0:EOF]
1532                  * extent lock (new_size:EOF for truncate).  It may seem
1533                  * excessive to send mtime/atime updates to OSTs when not
1534                  * setting times to past, but it is necessary due to possible
1535                  * time de-synchronization between MDT inode and OST objects */
1536                 rc = ll_setattr_ost(inode, attr);
1537         EXIT;
1538 out:
1539         if (op_data) {
1540                 if (op_data->op_ioepoch) {
1541                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1542                         if (!rc)
1543                                 rc = rc1;
1544                 }
1545                 ll_finish_md_op_data(op_data);
1546         }
1547         if (!S_ISDIR(inode->i_mode))
1548                 up_write(&lli->lli_trunc_sem);
1549
1550         ll_stats_ops_tally(ll_i2sbi(inode), (ia_valid & ATTR_SIZE) ?
1551                            LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1552
1553         return rc;
1554 }
1555
1556 int ll_setattr(struct dentry *de, struct iattr *attr)
1557 {
1558         int mode = de->d_inode->i_mode;
1559
1560         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1561                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1562                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1563
1564         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1565                                (ATTR_SIZE|ATTR_MODE)) &&
1566             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1567              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1568               !(attr->ia_mode & S_ISGID))))
1569                 attr->ia_valid |= ATTR_FORCE;
1570
1571         if ((mode & S_ISUID) &&
1572             !(attr->ia_mode & S_ISUID) &&
1573             !(attr->ia_valid & ATTR_KILL_SUID))
1574                 attr->ia_valid |= ATTR_KILL_SUID;
1575
1576         if (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1577             !(attr->ia_mode & S_ISGID) &&
1578             !(attr->ia_valid & ATTR_KILL_SGID))
1579                 attr->ia_valid |= ATTR_KILL_SGID;
1580
1581         return ll_setattr_raw(de, attr);
1582 }
1583
1584 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1585                        __u64 max_age, __u32 flags)
1586 {
1587         struct ll_sb_info *sbi = ll_s2sbi(sb);
1588         struct obd_statfs obd_osfs;
1589         int rc;
1590         ENTRY;
1591
1592         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1593         if (rc) {
1594                 CERROR("md_statfs fails: rc = %d\n", rc);
1595                 RETURN(rc);
1596         }
1597
1598         osfs->os_type = sb->s_magic;
1599
1600         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1601                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1602
1603         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1604                 flags |= OBD_STATFS_NODELAY;
1605
1606         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1607         if (rc) {
1608                 CERROR("obd_statfs fails: rc = %d\n", rc);
1609                 RETURN(rc);
1610         }
1611
1612         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1613                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1614                obd_osfs.os_files);
1615
1616         osfs->os_bsize = obd_osfs.os_bsize;
1617         osfs->os_blocks = obd_osfs.os_blocks;
1618         osfs->os_bfree = obd_osfs.os_bfree;
1619         osfs->os_bavail = obd_osfs.os_bavail;
1620
1621         /* If we don't have as many objects free on the OST as inodes
1622          * on the MDS, we reduce the total number of inodes to
1623          * compensate, so that the "inodes in use" number is correct.
1624          */
1625         if (obd_osfs.os_ffree < osfs->os_ffree) {
1626                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1627                         obd_osfs.os_ffree;
1628                 osfs->os_ffree = obd_osfs.os_ffree;
1629         }
1630
1631         RETURN(rc);
1632 }
1633 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1634 {
1635         struct super_block *sb = de->d_sb;
1636         struct obd_statfs osfs;
1637         int rc;
1638
1639         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1640         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1641
1642         /* Some amount of caching on the client is allowed */
1643         rc = ll_statfs_internal(sb, &osfs,
1644                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1645                                 0);
1646         if (rc)
1647                 return rc;
1648
1649         statfs_unpack(sfs, &osfs);
1650
1651         /* We need to downshift for all 32-bit kernels, because we can't
1652          * tell if the kernel is being called via sys_statfs64() or not.
1653          * Stop before overflowing f_bsize - in which case it is better
1654          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1655         if (sizeof(long) < 8) {
1656                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1657                         sfs->f_bsize <<= 1;
1658
1659                         osfs.os_blocks >>= 1;
1660                         osfs.os_bfree >>= 1;
1661                         osfs.os_bavail >>= 1;
1662                 }
1663         }
1664
1665         sfs->f_blocks = osfs.os_blocks;
1666         sfs->f_bfree = osfs.os_bfree;
1667         sfs->f_bavail = osfs.os_bavail;
1668
1669         return 0;
1670 }
1671
1672 void ll_inode_size_lock(struct inode *inode)
1673 {
1674         struct ll_inode_info *lli;
1675
1676         LASSERT(!S_ISDIR(inode->i_mode));
1677
1678         lli = ll_i2info(inode);
1679         LASSERT(lli->lli_size_sem_owner != current);
1680         down(&lli->lli_size_sem);
1681         LASSERT(lli->lli_size_sem_owner == NULL);
1682         lli->lli_size_sem_owner = current;
1683 }
1684
1685 void ll_inode_size_unlock(struct inode *inode)
1686 {
1687         struct ll_inode_info *lli;
1688
1689         lli = ll_i2info(inode);
1690         LASSERT(lli->lli_size_sem_owner == current);
1691         lli->lli_size_sem_owner = NULL;
1692         up(&lli->lli_size_sem);
1693 }
1694
1695 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1696 {
1697         struct ll_inode_info *lli = ll_i2info(inode);
1698         struct mdt_body *body = md->body;
1699         struct lov_stripe_md *lsm = md->lsm;
1700         struct ll_sb_info *sbi = ll_i2sbi(inode);
1701
1702         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1703         if (lsm != NULL) {
1704                 LASSERT(S_ISREG(inode->i_mode));
1705                 CDEBUG(D_INODE, "adding lsm %p to inode %lu/%u(%p)\n",
1706                                 lsm, inode->i_ino, inode->i_generation, inode);
1707                 /* cl_file_inode_init must go before lli_has_smd or a race
1708                  * is possible where client thinks the file has stripes,
1709                  * but lov raid0 is not setup yet and parallel e.g.
1710                  * glimpse would try to use uninitialized lov */
1711                 if (cl_file_inode_init(inode, md) == 0)
1712                         lli->lli_has_smd = true;
1713
1714                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1715                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1716                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1717                 if (md->lsm != NULL)
1718                         obd_free_memmd(ll_i2dtexp(inode), &md->lsm);
1719         }
1720
1721         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1722                 if (body->valid & OBD_MD_FLRMTPERM)
1723                         ll_update_remote_perm(inode, md->remote_perm);
1724         }
1725 #ifdef CONFIG_FS_POSIX_ACL
1726         else if (body->valid & OBD_MD_FLACL) {
1727                 spin_lock(&lli->lli_lock);
1728                 if (lli->lli_posix_acl)
1729                         posix_acl_release(lli->lli_posix_acl);
1730                 lli->lli_posix_acl = md->posix_acl;
1731                 spin_unlock(&lli->lli_lock);
1732         }
1733 #endif
1734         inode->i_ino = cl_fid_build_ino(&body->fid1, 0);
1735         inode->i_generation = cl_fid_build_gen(&body->fid1);
1736
1737         if (body->valid & OBD_MD_FLATIME) {
1738                 if (body->atime > LTIME_S(inode->i_atime))
1739                         LTIME_S(inode->i_atime) = body->atime;
1740                 lli->lli_lvb.lvb_atime = body->atime;
1741         }
1742         if (body->valid & OBD_MD_FLMTIME) {
1743                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1744                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1745                                "to "LPU64"\n", inode->i_ino,
1746                                LTIME_S(inode->i_mtime), body->mtime);
1747                         LTIME_S(inode->i_mtime) = body->mtime;
1748                 }
1749                 lli->lli_lvb.lvb_mtime = body->mtime;
1750         }
1751         if (body->valid & OBD_MD_FLCTIME) {
1752                 if (body->ctime > LTIME_S(inode->i_ctime))
1753                         LTIME_S(inode->i_ctime) = body->ctime;
1754                 lli->lli_lvb.lvb_ctime = body->ctime;
1755         }
1756         if (body->valid & OBD_MD_FLMODE)
1757                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1758         if (body->valid & OBD_MD_FLTYPE)
1759                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1760         LASSERT(inode->i_mode != 0);
1761         if (S_ISREG(inode->i_mode)) {
1762                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1763         } else {
1764                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1765         }
1766         if (body->valid & OBD_MD_FLUID)
1767                 inode->i_uid = body->uid;
1768         if (body->valid & OBD_MD_FLGID)
1769                 inode->i_gid = body->gid;
1770         if (body->valid & OBD_MD_FLFLAGS)
1771                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1772         if (body->valid & OBD_MD_FLNLINK)
1773                 set_nlink(inode, body->nlink);
1774         if (body->valid & OBD_MD_FLRDEV)
1775                 inode->i_rdev = old_decode_dev(body->rdev);
1776
1777         if (body->valid & OBD_MD_FLID) {
1778                 /* FID shouldn't be changed! */
1779                 if (fid_is_sane(&lli->lli_fid)) {
1780                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1781                                  "Trying to change FID "DFID
1782                                  " to the "DFID", inode %lu/%u(%p)\n",
1783                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1784                                  inode->i_ino, inode->i_generation, inode);
1785                 } else
1786                         lli->lli_fid = body->fid1;
1787         }
1788
1789         LASSERT(fid_seq(&lli->lli_fid) != 0);
1790
1791         if (body->valid & OBD_MD_FLSIZE) {
1792                 if (exp_connect_som(ll_i2mdexp(inode)) &&
1793                     S_ISREG(inode->i_mode) && lli->lli_has_smd) {
1794                         struct lustre_handle lockh;
1795                         ldlm_mode_t mode;
1796
1797                         /* As it is possible a blocking ast has been processed
1798                          * by this time, we need to check there is an UPDATE
1799                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
1800                          * it. */
1801                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
1802                                                &lockh, LDLM_FL_CBPENDING);
1803                         if (mode) {
1804                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
1805                                                       LLIF_EPOCH_PENDING |
1806                                                       LLIF_SOM_DIRTY)) {
1807                                         CERROR("ino %lu flags %u still has "
1808                                                "size authority! do not trust "
1809                                                "the size got from MDS\n",
1810                                                inode->i_ino, lli->lli_flags);
1811                                 } else {
1812                                         /* Use old size assignment to avoid
1813                                          * deadlock bz14138 & bz14326 */
1814                                         i_size_write(inode, body->size);
1815                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
1816                                 }
1817                                 ldlm_lock_decref(&lockh, mode);
1818                         }
1819                 } else {
1820                         /* Use old size assignment to avoid
1821                          * deadlock bz14138 & bz14326 */
1822                         i_size_write(inode, body->size);
1823
1824                         CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
1825                                inode->i_ino, (unsigned long long)body->size);
1826                 }
1827
1828                 if (body->valid & OBD_MD_FLBLOCKS)
1829                         inode->i_blocks = body->blocks;
1830         }
1831
1832         if (body->valid & OBD_MD_FLMDSCAPA) {
1833                 LASSERT(md->mds_capa);
1834                 ll_add_capa(inode, md->mds_capa);
1835         }
1836         if (body->valid & OBD_MD_FLOSSCAPA) {
1837                 LASSERT(md->oss_capa);
1838                 ll_add_capa(inode, md->oss_capa);
1839         }
1840 }
1841
1842 void ll_read_inode2(struct inode *inode, void *opaque)
1843 {
1844         struct lustre_md *md = opaque;
1845         struct ll_inode_info *lli = ll_i2info(inode);
1846         ENTRY;
1847
1848         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1849                PFID(&lli->lli_fid), inode);
1850
1851         LASSERT(!lli->lli_has_smd);
1852
1853         /* Core attributes from the MDS first.  This is a new inode, and
1854          * the VFS doesn't zero times in the core inode so we have to do
1855          * it ourselves.  They will be overwritten by either MDS or OST
1856          * attributes - we just need to make sure they aren't newer. */
1857         LTIME_S(inode->i_mtime) = 0;
1858         LTIME_S(inode->i_atime) = 0;
1859         LTIME_S(inode->i_ctime) = 0;
1860         inode->i_rdev = 0;
1861         ll_update_inode(inode, md);
1862
1863         /* OIDEBUG(inode); */
1864
1865         /* initializing backing dev info. */
1866         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
1867
1868
1869         if (S_ISREG(inode->i_mode)) {
1870                 struct ll_sb_info *sbi = ll_i2sbi(inode);
1871                 inode->i_op = &ll_file_inode_operations;
1872                 inode->i_fop = sbi->ll_fop;
1873                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
1874                 EXIT;
1875         } else if (S_ISDIR(inode->i_mode)) {
1876                 inode->i_op = &ll_dir_inode_operations;
1877                 inode->i_fop = &ll_dir_operations;
1878                 EXIT;
1879         } else if (S_ISLNK(inode->i_mode)) {
1880                 inode->i_op = &ll_fast_symlink_inode_operations;
1881                 EXIT;
1882         } else {
1883                 inode->i_op = &ll_special_inode_operations;
1884
1885                 init_special_inode(inode, inode->i_mode,
1886                                    inode->i_rdev);
1887
1888                 EXIT;
1889         }
1890 }
1891
1892 void ll_delete_inode(struct inode *inode)
1893 {
1894         struct cl_inode_info *lli = cl_i2info(inode);
1895         ENTRY;
1896
1897         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
1898                 /* discard all dirty pages before truncating them, required by
1899                  * osc_extent implementation at LU-1030. */
1900                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_DISCARD);
1901
1902         truncate_inode_pages(&inode->i_data, 0);
1903
1904         /* Workaround for LU-118 */
1905         if (inode->i_data.nrpages) {
1906                 TREE_READ_LOCK_IRQ(&inode->i_data);
1907                 TREE_READ_UNLOCK_IRQ(&inode->i_data);
1908                 LASSERTF(inode->i_data.nrpages == 0,
1909                          "inode=%lu/%u(%p) nrpages=%lu, see "
1910                          "http://jira.whamcloud.com/browse/LU-118\n",
1911                          inode->i_ino, inode->i_generation, inode,
1912                          inode->i_data.nrpages);
1913         }
1914         /* Workaround end */
1915
1916 #ifdef HAVE_SBOPS_EVICT_INODE
1917         ll_clear_inode(inode);
1918 #endif
1919         clear_inode(inode);
1920
1921         EXIT;
1922 }
1923
1924 int ll_iocontrol(struct inode *inode, struct file *file,
1925                  unsigned int cmd, unsigned long arg)
1926 {
1927         struct ll_sb_info *sbi = ll_i2sbi(inode);
1928         struct ptlrpc_request *req = NULL;
1929         int rc, flags = 0;
1930         ENTRY;
1931
1932         switch(cmd) {
1933         case FSFILT_IOC_GETFLAGS: {
1934                 struct mdt_body *body;
1935                 struct md_op_data *op_data;
1936
1937                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
1938                                              0, 0, LUSTRE_OPC_ANY,
1939                                              NULL);
1940                 if (IS_ERR(op_data))
1941                         RETURN(PTR_ERR(op_data));
1942
1943                 op_data->op_valid = OBD_MD_FLFLAGS;
1944                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
1945                 ll_finish_md_op_data(op_data);
1946                 if (rc) {
1947                         CERROR("failure %d inode %lu\n", rc, inode->i_ino);
1948                         RETURN(-abs(rc));
1949                 }
1950
1951                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
1952
1953                 flags = body->flags;
1954
1955                 ptlrpc_req_finished(req);
1956
1957                 RETURN(put_user(flags, (int *)arg));
1958         }
1959         case FSFILT_IOC_SETFLAGS: {
1960                 struct lov_stripe_md *lsm;
1961                 struct obd_info oinfo = { { { 0 } } };
1962                 struct md_op_data *op_data;
1963
1964                 if (get_user(flags, (int *)arg))
1965                         RETURN(-EFAULT);
1966
1967                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
1968                                              LUSTRE_OPC_ANY, NULL);
1969                 if (IS_ERR(op_data))
1970                         RETURN(PTR_ERR(op_data));
1971
1972                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
1973                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
1974                 rc = md_setattr(sbi->ll_md_exp, op_data,
1975                                 NULL, 0, NULL, 0, &req, NULL);
1976                 ll_finish_md_op_data(op_data);
1977                 ptlrpc_req_finished(req);
1978                 if (rc)
1979                         RETURN(rc);
1980
1981                 inode->i_flags = ll_ext_to_inode_flags(flags);
1982
1983                 lsm = ccc_inode_lsm_get(inode);
1984                 if (lsm == NULL)
1985                         RETURN(0);
1986
1987                 OBDO_ALLOC(oinfo.oi_oa);
1988                 if (!oinfo.oi_oa) {
1989                         ccc_inode_lsm_put(inode, lsm);
1990                         RETURN(-ENOMEM);
1991                 }
1992                 oinfo.oi_md = lsm;
1993                 oinfo.oi_oa->o_id = lsm->lsm_object_id;
1994                 oinfo.oi_oa->o_seq = lsm->lsm_object_seq;
1995                 oinfo.oi_oa->o_flags = flags;
1996                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
1997                                        OBD_MD_FLGROUP;
1998                 oinfo.oi_capa = ll_mdscapa_get(inode);
1999                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2000                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2001                 capa_put(oinfo.oi_capa);
2002                 OBDO_FREE(oinfo.oi_oa);
2003                 ccc_inode_lsm_put(inode, lsm);
2004
2005                 if (rc && rc != -EPERM && rc != -EACCES)
2006                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2007
2008                 RETURN(rc);
2009         }
2010         default:
2011                 RETURN(-ENOSYS);
2012         }
2013
2014         RETURN(0);
2015 }
2016
2017 int ll_flush_ctx(struct inode *inode)
2018 {
2019         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2020
2021         CDEBUG(D_SEC, "flush context for user %d\n", cfs_curproc_uid());
2022
2023         obd_set_info_async(NULL, sbi->ll_md_exp,
2024                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2025                            0, NULL, NULL);
2026         obd_set_info_async(NULL, sbi->ll_dt_exp,
2027                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2028                            0, NULL, NULL);
2029         return 0;
2030 }
2031
2032 /* umount -f client means force down, don't save state */
2033 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2034 void ll_umount_begin(struct vfsmount *vfsmnt, int flags)
2035 {
2036         struct super_block *sb = vfsmnt->mnt_sb;
2037 #else
2038 void ll_umount_begin(struct super_block *sb)
2039 {
2040 #endif
2041         struct ll_sb_info *sbi = ll_s2sbi(sb);
2042         struct obd_device *obd;
2043         struct obd_ioctl_data *ioc_data;
2044         ENTRY;
2045
2046 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2047         if (!(flags & MNT_FORCE)) {
2048                 EXIT;
2049                 return;
2050         }
2051 #endif
2052
2053         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2054                sb->s_count, atomic_read(&sb->s_active));
2055
2056         obd = class_exp2obd(sbi->ll_md_exp);
2057         if (obd == NULL) {
2058                 CERROR("Invalid MDC connection handle "LPX64"\n",
2059                        sbi->ll_md_exp->exp_handle.h_cookie);
2060                 EXIT;
2061                 return;
2062         }
2063         obd->obd_force = 1;
2064
2065         obd = class_exp2obd(sbi->ll_dt_exp);
2066         if (obd == NULL) {
2067                 CERROR("Invalid LOV connection handle "LPX64"\n",
2068                        sbi->ll_dt_exp->exp_handle.h_cookie);
2069                 EXIT;
2070                 return;
2071         }
2072         obd->obd_force = 1;
2073
2074         OBD_ALLOC_PTR(ioc_data);
2075         if (ioc_data) {
2076                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2077                               sizeof ioc_data, ioc_data, NULL);
2078
2079                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2080                               sizeof ioc_data, ioc_data, NULL);
2081
2082                 OBD_FREE_PTR(ioc_data);
2083         }
2084
2085
2086         /* Really, we'd like to wait until there are no requests outstanding,
2087          * and then continue.  For now, we just invalidate the requests,
2088          * schedule() and sleep one second if needed, and hope.
2089          */
2090         cfs_schedule();
2091 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2092         if (atomic_read(&vfsmnt->mnt_count) > 2) {
2093                 cfs_schedule_timeout_and_set_state(CFS_TASK_INTERRUPTIBLE,
2094                                                    cfs_time_seconds(1));
2095                 if (atomic_read(&vfsmnt->mnt_count) > 2)
2096                         LCONSOLE_WARN("Mount still busy with %d refs! You "
2097                                       "may try to umount it a bit later\n",
2098                                       atomic_read(&vfsmnt->mnt_count));
2099         }
2100 #endif
2101
2102         EXIT;
2103 }
2104
2105 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2106 {
2107         struct ll_sb_info *sbi = ll_s2sbi(sb);
2108         char *profilenm = get_profile_name(sb);
2109         int err;
2110         __u32 read_only;
2111
2112         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2113                 read_only = *flags & MS_RDONLY;
2114                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2115                                          sizeof(KEY_READ_ONLY),
2116                                          KEY_READ_ONLY, sizeof(read_only),
2117                                          &read_only, NULL);
2118                 if (err) {
2119                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2120                                       profilenm, read_only ?
2121                                       "read-only" : "read-write", err);
2122                         return err;
2123                 }
2124
2125                 if (read_only)
2126                         sb->s_flags |= MS_RDONLY;
2127                 else
2128                         sb->s_flags &= ~MS_RDONLY;
2129
2130                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2131                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2132                                       read_only ?  "read-only" : "read-write");
2133         }
2134         return 0;
2135 }
2136
2137 int ll_prep_inode(struct inode **inode,
2138                   struct ptlrpc_request *req,
2139                   struct super_block *sb)
2140 {
2141         struct ll_sb_info *sbi = NULL;
2142         struct lustre_md md;
2143         __u64 ibits;
2144         int rc;
2145         ENTRY;
2146
2147         LASSERT(*inode || sb);
2148         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2149         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2150                               sbi->ll_md_exp, &md);
2151         if (rc)
2152                 RETURN(rc);
2153
2154         if (*inode) {
2155                 ll_update_inode(*inode, &md);
2156         } else {
2157                 LASSERT(sb != NULL);
2158
2159                 /*
2160                  * At this point server returns to client's same fid as client
2161                  * generated for creating. So using ->fid1 is okay here.
2162                  */
2163                 LASSERT(fid_is_sane(&md.body->fid1));
2164
2165                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1, 0), &md);
2166                 if (*inode == NULL || IS_ERR(*inode)) {
2167                         if (md.lsm)
2168                                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2169 #ifdef CONFIG_FS_POSIX_ACL
2170                         if (md.posix_acl) {
2171                                 posix_acl_release(md.posix_acl);
2172                                 md.posix_acl = NULL;
2173                         }
2174 #endif
2175                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2176                         *inode = NULL;
2177                         CERROR("new_inode -fatal: rc %d\n", rc);
2178                         GOTO(out, rc);
2179                 }
2180         }
2181
2182         /* sanity check for LAYOUT lock. */
2183         ibits = MDS_INODELOCK_LAYOUT;
2184         if (S_ISREG(md.body->mode) && sbi->ll_flags & LL_SBI_LAYOUT_LOCK &&
2185             md.lsm != NULL && !ll_have_md_lock(*inode, &ibits, LCK_MINMODE)) {
2186                 CERROR("%s: inode "DFID" (%p) layout lock not granted.\n",
2187                         ll_get_fsname(sb, NULL, 0),
2188                         PFID(ll_inode2fid(*inode)), *inode);
2189         }
2190
2191 out:
2192         md_free_lustre_md(sbi->ll_md_exp, &md);
2193         RETURN(rc);
2194 }
2195
2196 int ll_obd_statfs(struct inode *inode, void *arg)
2197 {
2198         struct ll_sb_info *sbi = NULL;
2199         struct obd_export *exp;
2200         char *buf = NULL;
2201         struct obd_ioctl_data *data = NULL;
2202         __u32 type;
2203         __u32 flags;
2204         int len = 0, rc;
2205
2206         if (!inode || !(sbi = ll_i2sbi(inode)))
2207                 GOTO(out_statfs, rc = -EINVAL);
2208
2209         rc = obd_ioctl_getdata(&buf, &len, arg);
2210         if (rc)
2211                 GOTO(out_statfs, rc);
2212
2213         data = (void*)buf;
2214         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2215             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2216                 GOTO(out_statfs, rc = -EINVAL);
2217
2218         if (data->ioc_inllen1 != sizeof(__u32) ||
2219             data->ioc_inllen2 != sizeof(__u32) ||
2220             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2221             data->ioc_plen2 != sizeof(struct obd_uuid))
2222                 GOTO(out_statfs, rc = -EINVAL);
2223
2224         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2225         if (type & LL_STATFS_LMV)
2226                 exp = sbi->ll_md_exp;
2227         else if (type & LL_STATFS_LOV)
2228                 exp = sbi->ll_dt_exp;
2229         else
2230                 GOTO(out_statfs, rc = -ENODEV);
2231
2232         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2233         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2234         if (rc)
2235                 GOTO(out_statfs, rc);
2236 out_statfs:
2237         if (buf)
2238                 obd_ioctl_freedata(buf, len);
2239         return rc;
2240 }
2241
2242 int ll_process_config(struct lustre_cfg *lcfg)
2243 {
2244         char *ptr;
2245         void *sb;
2246         struct lprocfs_static_vars lvars;
2247         unsigned long x;
2248         int rc = 0;
2249
2250         lprocfs_llite_init_vars(&lvars);
2251
2252         /* The instance name contains the sb: lustre-client-aacfe000 */
2253         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2254         if (!ptr || !*(++ptr))
2255                 return -EINVAL;
2256         if (sscanf(ptr, "%lx", &x) != 1)
2257                 return -EINVAL;
2258         sb = (void *)x;
2259         /* This better be a real Lustre superblock! */
2260         LASSERT(s2lsi((struct super_block *)sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2261
2262         /* Note we have not called client_common_fill_super yet, so
2263            proc fns must be able to handle that! */
2264         rc = class_process_proc_param(PARAM_LLITE, lvars.obd_vars,
2265                                       lcfg, sb);
2266         if (rc > 0)
2267                 rc = 0;
2268         return(rc);
2269 }
2270
2271 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2272 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2273                                        struct inode *i1, struct inode *i2,
2274                                        const char *name, int namelen,
2275                                        int mode, __u32 opc, void *data)
2276 {
2277         LASSERT(i1 != NULL);
2278
2279         if (namelen > ll_i2sbi(i1)->ll_namelen)
2280                 return ERR_PTR(-ENAMETOOLONG);
2281
2282         if (op_data == NULL)
2283                 OBD_ALLOC_PTR(op_data);
2284
2285         if (op_data == NULL)
2286                 return ERR_PTR(-ENOMEM);
2287
2288         ll_i2gids(op_data->op_suppgids, i1, i2);
2289         op_data->op_fid1 = *ll_inode2fid(i1);
2290         op_data->op_capa1 = ll_mdscapa_get(i1);
2291
2292         if (i2) {
2293                 op_data->op_fid2 = *ll_inode2fid(i2);
2294                 op_data->op_capa2 = ll_mdscapa_get(i2);
2295         } else {
2296                 fid_zero(&op_data->op_fid2);
2297                 op_data->op_capa2 = NULL;
2298         }
2299
2300         op_data->op_name = name;
2301         op_data->op_namelen = namelen;
2302         op_data->op_mode = mode;
2303         op_data->op_mod_time = cfs_time_current_sec();
2304         op_data->op_fsuid = cfs_curproc_fsuid();
2305         op_data->op_fsgid = cfs_curproc_fsgid();
2306         op_data->op_cap = cfs_curproc_cap_pack();
2307         op_data->op_bias = MDS_CHECK_SPLIT;
2308         op_data->op_opc = opc;
2309         op_data->op_mds = 0;
2310         op_data->op_data = data;
2311
2312         /* If the file is being opened after mknod() (normally due to NFS)
2313          * try to use the default stripe data from parent directory for
2314          * allocating OST objects.  Try to pass the parent FID to MDS. */
2315         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2316             !ll_i2info(i2)->lli_has_smd) {
2317                 struct ll_inode_info *lli = ll_i2info(i2);
2318
2319                 spin_lock(&lli->lli_lock);
2320                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2321                         op_data->op_fid1 = lli->lli_pfid;
2322                 spin_unlock(&lli->lli_lock);
2323                 /** We ignore parent's capability temporary. */
2324         }
2325
2326         /* When called by ll_setattr_raw, file is i1. */
2327         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2328                 op_data->op_bias |= MDS_DATA_MODIFIED;
2329
2330         return op_data;
2331 }
2332
2333 void ll_finish_md_op_data(struct md_op_data *op_data)
2334 {
2335         capa_put(op_data->op_capa1);
2336         capa_put(op_data->op_capa2);
2337         OBD_FREE_PTR(op_data);
2338 }
2339
2340 #ifdef HAVE_SUPEROPS_USE_DENTRY
2341 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2342 #else
2343 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2344 #endif
2345 {
2346         struct ll_sb_info *sbi;
2347
2348 #ifdef HAVE_SUPEROPS_USE_DENTRY
2349         LASSERT((seq != NULL) && (dentry != NULL));
2350         sbi = ll_s2sbi(dentry->d_sb);
2351 #else
2352         LASSERT((seq != NULL) && (vfs != NULL));
2353         sbi = ll_s2sbi(vfs->mnt_sb);
2354 #endif
2355
2356         if (sbi->ll_flags & LL_SBI_NOLCK)
2357                 seq_puts(seq, ",nolock");
2358
2359         if (sbi->ll_flags & LL_SBI_FLOCK)
2360                 seq_puts(seq, ",flock");
2361
2362         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2363                 seq_puts(seq, ",localflock");
2364
2365         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2366                 seq_puts(seq, ",user_xattr");
2367
2368         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2369                 seq_puts(seq, ",lazystatfs");
2370
2371         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2372                 seq_puts(seq, ",user_fid2path");
2373
2374         RETURN(0);
2375 }
2376
2377 /**
2378  * Get obd name by cmd, and copy out to user space
2379  */
2380 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2381 {
2382         struct ll_sb_info *sbi = ll_i2sbi(inode);
2383         struct obd_device *obd;
2384         ENTRY;
2385
2386         if (cmd == OBD_IOC_GETDTNAME)
2387                 obd = class_exp2obd(sbi->ll_dt_exp);
2388         else if (cmd == OBD_IOC_GETMDNAME)
2389                 obd = class_exp2obd(sbi->ll_md_exp);
2390         else
2391                 RETURN(-EINVAL);
2392
2393         if (!obd)
2394                 RETURN(-ENOENT);
2395
2396         if (cfs_copy_to_user((void *)arg, obd->obd_name,
2397                              strlen(obd->obd_name) + 1))
2398                 RETURN(-EFAULT);
2399
2400         RETURN(0);
2401 }
2402
2403 /**
2404  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2405  * fsname will be returned in this buffer; otherwise, a static buffer will be
2406  * used to store the fsname and returned to caller.
2407  */
2408 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2409 {
2410         static char fsname_static[MTI_NAME_MAXLEN];
2411         struct lustre_sb_info *lsi = s2lsi(sb);
2412         char *ptr;
2413         int len;
2414
2415         if (buf == NULL) {
2416                 /* this means the caller wants to use static buffer
2417                  * and it doesn't care about race. Usually this is
2418                  * in error reporting path */
2419                 buf = fsname_static;
2420                 buflen = sizeof(fsname_static);
2421         }
2422
2423         len = strlen(lsi->lsi_lmd->lmd_profile);
2424         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2425         if (ptr && (strcmp(ptr, "-client") == 0))
2426                 len -= 7;
2427
2428         if (unlikely(len >= buflen))
2429                 len = buflen - 1;
2430         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2431         buf[len] = '\0';
2432
2433         return buf;
2434 }
2435
2436 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2437 {
2438         char *path = NULL;
2439
2440 #ifdef HAVE_FS_STRUCT_USE_PATH
2441         struct path p;
2442
2443         p.dentry = dentry;
2444         p.mnt = current->fs->root.mnt;
2445         path_get(&p);
2446         path = d_path(&p, buf, bufsize);
2447         path_put(&p);
2448 #else
2449         path = d_path(dentry, current->fs->rootmnt, buf, bufsize);
2450 #endif
2451
2452         return path;
2453 }
2454
2455 void ll_dirty_page_discard_warn(cfs_page_t *page, int ioret)
2456 {
2457         char *buf, *path = NULL;
2458         struct dentry *dentry = NULL;
2459         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2460
2461         /* this can be called inside spin lock so use GFP_ATOMIC. */
2462         buf = (char *)__get_free_page(GFP_ATOMIC);
2463         if (buf != NULL) {
2464                 dentry = d_find_alias(page->mapping->host);
2465                 if (dentry != NULL)
2466                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2467         }
2468
2469         CWARN("%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2470               "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2471               s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2472               PFID(&obj->cob_header.coh_lu.loh_fid),
2473               (path && !IS_ERR(path)) ? path : "", ioret);
2474
2475         if (dentry != NULL)
2476                 dput(dentry);
2477
2478         if (buf != NULL)
2479                 free_page((unsigned long)buf);
2480 }