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, 2013, 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 struct kmem_cache *ll_file_data_slab;
  60
  61 static LIST_HEAD(ll_super_blocks);
  62 static DEFINE_SPINLOCK(ll_sb_lock);
  63
  64 #ifndef log2
  65 #define log2(n) ffz(~(n))
  66 #endif
  67
  68 static struct ll_sb_info *ll_init_sbi(void)
  69 {
  70         struct ll_sb_info *sbi = NULL;
  71         unsigned long pages;
  72         unsigned long lru_page_max;
  73         struct sysinfo si;
  74         class_uuid_t uuid;
  75         int i;
  76         ENTRY;
  77
  78         OBD_ALLOC(sbi, sizeof(*sbi));
  79         if (!sbi)
  80                 RETURN(NULL);
  81
  82         spin_lock_init(&sbi->ll_lock);
  83         mutex_init(&sbi->ll_lco.lco_lock);
  84         spin_lock_init(&sbi->ll_pp_extent_lock);
  85         spin_lock_init(&sbi->ll_process_lock);
  86         sbi->ll_rw_stats_on = 0;
  87
  88         si_meminfo(&si);
  89         pages = si.totalram - si.totalhigh;
  90         lru_page_max = pages / 2;
  91
  92         /* initialize ll_cache data */
  93         atomic_set(&sbi->ll_cache.ccc_users, 0);
  94         sbi->ll_cache.ccc_lru_max = lru_page_max;
  95         atomic_set(&sbi->ll_cache.ccc_lru_left, lru_page_max);
  96         spin_lock_init(&sbi->ll_cache.ccc_lru_lock);
  97         CFS_INIT_LIST_HEAD(&sbi->ll_cache.ccc_lru);
  98
  99         atomic_set(&sbi->ll_cache.ccc_unstable_nr, 0);
 100         init_waitqueue_head(&sbi->ll_cache.ccc_unstable_waitq);
 101
 102         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 103                                            SBI_DEFAULT_READAHEAD_MAX);
 104         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 105         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 106                                            SBI_DEFAULT_READAHEAD_WHOLE_MAX;
 107         CFS_INIT_LIST_HEAD(&sbi->ll_conn_chain);
 108         CFS_INIT_LIST_HEAD(&sbi->ll_orphan_dentry_list);
 109
 110         ll_generate_random_uuid(uuid);
 111         class_uuid_unparse(uuid, &sbi->ll_sb_uuid);
 112         CDEBUG(D_CONFIG, "generated uuid: %s\n", sbi->ll_sb_uuid.uuid);
 113
 114         spin_lock(&ll_sb_lock);
 115         cfs_list_add_tail(&sbi->ll_list, &ll_super_blocks);
 116         spin_unlock(&ll_sb_lock);
 117
 118         sbi->ll_flags |= LL_SBI_VERBOSE;
 119 #ifdef ENABLE_CHECKSUM
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121 #endif
 122
 123 #ifdef HAVE_LRU_RESIZE_SUPPORT
 124         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 125 #endif
 126
 127         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 128                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 129                                pp_r_hist.oh_lock);
 130                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 131                                pp_w_hist.oh_lock);
 132         }
 133
 134         /* metadata statahead is enabled by default */
 135         sbi->ll_sa_max = LL_SA_RPC_DEF;
 136         atomic_set(&sbi->ll_sa_total, 0);
 137         atomic_set(&sbi->ll_sa_wrong, 0);
 138         atomic_set(&sbi->ll_agl_total, 0);
 139         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 140
 141         RETURN(sbi);
 142 }
 143
 144 static void ll_free_sbi(struct super_block *sb)
 145 {
 146         struct ll_sb_info *sbi = ll_s2sbi(sb);
 147         ENTRY;
 148
 149         if (sbi != NULL) {
 150                 spin_lock(&ll_sb_lock);
 151                 cfs_list_del(&sbi->ll_list);
 152                 spin_unlock(&ll_sb_lock);
 153                 OBD_FREE(sbi, sizeof(*sbi));
 154         }
 155         EXIT;
 156 }
 157
 158 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 159                                     struct vfsmount *mnt)
 160 {
 161         struct inode *root = 0;
 162         struct ll_sb_info *sbi = ll_s2sbi(sb);
 163         struct obd_device *obd;
 164         struct obd_capa *oc = NULL;
 165         struct obd_statfs *osfs = NULL;
 166         struct ptlrpc_request *request = NULL;
 167         struct obd_connect_data *data = NULL;
 168         struct obd_uuid *uuid;
 169         struct md_op_data *op_data;
 170         struct lustre_md lmd;
 171         obd_valid valid;
 172         int size, err, checksum;
 173         ENTRY;
 174
 175         obd = class_name2obd(md);
 176         if (!obd) {
 177                 CERROR("MD %s: not setup or attached\n", md);
 178                 RETURN(-EINVAL);
 179         }
 180
 181         OBD_ALLOC_PTR(data);
 182         if (data == NULL)
 183                 RETURN(-ENOMEM);
 184
 185         OBD_ALLOC_PTR(osfs);
 186         if (osfs == NULL) {
 187                 OBD_FREE_PTR(data);
 188                 RETURN(-ENOMEM);
 189         }
 190
 191         /* indicate the features supported by this client */
 192         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 193                                   OBD_CONNECT_ATTRFID  |
 194                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 195                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 196                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 197                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 198                                   OBD_CONNECT_RMT_CLIENT | OBD_CONNECT_VBR    |
 199                                   OBD_CONNECT_FULL20   | OBD_CONNECT_64BITHASH|
 200                                   OBD_CONNECT_EINPROGRESS |
 201                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 202                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS |
 203                                   OBD_CONNECT_MAX_EASIZE |
 204                                   OBD_CONNECT_FLOCK_DEAD |
 205                                   OBD_CONNECT_DISP_STRIPE;
 206
 207         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 208                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 209
 210 #ifdef HAVE_LRU_RESIZE_SUPPORT
 211         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 212                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 213 #endif
 214 #ifdef CONFIG_FS_POSIX_ACL
 215         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK;
 216 #endif
 217
 218         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 219                 /* flag mdc connection as lightweight, only used for test
 220                  * purpose, use with care */
 221                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 222
 223         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 224         data->ocd_version = LUSTRE_VERSION_CODE;
 225
 226         if (sb->s_flags & MS_RDONLY)
 227                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 228         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 229                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 230
 231 #ifdef HAVE_MS_FLOCK_LOCK
 232         /* force vfs to use lustre handler for flock() calls - bug 10743 */
 233         sb->s_flags |= MS_FLOCK_LOCK;
 234 #endif
 235 #ifdef MS_HAS_NEW_AOPS
 236         sb->s_flags |= MS_HAS_NEW_AOPS;
 237 #endif
 238
 239         if (sbi->ll_flags & LL_SBI_FLOCK)
 240                 sbi->ll_fop = &ll_file_operations_flock;
 241         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 242                 sbi->ll_fop = &ll_file_operations;
 243         else
 244                 sbi->ll_fop = &ll_file_operations_noflock;
 245
 246         /* real client */
 247         data->ocd_connect_flags |= OBD_CONNECT_REAL;
 248         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 249                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 250
 251         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 252
 253         err = obd_connect(NULL, &sbi->ll_md_exp, obd, &sbi->ll_sb_uuid, data, NULL);
 254         if (err == -EBUSY) {
 255                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 256                                    "recovery, of which this client is not a "
 257                                    "part. Please wait for recovery to complete,"
 258                                    " abort, or time out.\n", md);
 259                 GOTO(out, err);
 260         } else if (err) {
 261                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 262                 GOTO(out, err);
 263         }
 264
 265         sbi->ll_md_exp->exp_connect_data = *data;
 266
 267         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 268                            LUSTRE_SEQ_METADATA);
 269         if (err) {
 270                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 271                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 272                 GOTO(out_md, err);
 273         }
 274
 275         /* For mount, we only need fs info from MDT0, and also in DNE, it
 276          * can make sure the client can be mounted as long as MDT0 is
 277          * avaible */
 278         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 279                         cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 280                         OBD_STATFS_FOR_MDT0);
 281         if (err)
 282                 GOTO(out_md_fid, err);
 283
 284         /* This needs to be after statfs to ensure connect has finished.
 285          * Note that "data" does NOT contain the valid connect reply.
 286          * If connecting to a 1.8 server there will be no LMV device, so
 287          * we can access the MDC export directly and exp_connect_flags will
 288          * be non-zero, but if accessing an upgraded 2.1 server it will
 289          * have the correct flags filled in.
 290          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 291         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 292         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 293             valid != CLIENT_CONNECT_MDT_REQD) {
 294                 char *buf;
 295
 296                 OBD_ALLOC_WAIT(buf, PAGE_CACHE_SIZE);
 297                 obd_connect_flags2str(buf, PAGE_CACHE_SIZE,
 298                                       valid ^ CLIENT_CONNECT_MDT_REQD, ",");
 299                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 300                                    "feature(s) needed for correct operation "
 301                                    "of this client (%s). Please upgrade "
 302                                    "server or downgrade client.\n",
 303                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 304                 OBD_FREE(buf, PAGE_CACHE_SIZE);
 305                 GOTO(out_md_fid, err = -EPROTO);
 306         }
 307
 308         size = sizeof(*data);
 309         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 310                            KEY_CONN_DATA,  &size, data, NULL);
 311         if (err) {
 312                 CERROR("%s: Get connect data failed: rc = %d\n",
 313                        sbi->ll_md_exp->exp_obd->obd_name, err);
 314                 GOTO(out_md_fid, err);
 315         }
 316
 317         LASSERT(osfs->os_bsize);
 318         sb->s_blocksize = osfs->os_bsize;
 319         sb->s_blocksize_bits = log2(osfs->os_bsize);
 320         sb->s_magic = LL_SUPER_MAGIC;
 321         sb->s_maxbytes = MAX_LFS_FILESIZE;
 322         sbi->ll_namelen = osfs->os_namelen;
 323         sbi->ll_max_rw_chunk = LL_DEFAULT_MAX_RW_CHUNK;
 324
 325         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 326             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 327                 LCONSOLE_INFO("Disabling user_xattr feature because "
 328                               "it is not supported on the server\n");
 329                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 330         }
 331
 332         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 333 #ifdef MS_POSIXACL
 334                 sb->s_flags |= MS_POSIXACL;
 335 #endif
 336                 sbi->ll_flags |= LL_SBI_ACL;
 337         } else {
 338                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 339 #ifdef MS_POSIXACL
 340                 sb->s_flags &= ~MS_POSIXACL;
 341 #endif
 342                 sbi->ll_flags &= ~LL_SBI_ACL;
 343         }
 344
 345         if (data->ocd_connect_flags & OBD_CONNECT_RMT_CLIENT) {
 346                 if (!(sbi->ll_flags & LL_SBI_RMT_CLIENT)) {
 347                         sbi->ll_flags |= LL_SBI_RMT_CLIENT;
 348                         LCONSOLE_INFO("client is set as remote by default.\n");
 349                 }
 350         } else {
 351                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 352                         sbi->ll_flags &= ~LL_SBI_RMT_CLIENT;
 353                         LCONSOLE_INFO("client claims to be remote, but server "
 354                                       "rejected, forced to be local.\n");
 355                 }
 356         }
 357
 358         if (data->ocd_connect_flags & OBD_CONNECT_MDS_CAPA) {
 359                 LCONSOLE_INFO("client enabled MDS capability!\n");
 360                 sbi->ll_flags |= LL_SBI_MDS_CAPA;
 361         }
 362
 363         if (data->ocd_connect_flags & OBD_CONNECT_OSS_CAPA) {
 364                 LCONSOLE_INFO("client enabled OSS capability!\n");
 365                 sbi->ll_flags |= LL_SBI_OSS_CAPA;
 366         }
 367
 368         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 369                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 370
 371         if (data->ocd_connect_flags & OBD_CONNECT_BRW_SIZE)
 372                 sbi->ll_md_brw_size = data->ocd_brw_size;
 373         else
 374                 sbi->ll_md_brw_size = PAGE_CACHE_SIZE;
 375
 376         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 377                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 378
 379         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 380                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 381                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 382                                       "unknown maximum xattr size.\n", dt);
 383                 } else {
 384                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 385                         sbi->ll_xattr_cache_enabled = 1;
 386                 }
 387         }
 388
 389         obd = class_name2obd(dt);
 390         if (!obd) {
 391                 CERROR("DT %s: not setup or attached\n", dt);
 392                 GOTO(out_md_fid, err = -ENODEV);
 393         }
 394
 395         data->ocd_connect_flags = OBD_CONNECT_GRANT     | OBD_CONNECT_VERSION  |
 396                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 397                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID      |
 398                                   OBD_CONNECT_SRVLOCK   | OBD_CONNECT_TRUNCLOCK|
 399                                   OBD_CONNECT_AT | OBD_CONNECT_RMT_CLIENT |
 400                                   OBD_CONNECT_OSS_CAPA | OBD_CONNECT_VBR|
 401                                   OBD_CONNECT_FULL20 | OBD_CONNECT_64BITHASH |
 402                                   OBD_CONNECT_MAXBYTES |
 403                                   OBD_CONNECT_EINPROGRESS |
 404                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 405                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
 406
 407         if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
 408                 data->ocd_connect_flags |= OBD_CONNECT_SOM;
 409
 410         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
 411                 /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 412                  * disabled by default, because it can still be enabled on the
 413                  * fly via /proc. As a consequence, we still need to come to an
 414                  * agreement on the supported algorithms at connect time */
 415                 data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 416
 417                 if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 418                         data->ocd_cksum_types = OBD_CKSUM_ADLER;
 419                 else
 420                         data->ocd_cksum_types = cksum_types_supported_client();
 421         }
 422
 423 #ifdef HAVE_LRU_RESIZE_SUPPORT
 424         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 425 #endif
 426         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 427                 data->ocd_connect_flags |= OBD_CONNECT_RMT_CLIENT_FORCE;
 428
 429         CDEBUG(D_RPCTRACE, "ocd_connect_flags: "LPX64" ocd_version: %d "
 430                "ocd_grant: %d\n", data->ocd_connect_flags,
 431                data->ocd_version, data->ocd_grant);
 432
 433         obd->obd_upcall.onu_owner = &sbi->ll_lco;
 434         obd->obd_upcall.onu_upcall = cl_ocd_update;
 435
 436         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 437
 438         err = obd_connect(NULL, &sbi->ll_dt_exp, obd, &sbi->ll_sb_uuid, data,
 439                           NULL);
 440         if (err == -EBUSY) {
 441                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 442                                    "recovery, of which this client is not a "
 443                                    "part.  Please wait for recovery to "
 444                                    "complete, abort, or time out.\n", dt);
 445                 GOTO(out_md, err);
 446         } else if (err) {
 447                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 448                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 449                 GOTO(out_md, err);
 450         }
 451
 452         sbi->ll_dt_exp->exp_connect_data = *data;
 453
 454         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 455                            LUSTRE_SEQ_METADATA);
 456         if (err) {
 457                 CERROR("%s: Can't init data layer FID infrastructure, "
 458                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 459                 GOTO(out_dt, err);
 460         }
 461
 462         mutex_lock(&sbi->ll_lco.lco_lock);
 463         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 464         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 465         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 466         mutex_unlock(&sbi->ll_lco.lco_lock);
 467
 468         fid_zero(&sbi->ll_root_fid);
 469         err = md_getstatus(sbi->ll_md_exp, &sbi->ll_root_fid, &oc);
 470         if (err) {
 471                 CERROR("cannot mds_connect: rc = %d\n", err);
 472                 GOTO(out_lock_cn_cb, err);
 473         }
 474         if (!fid_is_sane(&sbi->ll_root_fid)) {
 475                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 476                        sbi->ll_md_exp->exp_obd->obd_name,
 477                        PFID(&sbi->ll_root_fid));
 478                 GOTO(out_lock_cn_cb, err = -EINVAL);
 479         }
 480         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 481
 482         sb->s_op = &lustre_super_operations;
 483 #if THREAD_SIZE >= 8192 /*b=17630*/
 484         sb->s_export_op = &lustre_export_operations;
 485 #endif
 486
 487         /* make root inode
 488          * XXX: move this to after cbd setup? */
 489         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMDSCAPA |
 490                 OBD_MD_FLMODEASIZE;
 491         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 492                 valid |= OBD_MD_FLRMTPERM;
 493         else if (sbi->ll_flags & LL_SBI_ACL)
 494                 valid |= OBD_MD_FLACL;
 495
 496         OBD_ALLOC_PTR(op_data);
 497         if (op_data == NULL)
 498                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 499
 500         op_data->op_fid1 = sbi->ll_root_fid;
 501         op_data->op_mode = 0;
 502         op_data->op_capa1 = oc;
 503         op_data->op_valid = valid;
 504
 505         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 506         if (oc)
 507                 capa_put(oc);
 508         OBD_FREE_PTR(op_data);
 509         if (err) {
 510                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 511                        sbi->ll_md_exp->exp_obd->obd_name, err);
 512                 GOTO(out_lock_cn_cb, err);
 513         }
 514
 515         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 516                                sbi->ll_md_exp, &lmd);
 517         if (err) {
 518                 CERROR("failed to understand root inode md: rc = %d\n", err);
 519                 ptlrpc_req_finished(request);
 520                 GOTO(out_lock_cn_cb, err);
 521         }
 522
 523         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 524         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 525                                             sbi->ll_flags & LL_SBI_32BIT_API),
 526                        &lmd);
 527         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 528         ptlrpc_req_finished(request);
 529
 530         if (root == NULL || IS_ERR(root)) {
 531                 if (lmd.lsm)
 532                         obd_free_memmd(sbi->ll_dt_exp, &lmd.lsm);
 533 #ifdef CONFIG_FS_POSIX_ACL
 534                 if (lmd.posix_acl) {
 535                         posix_acl_release(lmd.posix_acl);
 536                         lmd.posix_acl = NULL;
 537                 }
 538 #endif
 539                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 540                 root = NULL;
 541                 CERROR("lustre_lite: bad iget4 for root\n");
 542                 GOTO(out_root, err);
 543         }
 544
 545         err = ll_close_thread_start(&sbi->ll_lcq);
 546         if (err) {
 547                 CERROR("cannot start close thread: rc %d\n", err);
 548                 GOTO(out_root, err);
 549         }
 550
 551 #ifdef CONFIG_FS_POSIX_ACL
 552         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 553                 rct_init(&sbi->ll_rct);
 554                 et_init(&sbi->ll_et);
 555         }
 556 #endif
 557
 558         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 559         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 560                                  KEY_CHECKSUM, sizeof(checksum), &checksum,
 561                                  NULL);
 562         cl_sb_init(sb);
 563
 564         err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CACHE_SET),
 565                                  KEY_CACHE_SET, sizeof(sbi->ll_cache),
 566                                  &sbi->ll_cache, NULL);
 567
 568         sb->s_root = d_make_root(root);
 569         if (sb->s_root == NULL) {
 570                 CERROR("%s: can't make root dentry\n",
 571                         ll_get_fsname(sb, NULL, 0));
 572                 GOTO(out_root, err = -ENOMEM);
 573         }
 574 #ifdef HAVE_DCACHE_LOCK
 575         sb->s_root->d_op = &ll_d_ops;
 576 #endif
 577
 578         sbi->ll_sdev_orig = sb->s_dev;
 579
 580         /* We set sb->s_dev equal on all lustre clients in order to support
 581          * NFS export clustering.  NFSD requires that the FSID be the same
 582          * on all clients. */
 583         /* s_dev is also used in lt_compare() to compare two fs, but that is
 584          * only a node-local comparison. */
 585         uuid = obd_get_uuid(sbi->ll_md_exp);
 586         if (uuid != NULL)
 587                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 588
 589         if (data != NULL)
 590                 OBD_FREE_PTR(data);
 591         if (osfs != NULL)
 592                 OBD_FREE_PTR(osfs);
 593         if (proc_lustre_fs_root != NULL) {
 594                 err = lprocfs_register_mountpoint(proc_lustre_fs_root, sb,
 595                                                   dt, md);
 596                 if (err < 0) {
 597                         CERROR("%s: could not register mount in lprocfs: "
 598                                "rc = %d\n", ll_get_fsname(sb, NULL, 0), err);
 599                         err = 0;
 600                 }
 601         }
 602
 603         RETURN(err);
 604 out_root:
 605         if (root)
 606                 iput(root);
 607 out_lock_cn_cb:
 608         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 609 out_dt:
 610         obd_disconnect(sbi->ll_dt_exp);
 611         sbi->ll_dt_exp = NULL;
 612         /* Make sure all OScs are gone, since cl_cache is accessing sbi. */
 613         obd_zombie_barrier();
 614 out_md_fid:
 615         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 616 out_md:
 617         obd_disconnect(sbi->ll_md_exp);
 618         sbi->ll_md_exp = NULL;
 619 out:
 620         if (data != NULL)
 621                 OBD_FREE_PTR(data);
 622         if (osfs != NULL)
 623                 OBD_FREE_PTR(osfs);
 624         return err;
 625 }
 626
 627 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 628 {
 629         int size, rc;
 630
 631         *lmmsize = obd_size_diskmd(sbi->ll_dt_exp, NULL);
 632         size = sizeof(int);
 633         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 634                           KEY_MAX_EASIZE, &size, lmmsize, NULL);
 635         if (rc)
 636                 CERROR("Get max mdsize error rc %d\n", rc);
 637
 638         RETURN(rc);
 639 }
 640
 641 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 642 {
 643         int size, rc;
 644
 645         size = sizeof(int);
 646         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 647                          KEY_DEFAULT_EASIZE, &size, lmmsize, NULL);
 648         if (rc)
 649                 CERROR("Get default mdsize error rc %d\n", rc);
 650
 651         RETURN(rc);
 652 }
 653
 654 int ll_get_max_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 655 {
 656         int size, rc;
 657
 658         size = sizeof(int);
 659         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_COOKIESIZE),
 660                           KEY_MAX_COOKIESIZE, &size, lmmsize, NULL);
 661         if (rc)
 662                 CERROR("Get max cookiesize error rc %d\n", rc);
 663
 664         RETURN(rc);
 665 }
 666
 667 int ll_get_default_cookiesize(struct ll_sb_info *sbi, int *lmmsize)
 668 {
 669         int size, rc;
 670
 671         size = sizeof(int);
 672         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_COOKIESIZE),
 673                           KEY_DEFAULT_COOKIESIZE, &size, lmmsize, NULL);
 674         if (rc)
 675                 CERROR("Get default cookiesize error rc %d\n", rc);
 676
 677         RETURN(rc);
 678 }
 679
 680 static void ll_dump_inode(struct inode *inode)
 681 {
 682         struct ll_d_hlist_node *tmp;
 683         int dentry_count = 0;
 684
 685         LASSERT(inode != NULL);
 686
 687         ll_d_hlist_for_each(tmp, &inode->i_dentry)
 688                 dentry_count++;
 689
 690         CERROR("%s: inode %p dump: dev=%s fid="DFID
 691                " mode=%o count=%u, %d dentries\n",
 692                ll_get_fsname(inode->i_sb, NULL, 0), inode,
 693                ll_i2mdexp(inode)->exp_obd->obd_name, PFID(ll_inode2fid(inode)),
 694                inode->i_mode, atomic_read(&inode->i_count), dentry_count);
 695 }
 696
 697 void lustre_dump_dentry(struct dentry *dentry, int recur)
 698 {
 699         struct list_head *tmp;
 700         int subdirs = 0;
 701
 702         LASSERT(dentry != NULL);
 703
 704         list_for_each(tmp, &dentry->d_subdirs)
 705                 subdirs++;
 706
 707         CERROR("dentry %p dump: name=%.*s parent=%.*s (%p), inode=%p, count=%u,"
 708                " flags=0x%x, fsdata=%p, %d subdirs\n", dentry,
 709                dentry->d_name.len, dentry->d_name.name,
 710                dentry->d_parent->d_name.len, dentry->d_parent->d_name.name,
 711                dentry->d_parent, dentry->d_inode, d_count(dentry),
 712                dentry->d_flags, dentry->d_fsdata, subdirs);
 713         if (dentry->d_inode != NULL)
 714                 ll_dump_inode(dentry->d_inode);
 715
 716         if (recur == 0)
 717                 return;
 718
 719         list_for_each(tmp, &dentry->d_subdirs) {
 720                 struct dentry *d = list_entry(tmp, struct dentry, d_u.d_child);
 721                 lustre_dump_dentry(d, recur - 1);
 722         }
 723 }
 724
 725 static void client_common_put_super(struct super_block *sb)
 726 {
 727         struct ll_sb_info *sbi = ll_s2sbi(sb);
 728         ENTRY;
 729
 730 #ifdef CONFIG_FS_POSIX_ACL
 731         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
 732                 et_fini(&sbi->ll_et);
 733                 rct_fini(&sbi->ll_rct);
 734         }
 735 #endif
 736
 737         ll_close_thread_shutdown(sbi->ll_lcq);
 738
 739         cl_sb_fini(sb);
 740
 741         cfs_list_del(&sbi->ll_conn_chain);
 742
 743         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 744         obd_disconnect(sbi->ll_dt_exp);
 745         sbi->ll_dt_exp = NULL;
 746         /* wait till all OSCs are gone, since cl_cache is accessing sbi.
 747          * see LU-2543. */
 748         obd_zombie_barrier();
 749
 750         lprocfs_unregister_mountpoint(sbi);
 751
 752         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 753         obd_disconnect(sbi->ll_md_exp);
 754         sbi->ll_md_exp = NULL;
 755
 756         EXIT;
 757 }
 758
 759 void ll_kill_super(struct super_block *sb)
 760 {
 761         struct ll_sb_info *sbi;
 762
 763         ENTRY;
 764
 765         /* not init sb ?*/
 766         if (!(sb->s_flags & MS_ACTIVE))
 767                 return;
 768
 769         sbi = ll_s2sbi(sb);
 770         /* we need restore s_dev from changed for clustred NFS before put_super
 771          * because new kernels have cached s_dev and change sb->s_dev in
 772          * put_super not affected real removing devices */
 773         if (sbi) {
 774                 sb->s_dev = sbi->ll_sdev_orig;
 775                 sbi->ll_umounting = 1;
 776         }
 777         EXIT;
 778 }
 779
 780 static inline int ll_set_opt(const char *opt, char *data, int fl)
 781 {
 782         if (strncmp(opt, data, strlen(opt)) != 0)
 783                 return(0);
 784         else
 785                 return(fl);
 786 }
 787
 788 /* non-client-specific mount options are parsed in lmd_parse */
 789 static int ll_options(char *options, int *flags)
 790 {
 791         int tmp;
 792         char *s1 = options, *s2;
 793         ENTRY;
 794
 795         if (!options)
 796                 RETURN(0);
 797
 798         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 799
 800         while (*s1) {
 801                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 802                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 803                 if (tmp) {
 804                         *flags |= tmp;
 805                         goto next;
 806                 }
 807                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 808                 if (tmp) {
 809                         *flags |= tmp;
 810                         goto next;
 811                 }
 812                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 813                 if (tmp) {
 814                         *flags |= tmp;
 815                         goto next;
 816                 }
 817                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 818                 if (tmp) {
 819                         *flags &= ~tmp;
 820                         goto next;
 821                 }
 822                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 823                 if (tmp) {
 824                         *flags |= tmp;
 825                         goto next;
 826                 }
 827                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 828                 if (tmp) {
 829                         *flags &= ~tmp;
 830                         goto next;
 831                 }
 832 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 6, 51, 0)
 833                 tmp = ll_set_opt("acl", s1, LL_SBI_ACL);
 834                 if (tmp) {
 835                         /* Ignore deprecated mount option.  The client will
 836                          * always try to mount with ACL support, whether this
 837                          * is used depends on whether server supports it. */
 838                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 839                                                   "mount option 'acl'.\n");
 840                         goto next;
 841                 }
 842                 tmp = ll_set_opt("noacl", s1, LL_SBI_ACL);
 843                 if (tmp) {
 844                         LCONSOLE_ERROR_MSG(0x152, "Ignoring deprecated "
 845                                                   "mount option 'noacl'.\n");
 846                         goto next;
 847                 }
 848 #else
 849 #warning "{no}acl options have been deprecated since 1.8, please remove them"
 850 #endif
 851                 tmp = ll_set_opt("remote_client", s1, LL_SBI_RMT_CLIENT);
 852                 if (tmp) {
 853                         *flags |= tmp;
 854                         goto next;
 855                 }
 856                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 857                 if (tmp) {
 858                         *flags |= tmp;
 859                         goto next;
 860                 }
 861                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 862                 if (tmp) {
 863                         *flags &= ~tmp;
 864                         goto next;
 865                 }
 866
 867                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 878                 if (tmp) {
 879                         *flags |= tmp;
 880                         goto next;
 881                 }
 882                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 883                 if (tmp) {
 884                         *flags &= ~tmp;
 885                         goto next;
 886                 }
 887                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 888                 if (tmp) {
 889                         *flags |= tmp;
 890                         goto next;
 891                 }
 892                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 893                 if (tmp) {
 894                         *flags &= ~tmp;
 895                         goto next;
 896                 }
 897                 tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
 898                 if (tmp) {
 899                         *flags |= tmp;
 900                         goto next;
 901                 }
 902                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 903                 if (tmp) {
 904                         *flags |= tmp;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 908                 if (tmp) {
 909                         *flags |= tmp;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 913                 if (tmp) {
 914                         *flags &= ~tmp;
 915                         goto next;
 916                 }
 917                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 918                                    s1);
 919                 RETURN(-EINVAL);
 920
 921 next:
 922                 /* Find next opt */
 923                 s2 = strchr(s1, ',');
 924                 if (s2 == NULL)
 925                         break;
 926                 s1 = s2 + 1;
 927         }
 928         RETURN(0);
 929 }
 930
 931 void ll_lli_init(struct ll_inode_info *lli)
 932 {
 933         lli->lli_inode_magic = LLI_INODE_MAGIC;
 934         lli->lli_flags = 0;
 935         lli->lli_ioepoch = 0;
 936         lli->lli_maxbytes = MAX_LFS_FILESIZE;
 937         spin_lock_init(&lli->lli_lock);
 938         lli->lli_posix_acl = NULL;
 939         lli->lli_remote_perms = NULL;
 940         mutex_init(&lli->lli_rmtperm_mutex);
 941         /* Do not set lli_fid, it has been initialized already. */
 942         fid_zero(&lli->lli_pfid);
 943         INIT_LIST_HEAD(&lli->lli_close_list);
 944         INIT_LIST_HEAD(&lli->lli_oss_capas);
 945         atomic_set(&lli->lli_open_count, 0);
 946         lli->lli_mds_capa = NULL;
 947         lli->lli_rmtperm_time = 0;
 948         lli->lli_pending_och = NULL;
 949         lli->lli_mds_read_och = NULL;
 950         lli->lli_mds_write_och = NULL;
 951         lli->lli_mds_exec_och = NULL;
 952         lli->lli_open_fd_read_count = 0;
 953         lli->lli_open_fd_write_count = 0;
 954         lli->lli_open_fd_exec_count = 0;
 955         mutex_init(&lli->lli_och_mutex);
 956         spin_lock_init(&lli->lli_agl_lock);
 957         lli->lli_has_smd = false;
 958         spin_lock_init(&lli->lli_layout_lock);
 959         ll_layout_version_set(lli, LL_LAYOUT_GEN_NONE);
 960         lli->lli_clob = NULL;
 961
 962         init_rwsem(&lli->lli_xattrs_list_rwsem);
 963         mutex_init(&lli->lli_xattrs_enq_lock);
 964
 965         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 966         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 967                 mutex_init(&lli->lli_readdir_mutex);
 968                 lli->lli_opendir_key = NULL;
 969                 lli->lli_sai = NULL;
 970                 spin_lock_init(&lli->lli_sa_lock);
 971                 lli->lli_opendir_pid = 0;
 972         } else {
 973                 mutex_init(&lli->lli_size_mutex);
 974                 lli->lli_symlink_name = NULL;
 975                 init_rwsem(&lli->lli_trunc_sem);
 976                 mutex_init(&lli->lli_write_mutex);
 977                 init_rwsem(&lli->lli_glimpse_sem);
 978                 lli->lli_glimpse_time = 0;
 979                 CFS_INIT_LIST_HEAD(&lli->lli_agl_list);
 980                 lli->lli_agl_index = 0;
 981                 lli->lli_async_rc = 0;
 982         }
 983         mutex_init(&lli->lli_layout_mutex);
 984 }
 985
 986 static inline int ll_bdi_register(struct backing_dev_info *bdi)
 987 {
 988         static atomic_t ll_bdi_num = ATOMIC_INIT(0);
 989
 990         bdi->name = "lustre";
 991         return bdi_register(bdi, NULL, "lustre-%d",
 992                             atomic_inc_return(&ll_bdi_num));
 993 }
 994
 995 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
 996 {
 997         struct lustre_profile *lprof = NULL;
 998         struct lustre_sb_info *lsi = s2lsi(sb);
 999         struct ll_sb_info *sbi;
1000         char  *dt = NULL, *md = NULL;
1001         char  *profilenm = get_profile_name(sb);
1002         struct config_llog_instance *cfg;
1003         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1004         const int instlen = sizeof(cfg->cfg_instance) * 2 + 2;
1005         int    err;
1006         ENTRY;
1007
1008         CDEBUG(D_VFSTRACE, "VFS Op: sb %p\n", sb);
1009
1010         OBD_ALLOC_PTR(cfg);
1011         if (cfg == NULL)
1012                 RETURN(-ENOMEM);
1013
1014         try_module_get(THIS_MODULE);
1015
1016         /* client additional sb info */
1017         lsi->lsi_llsbi = sbi = ll_init_sbi();
1018         if (!sbi) {
1019                 module_put(THIS_MODULE);
1020                 OBD_FREE_PTR(cfg);
1021                 RETURN(-ENOMEM);
1022         }
1023
1024         err = ll_options(lsi->lsi_lmd->lmd_opts, &sbi->ll_flags);
1025         if (err)
1026                 GOTO(out_free, err);
1027
1028         err = bdi_init(&lsi->lsi_bdi);
1029         if (err)
1030                 GOTO(out_free, err);
1031         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1032         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1033         err = ll_bdi_register(&lsi->lsi_bdi);
1034         if (err)
1035                 GOTO(out_free, err);
1036
1037         sb->s_bdi = &lsi->lsi_bdi;
1038 #ifndef HAVE_DCACHE_LOCK
1039         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1040         sb->s_d_op = &ll_d_ops;
1041 #endif
1042
1043         /* Generate a string unique to this super, in case some joker tries
1044            to mount the same fs at two mount points.
1045            Use the address of the super itself.*/
1046         cfg->cfg_instance = sb;
1047         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1048         cfg->cfg_callback = class_config_llog_handler;
1049         /* set up client obds */
1050         err = lustre_process_log(sb, profilenm, cfg);
1051         if (err < 0)
1052                 GOTO(out_free, err);
1053
1054         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1055         lprof = class_get_profile(profilenm);
1056         if (lprof == NULL) {
1057                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1058                                    " read from the MGS.  Does that filesystem "
1059                                    "exist?\n", profilenm);
1060                 GOTO(out_free, err = -EINVAL);
1061         }
1062         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1063                lprof->lp_md, lprof->lp_dt);
1064
1065         OBD_ALLOC(dt, strlen(lprof->lp_dt) + instlen + 2);
1066         if (!dt)
1067                 GOTO(out_free, err = -ENOMEM);
1068         sprintf(dt, "%s-%p", lprof->lp_dt, cfg->cfg_instance);
1069
1070         OBD_ALLOC(md, strlen(lprof->lp_md) + instlen + 2);
1071         if (!md)
1072                 GOTO(out_free, err = -ENOMEM);
1073         sprintf(md, "%s-%p", lprof->lp_md, cfg->cfg_instance);
1074
1075         /* connections, registrations, sb setup */
1076         err = client_common_fill_super(sb, md, dt, mnt);
1077
1078 out_free:
1079         if (md)
1080                 OBD_FREE(md, strlen(lprof->lp_md) + instlen + 2);
1081         if (dt)
1082                 OBD_FREE(dt, strlen(lprof->lp_dt) + instlen + 2);
1083         if (err)
1084                 ll_put_super(sb);
1085         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1086                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1087
1088         OBD_FREE_PTR(cfg);
1089         RETURN(err);
1090 } /* ll_fill_super */
1091
1092 void ll_put_super(struct super_block *sb)
1093 {
1094         struct config_llog_instance cfg, params_cfg;
1095         struct obd_device *obd;
1096         struct lustre_sb_info *lsi = s2lsi(sb);
1097         struct ll_sb_info *sbi = ll_s2sbi(sb);
1098         char *profilenm = get_profile_name(sb);
1099         int ccc_count, next, force = 1, rc = 0;
1100         ENTRY;
1101
1102         CDEBUG(D_VFSTRACE, "VFS Op: sb %p - %s\n", sb, profilenm);
1103
1104         ll_print_capa_stat(sbi);
1105
1106         cfg.cfg_instance = sb;
1107         lustre_end_log(sb, profilenm, &cfg);
1108
1109         params_cfg.cfg_instance = sb;
1110         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1111
1112         if (sbi->ll_md_exp) {
1113                 obd = class_exp2obd(sbi->ll_md_exp);
1114                 if (obd)
1115                         force = obd->obd_force;
1116         }
1117
1118         /* Wait for unstable pages to be committed to stable storage */
1119         if (force == 0) {
1120                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1121                 rc = l_wait_event(sbi->ll_cache.ccc_unstable_waitq,
1122                         atomic_read(&sbi->ll_cache.ccc_unstable_nr) == 0,
1123                         &lwi);
1124         }
1125
1126         ccc_count = atomic_read(&sbi->ll_cache.ccc_unstable_nr);
1127         if (force == 0 && rc != -EINTR)
1128                 LASSERTF(ccc_count == 0, "count: %i\n", ccc_count);
1129
1130
1131         /* We need to set force before the lov_disconnect in
1132            lustre_common_put_super, since l_d cleans up osc's as well. */
1133         if (force) {
1134                 next = 0;
1135                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1136                                                      &next)) != NULL) {
1137                         obd->obd_force = force;
1138                 }
1139         }
1140
1141         if (sbi->ll_lcq) {
1142                 /* Only if client_common_fill_super succeeded */
1143                 client_common_put_super(sb);
1144         }
1145
1146         next = 0;
1147         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1148                 class_manual_cleanup(obd);
1149         }
1150
1151         if (sbi->ll_flags & LL_SBI_VERBOSE)
1152                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1153
1154         if (profilenm)
1155                 class_del_profile(profilenm);
1156
1157         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1158                 bdi_destroy(&lsi->lsi_bdi);
1159                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1160         }
1161
1162         ll_free_sbi(sb);
1163         lsi->lsi_llsbi = NULL;
1164
1165         lustre_common_put_super(sb);
1166
1167         cl_env_cache_purge(~0);
1168
1169         module_put(THIS_MODULE);
1170
1171         EXIT;
1172 } /* client_put_super */
1173
1174 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1175 {
1176         struct inode *inode = NULL;
1177
1178         /* NOTE: we depend on atomic igrab() -bzzz */
1179         lock_res_and_lock(lock);
1180         if (lock->l_resource->lr_lvb_inode) {
1181                 struct ll_inode_info * lli;
1182                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1183                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1184                         inode = igrab(lock->l_resource->lr_lvb_inode);
1185                 } else {
1186                         inode = lock->l_resource->lr_lvb_inode;
1187                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1188                                          D_WARNING, lock, "lr_lvb_inode %p is "
1189                                          "bogus: magic %08x",
1190                                          lock->l_resource->lr_lvb_inode,
1191                                          lli->lli_inode_magic);
1192                         inode = NULL;
1193                 }
1194         }
1195         unlock_res_and_lock(lock);
1196         return inode;
1197 }
1198
1199 static void ll_dir_clear_lsm_md(struct inode *inode)
1200 {
1201         struct ll_inode_info *lli = ll_i2info(inode);
1202
1203         LASSERT(S_ISDIR(inode->i_mode));
1204
1205         if (lli->lli_lsm_md != NULL) {
1206                 lmv_free_memmd(lli->lli_lsm_md);
1207                 lli->lli_lsm_md = NULL;
1208         }
1209 }
1210
1211 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1212                                       const struct lu_fid *fid,
1213                                       struct lustre_md *md)
1214 {
1215         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1216         struct mdt_body         *body = md->body;
1217         struct inode            *inode;
1218         ino_t                   ino;
1219         ENTRY;
1220
1221         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1222         inode = iget_locked(sb, ino);
1223         if (inode == NULL) {
1224                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1225                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1226                 RETURN(ERR_PTR(-ENOENT));
1227         }
1228
1229         if (inode->i_state & I_NEW) {
1230                 struct ll_inode_info *lli = ll_i2info(inode);
1231                 struct lmv_stripe_md *lsm = md->lmv;
1232
1233                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1234                                 (body->mode & S_IFMT);
1235                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1236                          PFID(fid));
1237
1238                 LTIME_S(inode->i_mtime) = 0;
1239                 LTIME_S(inode->i_atime) = 0;
1240                 LTIME_S(inode->i_ctime) = 0;
1241                 inode->i_rdev = 0;
1242
1243                 /* initializing backing dev info. */
1244                 inode->i_mapping->backing_dev_info =
1245                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1246                 inode->i_op = &ll_dir_inode_operations;
1247                 inode->i_fop = &ll_dir_operations;
1248                 lli->lli_fid = *fid;
1249                 ll_lli_init(lli);
1250
1251                 LASSERT(lsm != NULL);
1252                 /* master stripe FID */
1253                 lli->lli_pfid = lsm->lsm_md_oinfo[0].lmo_fid;
1254                 CDEBUG(D_INODE, "lli %p master "DFID" slave "DFID"\n",
1255                        lli, PFID(fid), PFID(&lli->lli_pfid));
1256                 unlock_new_inode(inode);
1257         }
1258
1259         RETURN(inode);
1260 }
1261
1262 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1263 {
1264         struct lu_fid *fid;
1265         struct lmv_stripe_md *lsm = md->lmv;
1266         int i;
1267
1268         LASSERT(lsm != NULL);
1269         /* XXX sigh, this lsm_root initialization should be in
1270          * LMV layer, but it needs ll_iget right now, so we
1271          * put this here right now. */
1272         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1273                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1274                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1275                 if (i == 0) {
1276                         lsm->lsm_md_oinfo[i].lmo_root = inode;
1277                 } else {
1278                         /* Unfortunately ll_iget will call ll_update_inode,
1279                          * where the initialization of slave inode is slightly
1280                          * different, so it reset lsm_md to NULL to avoid
1281                          * initializing lsm for slave inode. */
1282                         lsm->lsm_md_oinfo[i].lmo_root =
1283                                         ll_iget_anon_dir(inode->i_sb, fid, md);
1284                         if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1285                                 int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1286
1287                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1288                                 return rc;
1289                         }
1290                 }
1291         }
1292
1293         /* Here is where the lsm is being initialized(fill lmo_info) after
1294          * client retrieve MD stripe information from MDT. */
1295         return md_update_lsm_md(ll_i2mdexp(inode), lsm, md->body,
1296                                 ll_md_blocking_ast);
1297 }
1298
1299 static inline int lli_lsm_md_eq(const struct lmv_stripe_md *lsm_md1,
1300                                 const struct lmv_stripe_md *lsm_md2)
1301 {
1302         return lsm_md1->lsm_md_magic == lsm_md2->lsm_md_magic &&
1303                lsm_md1->lsm_md_stripe_count == lsm_md2->lsm_md_stripe_count &&
1304                lsm_md1->lsm_md_master_mdt_index ==
1305                                         lsm_md2->lsm_md_master_mdt_index &&
1306                lsm_md1->lsm_md_hash_type == lsm_md2->lsm_md_hash_type &&
1307                lsm_md1->lsm_md_layout_version ==
1308                                         lsm_md2->lsm_md_layout_version &&
1309                strcmp(lsm_md1->lsm_md_pool_name,
1310                       lsm_md2->lsm_md_pool_name) == 0;
1311 }
1312
1313 static void ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1314 {
1315         struct ll_inode_info *lli = ll_i2info(inode);
1316         struct lmv_stripe_md *lsm = md->lmv;
1317         int idx;
1318         ENTRY;
1319
1320         LASSERT(S_ISDIR(inode->i_mode));
1321         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1322                PFID(ll_inode2fid(inode)));
1323
1324         /* no striped information from request. */
1325         if (lsm == NULL) {
1326                 if (lli->lli_lsm_md == NULL) {
1327                         RETURN_EXIT;
1328                 } else if (lli->lli_lsm_md->lsm_md_magic == LMV_MAGIC_MIGRATE) {
1329                         /* migration is done, the temporay MIGRATE layout has
1330                          * been removed */
1331                         CDEBUG(D_INODE, DFID" finish migration.\n",
1332                                PFID(ll_inode2fid(inode)));
1333                         lmv_free_memmd(lli->lli_lsm_md);
1334                         lli->lli_lsm_md = NULL;
1335                         RETURN_EXIT;
1336                 } else {
1337                         /* The lustre_md from req does not include stripeEA,
1338                          * see ll_md_setattr */
1339                         RETURN_EXIT;
1340                 }
1341         }
1342
1343         /* set the directory layout */
1344         if (lli->lli_lsm_md == NULL) {
1345                 int rc;
1346
1347                 rc = ll_init_lsm_md(inode, md);
1348                 if (rc != 0) {
1349                         CERROR("%s: init "DFID" failed: rc = %d\n",
1350                                ll_get_fsname(inode->i_sb, NULL, 0),
1351                                PFID(&lli->lli_fid), rc);
1352                         return;
1353                 }
1354                 lli->lli_lsm_md = lsm;
1355                 /* set lsm_md to NULL, so the following free lustre_md
1356                  * will not free this lsm */
1357                 md->lmv = NULL;
1358                 CDEBUG(D_INODE, "Set lsm %p magic %x to "DFID"\n", lsm,
1359                        lsm->lsm_md_magic, PFID(ll_inode2fid(inode)));
1360                 RETURN_EXIT;
1361         }
1362
1363         /* Compare the old and new stripe information */
1364         if (!lli_lsm_md_eq(lli->lli_lsm_md, lsm)) {
1365                 CERROR("inode %p %lu mismatch\n"
1366                        "    new(%p)     vs     lli_lsm_md(%p):\n"
1367                        "    magic:      %x                   %x\n"
1368                        "    count:      %x                   %x\n"
1369                        "    master:     %x                   %x\n"
1370                        "    hash_type:  %x                   %x\n"
1371                        "    layout:     %x                   %x\n"
1372                        "    pool:       %s                   %s\n",
1373                        inode, inode->i_ino, lsm, lli->lli_lsm_md,
1374                        lsm->lsm_md_magic, lli->lli_lsm_md->lsm_md_magic,
1375                        lsm->lsm_md_stripe_count,
1376                        lli->lli_lsm_md->lsm_md_stripe_count,
1377                        lsm->lsm_md_master_mdt_index,
1378                        lli->lli_lsm_md->lsm_md_master_mdt_index,
1379                        lsm->lsm_md_hash_type, lli->lli_lsm_md->lsm_md_hash_type,
1380                        lsm->lsm_md_layout_version,
1381                        lli->lli_lsm_md->lsm_md_layout_version,
1382                        lsm->lsm_md_pool_name,
1383                        lli->lli_lsm_md->lsm_md_pool_name);
1384                 return;
1385         }
1386
1387         for (idx = 0; idx < lli->lli_lsm_md->lsm_md_stripe_count; idx++) {
1388                 if (!lu_fid_eq(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid,
1389                                &lsm->lsm_md_oinfo[idx].lmo_fid)) {
1390                         CERROR("%s: FID in lsm mismatch idx %d, old: "DFID
1391                                "new:"DFID"\n",
1392                                ll_get_fsname(inode->i_sb, NULL, 0), idx,
1393                              PFID(&lli->lli_lsm_md->lsm_md_oinfo[idx].lmo_fid),
1394                                PFID(&lsm->lsm_md_oinfo[idx].lmo_fid));
1395                         return;
1396                 }
1397         }
1398
1399         md_update_lsm_md(ll_i2mdexp(inode), ll_i2info(inode)->lli_lsm_md,
1400                          md->body, ll_md_blocking_ast);
1401
1402         RETURN_EXIT;
1403 }
1404
1405 void ll_clear_inode(struct inode *inode)
1406 {
1407         struct ll_inode_info *lli = ll_i2info(inode);
1408         struct ll_sb_info *sbi = ll_i2sbi(inode);
1409         ENTRY;
1410
1411         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1412                PFID(ll_inode2fid(inode)), inode);
1413
1414         if (S_ISDIR(inode->i_mode)) {
1415                 /* these should have been cleared in ll_file_release */
1416                 LASSERT(lli->lli_opendir_key == NULL);
1417                 LASSERT(lli->lli_sai == NULL);
1418                 LASSERT(lli->lli_opendir_pid == 0);
1419         }
1420
1421         spin_lock(&lli->lli_lock);
1422         ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
1423         spin_unlock(&lli->lli_lock);
1424         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1425
1426         LASSERT(!lli->lli_open_fd_write_count);
1427         LASSERT(!lli->lli_open_fd_read_count);
1428         LASSERT(!lli->lli_open_fd_exec_count);
1429
1430         if (lli->lli_mds_write_och)
1431                 ll_md_real_close(inode, FMODE_WRITE);
1432         if (lli->lli_mds_exec_och)
1433                 ll_md_real_close(inode, FMODE_EXEC);
1434         if (lli->lli_mds_read_och)
1435                 ll_md_real_close(inode, FMODE_READ);
1436
1437         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1438                 OBD_FREE(lli->lli_symlink_name,
1439                          strlen(lli->lli_symlink_name) + 1);
1440                 lli->lli_symlink_name = NULL;
1441         }
1442
1443         ll_xattr_cache_destroy(inode);
1444
1445         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1446                 LASSERT(lli->lli_posix_acl == NULL);
1447                 if (lli->lli_remote_perms) {
1448                         free_rmtperm_hash(lli->lli_remote_perms);
1449                         lli->lli_remote_perms = NULL;
1450                 }
1451         }
1452 #ifdef CONFIG_FS_POSIX_ACL
1453         else if (lli->lli_posix_acl) {
1454                 LASSERT(atomic_read(&lli->lli_posix_acl->a_refcount) == 1);
1455                 LASSERT(lli->lli_remote_perms == NULL);
1456                 posix_acl_release(lli->lli_posix_acl);
1457                 lli->lli_posix_acl = NULL;
1458         }
1459 #endif
1460         lli->lli_inode_magic = LLI_INODE_DEAD;
1461
1462         ll_clear_inode_capas(inode);
1463         if (S_ISDIR(inode->i_mode))
1464                 ll_dir_clear_lsm_md(inode);
1465         else
1466                 LASSERT(list_empty(&lli->lli_agl_list));
1467
1468         /*
1469          * XXX This has to be done before lsm is freed below, because
1470          * cl_object still uses inode lsm.
1471          */
1472         cl_inode_fini(inode);
1473         lli->lli_has_smd = false;
1474
1475         EXIT;
1476 }
1477
1478 int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data,
1479                   struct md_open_data **mod)
1480 {
1481         struct lustre_md md;
1482         struct inode *inode = dentry->d_inode;
1483         struct ll_sb_info *sbi = ll_i2sbi(inode);
1484         struct ptlrpc_request *request = NULL;
1485         int rc, ia_valid;
1486         ENTRY;
1487
1488         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1489                                      LUSTRE_OPC_ANY, NULL);
1490         if (IS_ERR(op_data))
1491                 RETURN(PTR_ERR(op_data));
1492
1493         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, NULL, 0,
1494                         &request, mod);
1495         if (rc) {
1496                 ptlrpc_req_finished(request);
1497                 if (rc == -ENOENT) {
1498                         clear_nlink(inode);
1499                         /* Unlinked special device node? Or just a race?
1500                          * Pretend we done everything. */
1501                         if (!S_ISREG(inode->i_mode) &&
1502                             !S_ISDIR(inode->i_mode)) {
1503                                 ia_valid = op_data->op_attr.ia_valid;
1504                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1505                                 rc = simple_setattr(dentry, &op_data->op_attr);
1506                                 op_data->op_attr.ia_valid = ia_valid;
1507                         }
1508                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1509                         CERROR("md_setattr fails: rc = %d\n", rc);
1510                 }
1511                 RETURN(rc);
1512         }
1513
1514         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1515                               sbi->ll_md_exp, &md);
1516         if (rc) {
1517                 ptlrpc_req_finished(request);
1518                 RETURN(rc);
1519         }
1520
1521         ia_valid = op_data->op_attr.ia_valid;
1522         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1523          * cache is not cleared yet. */
1524         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1525         rc = simple_setattr(dentry, &op_data->op_attr);
1526         op_data->op_attr.ia_valid = ia_valid;
1527
1528         /* Extract epoch data if obtained. */
1529         op_data->op_handle = md.body->handle;
1530         op_data->op_ioepoch = md.body->ioepoch;
1531
1532         ll_update_inode(inode, &md);
1533         ptlrpc_req_finished(request);
1534
1535         RETURN(rc);
1536 }
1537
1538 /* Close IO epoch and send Size-on-MDS attribute update. */
1539 static int ll_setattr_done_writing(struct inode *inode,
1540                                    struct md_op_data *op_data,
1541                                    struct md_open_data *mod)
1542 {
1543         struct ll_inode_info *lli = ll_i2info(inode);
1544         int rc = 0;
1545         ENTRY;
1546
1547         LASSERT(op_data != NULL);
1548         if (!S_ISREG(inode->i_mode))
1549                 RETURN(0);
1550
1551         CDEBUG(D_INODE, "Epoch "LPU64" closed on "DFID" for truncate\n",
1552                op_data->op_ioepoch, PFID(&lli->lli_fid));
1553
1554         op_data->op_flags = MF_EPOCH_CLOSE;
1555         ll_done_writing_attr(inode, op_data);
1556         ll_pack_inode2opdata(inode, op_data, NULL);
1557
1558         rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
1559         if (rc == -EAGAIN) {
1560                 /* MDS has instructed us to obtain Size-on-MDS attribute
1561                  * from OSTs and send setattr to back to MDS. */
1562                 rc = ll_som_update(inode, op_data);
1563         } else if (rc) {
1564                 CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
1565                        ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
1566                        PFID(ll_inode2fid(inode)), rc);
1567         }
1568         RETURN(rc);
1569 }
1570
1571 static int ll_setattr_ost(struct inode *inode, struct iattr *attr)
1572 {
1573         struct obd_capa *capa;
1574         int rc;
1575
1576         if (attr->ia_valid & ATTR_SIZE)
1577                 capa = ll_osscapa_get(inode, CAPA_OPC_OSS_TRUNC);
1578         else
1579                 capa = ll_mdscapa_get(inode);
1580
1581         rc = cl_setattr_ost(inode, attr, capa);
1582
1583         if (attr->ia_valid & ATTR_SIZE)
1584                 ll_truncate_free_capa(capa);
1585         else
1586                 capa_put(capa);
1587
1588         return rc;
1589 }
1590
1591 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1592  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1593  * keep these values until such a time that objects are allocated for it.
1594  * We do the MDS operations first, as it is checking permissions for us.
1595  * We don't to the MDS RPC if there is nothing that we want to store there,
1596  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1597  * going to do an RPC anyways.
1598  *
1599  * If we are doing a truncate, we will send the mtime and ctime updates
1600  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1601  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1602  * at the same time.
1603  *
1604  * In case of HSMimport, we only set attr on MDS.
1605  */
1606 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr, bool hsm_import)
1607 {
1608         struct inode *inode = dentry->d_inode;
1609         struct ll_inode_info *lli = ll_i2info(inode);
1610         struct md_op_data *op_data = NULL;
1611         struct md_open_data *mod = NULL;
1612         bool file_is_released = false;
1613         int rc = 0, rc1 = 0;
1614         ENTRY;
1615
1616         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1617                "valid %x, hsm_import %d\n",
1618                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1619                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1620                hsm_import);
1621
1622         if (attr->ia_valid & ATTR_SIZE) {
1623                 /* Check new size against VFS/VM file size limit and rlimit */
1624                 rc = inode_newsize_ok(inode, attr->ia_size);
1625                 if (rc)
1626                         RETURN(rc);
1627
1628                 /* The maximum Lustre file size is variable, based on the
1629                  * OST maximum object size and number of stripes.  This
1630                  * needs another check in addition to the VFS check above. */
1631                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1632                         CDEBUG(D_INODE,"file "DFID" too large %llu > "LPU64"\n",
1633                                PFID(&lli->lli_fid), attr->ia_size,
1634                                ll_file_maxbytes(inode));
1635                         RETURN(-EFBIG);
1636                 }
1637
1638                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1639         }
1640
1641         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1642         if (attr->ia_valid & TIMES_SET_FLAGS) {
1643                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1644                     !cfs_capable(CFS_CAP_FOWNER))
1645                         RETURN(-EPERM);
1646         }
1647
1648         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1649         if (attr->ia_valid & ATTR_CTIME) {
1650                 attr->ia_ctime = CFS_CURRENT_TIME;
1651                 attr->ia_valid |= ATTR_CTIME_SET;
1652         }
1653         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1654             (attr->ia_valid & ATTR_ATIME)) {
1655                 attr->ia_atime = CFS_CURRENT_TIME;
1656                 attr->ia_valid |= ATTR_ATIME_SET;
1657         }
1658         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1659             (attr->ia_valid & ATTR_MTIME)) {
1660                 attr->ia_mtime = CFS_CURRENT_TIME;
1661                 attr->ia_valid |= ATTR_MTIME_SET;
1662         }
1663
1664         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1665                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu, now = %lu\n",
1666                        LTIME_S(attr->ia_mtime), LTIME_S(attr->ia_ctime),
1667                        cfs_time_current_sec());
1668
1669         /* We always do an MDS RPC, even if we're only changing the size;
1670          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1671
1672         OBD_ALLOC_PTR(op_data);
1673         if (op_data == NULL)
1674                 RETURN(-ENOMEM);
1675
1676         if (!S_ISDIR(inode->i_mode)) {
1677                 if (attr->ia_valid & ATTR_SIZE)
1678                         inode_dio_write_done(inode);
1679                 mutex_unlock(&inode->i_mutex);
1680         }
1681
1682         /* truncate on a released file must failed with -ENODATA,
1683          * so size must not be set on MDS for released file
1684          * but other attributes must be set
1685          */
1686         if (S_ISREG(inode->i_mode)) {
1687                 struct lov_stripe_md *lsm;
1688                 __u32 gen;
1689
1690                 ll_layout_refresh(inode, &gen);
1691                 lsm = ccc_inode_lsm_get(inode);
1692                 if (lsm && lsm->lsm_pattern & LOV_PATTERN_F_RELEASED)
1693                         file_is_released = true;
1694                 ccc_inode_lsm_put(inode, lsm);
1695
1696                 if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1697                         if (file_is_released) {
1698                                 rc = ll_layout_restore(inode, 0, attr->ia_size);
1699                                 if (rc < 0)
1700                                         GOTO(out, rc);
1701
1702                                 file_is_released = false;
1703                                 ll_layout_refresh(inode, &gen);
1704                         }
1705
1706                         /* If we are changing file size, file content is
1707                          * modified, flag it. */
1708                         attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1709                         spin_lock(&lli->lli_lock);
1710                         lli->lli_flags |= LLIF_DATA_MODIFIED;
1711                         spin_unlock(&lli->lli_lock);
1712                         op_data->op_bias |= MDS_DATA_MODIFIED;
1713                 }
1714         }
1715
1716         memcpy(&op_data->op_attr, attr, sizeof(*attr));
1717
1718         /* Open epoch for truncate. */
1719         if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
1720             (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
1721                 op_data->op_flags = MF_EPOCH_OPEN;
1722
1723         rc = ll_md_setattr(dentry, op_data, &mod);
1724         if (rc)
1725                 GOTO(out, rc);
1726
1727         /* RPC to MDT is sent, cancel data modification flag */
1728         if (rc == 0 && (op_data->op_bias & MDS_DATA_MODIFIED)) {
1729                 spin_lock(&lli->lli_lock);
1730                 lli->lli_flags &= ~LLIF_DATA_MODIFIED;
1731                 spin_unlock(&lli->lli_lock);
1732         }
1733
1734         ll_ioepoch_open(lli, op_data->op_ioepoch);
1735         if (!S_ISREG(inode->i_mode) || file_is_released)
1736                 GOTO(out, rc = 0);
1737
1738         if (attr->ia_valid & (ATTR_SIZE |
1739                               ATTR_ATIME | ATTR_ATIME_SET |
1740                               ATTR_MTIME | ATTR_MTIME_SET)) {
1741                 /* For truncate and utimes sending attributes to OSTs, setting
1742                  * mtime/atime to the past will be performed under PW [0:EOF]
1743                  * extent lock (new_size:EOF for truncate).  It may seem
1744                  * excessive to send mtime/atime updates to OSTs when not
1745                  * setting times to past, but it is necessary due to possible
1746                  * time de-synchronization between MDT inode and OST objects */
1747                 if (attr->ia_valid & ATTR_SIZE)
1748                         down_write(&lli->lli_trunc_sem);
1749                 rc = ll_setattr_ost(inode, attr);
1750                 if (attr->ia_valid & ATTR_SIZE)
1751                         up_write(&lli->lli_trunc_sem);
1752         }
1753         EXIT;
1754 out:
1755         if (op_data) {
1756                 if (op_data->op_ioepoch) {
1757                         rc1 = ll_setattr_done_writing(inode, op_data, mod);
1758                         if (!rc)
1759                                 rc = rc1;
1760                 }
1761                 ll_finish_md_op_data(op_data);
1762         }
1763         if (!S_ISDIR(inode->i_mode)) {
1764                 mutex_lock(&inode->i_mutex);
1765                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1766                         inode_dio_wait(inode);
1767         }
1768
1769         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1770                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1771
1772         return rc;
1773 }
1774
1775 int ll_setattr(struct dentry *de, struct iattr *attr)
1776 {
1777         int mode = de->d_inode->i_mode;
1778
1779         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1780                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1781                 attr->ia_valid |= MDS_OPEN_OWNEROVERRIDE;
1782
1783         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1784                                (ATTR_SIZE|ATTR_MODE)) &&
1785             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1786              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1787               !(attr->ia_mode & S_ISGID))))
1788                 attr->ia_valid |= ATTR_FORCE;
1789
1790         if ((attr->ia_valid & ATTR_MODE) &&
1791             (mode & S_ISUID) &&
1792             !(attr->ia_mode & S_ISUID) &&
1793             !(attr->ia_valid & ATTR_KILL_SUID))
1794                 attr->ia_valid |= ATTR_KILL_SUID;
1795
1796         if ((attr->ia_valid & ATTR_MODE) &&
1797             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1798             !(attr->ia_mode & S_ISGID) &&
1799             !(attr->ia_valid & ATTR_KILL_SGID))
1800                 attr->ia_valid |= ATTR_KILL_SGID;
1801
1802         return ll_setattr_raw(de, attr, false);
1803 }
1804
1805 int ll_statfs_internal(struct super_block *sb, struct obd_statfs *osfs,
1806                        __u64 max_age, __u32 flags)
1807 {
1808         struct ll_sb_info *sbi = ll_s2sbi(sb);
1809         struct obd_statfs obd_osfs;
1810         int rc;
1811         ENTRY;
1812
1813         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1814         if (rc) {
1815                 CERROR("md_statfs fails: rc = %d\n", rc);
1816                 RETURN(rc);
1817         }
1818
1819         osfs->os_type = sb->s_magic;
1820
1821         CDEBUG(D_SUPER, "MDC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1822                osfs->os_bavail, osfs->os_blocks, osfs->os_ffree,osfs->os_files);
1823
1824         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1825                 flags |= OBD_STATFS_NODELAY;
1826
1827         rc = obd_statfs_rqset(sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1828         if (rc) {
1829                 CERROR("obd_statfs fails: rc = %d\n", rc);
1830                 RETURN(rc);
1831         }
1832
1833         CDEBUG(D_SUPER, "OSC blocks "LPU64"/"LPU64" objects "LPU64"/"LPU64"\n",
1834                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1835                obd_osfs.os_files);
1836
1837         osfs->os_bsize = obd_osfs.os_bsize;
1838         osfs->os_blocks = obd_osfs.os_blocks;
1839         osfs->os_bfree = obd_osfs.os_bfree;
1840         osfs->os_bavail = obd_osfs.os_bavail;
1841
1842         /* If we don't have as many objects free on the OST as inodes
1843          * on the MDS, we reduce the total number of inodes to
1844          * compensate, so that the "inodes in use" number is correct.
1845          */
1846         if (obd_osfs.os_ffree < osfs->os_ffree) {
1847                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1848                         obd_osfs.os_ffree;
1849                 osfs->os_ffree = obd_osfs.os_ffree;
1850         }
1851
1852         RETURN(rc);
1853 }
1854 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1855 {
1856         struct super_block *sb = de->d_sb;
1857         struct obd_statfs osfs;
1858         __u64 fsid = huge_encode_dev(sb->s_dev);
1859         int rc;
1860
1861         CDEBUG(D_VFSTRACE, "VFS Op: at "LPU64" jiffies\n", get_jiffies_64());
1862         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1863
1864         /* Some amount of caching on the client is allowed */
1865         rc = ll_statfs_internal(sb, &osfs,
1866                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
1867                                 0);
1868         if (rc)
1869                 return rc;
1870
1871         statfs_unpack(sfs, &osfs);
1872
1873         /* We need to downshift for all 32-bit kernels, because we can't
1874          * tell if the kernel is being called via sys_statfs64() or not.
1875          * Stop before overflowing f_bsize - in which case it is better
1876          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1877         if (sizeof(long) < 8) {
1878                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1879                         sfs->f_bsize <<= 1;
1880
1881                         osfs.os_blocks >>= 1;
1882                         osfs.os_bfree >>= 1;
1883                         osfs.os_bavail >>= 1;
1884                 }
1885         }
1886
1887         sfs->f_blocks = osfs.os_blocks;
1888         sfs->f_bfree = osfs.os_bfree;
1889         sfs->f_bavail = osfs.os_bavail;
1890         sfs->f_fsid.val[0] = (__u32)fsid;
1891         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1892         return 0;
1893 }
1894
1895 void ll_inode_size_lock(struct inode *inode)
1896 {
1897         struct ll_inode_info *lli;
1898
1899         LASSERT(!S_ISDIR(inode->i_mode));
1900
1901         lli = ll_i2info(inode);
1902         mutex_lock(&lli->lli_size_mutex);
1903 }
1904
1905 void ll_inode_size_unlock(struct inode *inode)
1906 {
1907         struct ll_inode_info *lli;
1908
1909         lli = ll_i2info(inode);
1910         mutex_unlock(&lli->lli_size_mutex);
1911 }
1912
1913 void ll_update_inode(struct inode *inode, struct lustre_md *md)
1914 {
1915         struct ll_inode_info *lli = ll_i2info(inode);
1916         struct mdt_body *body = md->body;
1917         struct lov_stripe_md *lsm = md->lsm;
1918         struct ll_sb_info *sbi = ll_i2sbi(inode);
1919
1920         LASSERT ((lsm != NULL) == ((body->valid & OBD_MD_FLEASIZE) != 0));
1921         if (lsm != NULL) {
1922                 if (!lli->lli_has_smd &&
1923                     !(sbi->ll_flags & LL_SBI_LAYOUT_LOCK))
1924                         cl_file_inode_init(inode, md);
1925
1926                 lli->lli_maxbytes = lsm->lsm_maxbytes;
1927                 if (lli->lli_maxbytes > MAX_LFS_FILESIZE)
1928                         lli->lli_maxbytes = MAX_LFS_FILESIZE;
1929         }
1930
1931         if (S_ISDIR(inode->i_mode))
1932                 ll_update_lsm_md(inode, md);
1933
1934         if (sbi->ll_flags & LL_SBI_RMT_CLIENT) {
1935                 if (body->valid & OBD_MD_FLRMTPERM)
1936                         ll_update_remote_perm(inode, md->remote_perm);
1937         }
1938 #ifdef CONFIG_FS_POSIX_ACL
1939         else if (body->valid & OBD_MD_FLACL) {
1940                 spin_lock(&lli->lli_lock);
1941                 if (lli->lli_posix_acl)
1942                         posix_acl_release(lli->lli_posix_acl);
1943                 lli->lli_posix_acl = md->posix_acl;
1944                 spin_unlock(&lli->lli_lock);
1945         }
1946 #endif
1947         inode->i_ino = cl_fid_build_ino(&body->fid1,
1948                                         sbi->ll_flags & LL_SBI_32BIT_API);
1949         inode->i_generation = cl_fid_build_gen(&body->fid1);
1950
1951         if (body->valid & OBD_MD_FLATIME) {
1952                 if (body->atime > LTIME_S(inode->i_atime))
1953                         LTIME_S(inode->i_atime) = body->atime;
1954                 lli->lli_lvb.lvb_atime = body->atime;
1955         }
1956         if (body->valid & OBD_MD_FLMTIME) {
1957                 if (body->mtime > LTIME_S(inode->i_mtime)) {
1958                         CDEBUG(D_INODE, "setting ino %lu mtime from %lu "
1959                                "to "LPU64"\n", inode->i_ino,
1960                                LTIME_S(inode->i_mtime), body->mtime);
1961                         LTIME_S(inode->i_mtime) = body->mtime;
1962                 }
1963                 lli->lli_lvb.lvb_mtime = body->mtime;
1964         }
1965         if (body->valid & OBD_MD_FLCTIME) {
1966                 if (body->ctime > LTIME_S(inode->i_ctime))
1967                         LTIME_S(inode->i_ctime) = body->ctime;
1968                 lli->lli_lvb.lvb_ctime = body->ctime;
1969         }
1970         if (body->valid & OBD_MD_FLMODE)
1971                 inode->i_mode = (inode->i_mode & S_IFMT)|(body->mode & ~S_IFMT);
1972         if (body->valid & OBD_MD_FLTYPE)
1973                 inode->i_mode = (inode->i_mode & ~S_IFMT)|(body->mode & S_IFMT);
1974         LASSERT(inode->i_mode != 0);
1975         if (S_ISREG(inode->i_mode)) {
1976                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1, LL_MAX_BLKSIZE_BITS);
1977         } else {
1978                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1979         }
1980         if (body->valid & OBD_MD_FLUID)
1981                 inode->i_uid = make_kuid(&init_user_ns, body->uid);
1982         if (body->valid & OBD_MD_FLGID)
1983                 inode->i_gid = make_kgid(&init_user_ns, body->gid);
1984         if (body->valid & OBD_MD_FLFLAGS)
1985                 inode->i_flags = ll_ext_to_inode_flags(body->flags);
1986         if (body->valid & OBD_MD_FLNLINK)
1987                 set_nlink(inode, body->nlink);
1988         if (body->valid & OBD_MD_FLRDEV)
1989                 inode->i_rdev = old_decode_dev(body->rdev);
1990
1991         if (body->valid & OBD_MD_FLID) {
1992                 /* FID shouldn't be changed! */
1993                 if (fid_is_sane(&lli->lli_fid)) {
1994                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->fid1),
1995                                  "Trying to change FID "DFID
1996                                  " to the "DFID", inode "DFID"(%p)\n",
1997                                  PFID(&lli->lli_fid), PFID(&body->fid1),
1998                                  PFID(ll_inode2fid(inode)), inode);
1999                 } else {
2000                         lli->lli_fid = body->fid1;
2001                 }
2002         }
2003
2004         LASSERT(fid_seq(&lli->lli_fid) != 0);
2005
2006         if (body->valid & OBD_MD_FLSIZE) {
2007                 if (exp_connect_som(ll_i2mdexp(inode)) &&
2008                     S_ISREG(inode->i_mode)) {
2009                         struct lustre_handle lockh;
2010                         ldlm_mode_t mode;
2011
2012                         /* As it is possible a blocking ast has been processed
2013                          * by this time, we need to check there is an UPDATE
2014                          * lock on the client and set LLIF_MDS_SIZE_LOCK holding
2015                          * it. */
2016                         mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
2017                                                &lockh, LDLM_FL_CBPENDING,
2018                                                LCK_CR | LCK_CW |
2019                                                LCK_PR | LCK_PW);
2020                         if (mode) {
2021                                 if (lli->lli_flags & (LLIF_DONE_WRITING |
2022                                                       LLIF_EPOCH_PENDING |
2023                                                       LLIF_SOM_DIRTY)) {
2024                                         CERROR("%s: inode "DFID" flags %u still"
2025                                                " has size authority! do not "
2026                                                "trust the size from MDS\n",
2027                                                sbi->ll_md_exp->exp_obd->obd_name,
2028                                                PFID(ll_inode2fid(inode)),
2029                                                lli->lli_flags);
2030                                 } else {
2031                                         /* Use old size assignment to avoid
2032                                          * deadlock bz14138 & bz14326 */
2033                                         i_size_write(inode, body->size);
2034                                         spin_lock(&lli->lli_lock);
2035                                         lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
2036                                         spin_unlock(&lli->lli_lock);
2037                                 }
2038                                 ldlm_lock_decref(&lockh, mode);
2039                         }
2040                 } else {
2041                         /* Use old size assignment to avoid
2042                          * deadlock bz14138 & bz14326 */
2043                         i_size_write(inode, body->size);
2044
2045                         CDEBUG(D_VFSTRACE,
2046                                "inode="DFID", updating i_size %llu\n",
2047                                PFID(ll_inode2fid(inode)),
2048                                (unsigned long long)body->size);
2049                 }
2050
2051                 if (body->valid & OBD_MD_FLBLOCKS)
2052                         inode->i_blocks = body->blocks;
2053         }
2054
2055         if (body->valid & OBD_MD_FLMDSCAPA) {
2056                 LASSERT(md->mds_capa);
2057                 ll_add_capa(inode, md->mds_capa);
2058         }
2059         if (body->valid & OBD_MD_FLOSSCAPA) {
2060                 LASSERT(md->oss_capa);
2061                 ll_add_capa(inode, md->oss_capa);
2062         }
2063
2064         if (body->valid & OBD_MD_TSTATE) {
2065                 if (body->t_state & MS_RESTORE)
2066                         lli->lli_flags |= LLIF_FILE_RESTORING;
2067         }
2068 }
2069
2070 void ll_read_inode2(struct inode *inode, void *opaque)
2071 {
2072         struct lustre_md *md = opaque;
2073         struct ll_inode_info *lli = ll_i2info(inode);
2074         ENTRY;
2075
2076         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2077                PFID(&lli->lli_fid), inode);
2078
2079         LASSERT(!lli->lli_has_smd);
2080
2081         /* Core attributes from the MDS first.  This is a new inode, and
2082          * the VFS doesn't zero times in the core inode so we have to do
2083          * it ourselves.  They will be overwritten by either MDS or OST
2084          * attributes - we just need to make sure they aren't newer. */
2085         LTIME_S(inode->i_mtime) = 0;
2086         LTIME_S(inode->i_atime) = 0;
2087         LTIME_S(inode->i_ctime) = 0;
2088         inode->i_rdev = 0;
2089         ll_update_inode(inode, md);
2090
2091         /* OIDEBUG(inode); */
2092
2093         /* initializing backing dev info. */
2094         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2095
2096
2097         if (S_ISREG(inode->i_mode)) {
2098                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2099                 inode->i_op = &ll_file_inode_operations;
2100                 inode->i_fop = sbi->ll_fop;
2101                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2102                 EXIT;
2103         } else if (S_ISDIR(inode->i_mode)) {
2104                 inode->i_op = &ll_dir_inode_operations;
2105                 inode->i_fop = &ll_dir_operations;
2106                 EXIT;
2107         } else if (S_ISLNK(inode->i_mode)) {
2108                 inode->i_op = &ll_fast_symlink_inode_operations;
2109                 EXIT;
2110         } else {
2111                 inode->i_op = &ll_special_inode_operations;
2112
2113                 init_special_inode(inode, inode->i_mode,
2114                                    inode->i_rdev);
2115
2116                 EXIT;
2117         }
2118 }
2119
2120 void ll_delete_inode(struct inode *inode)
2121 {
2122         struct cl_inode_info *lli = cl_i2info(inode);
2123         ENTRY;
2124
2125         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2126                 /* discard all dirty pages before truncating them, required by
2127                  * osc_extent implementation at LU-1030. */
2128                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF,
2129                                    CL_FSYNC_DISCARD, 1);
2130
2131         truncate_inode_pages(&inode->i_data, 0);
2132
2133         /* Workaround for LU-118 */
2134         if (inode->i_data.nrpages) {
2135                 spin_lock_irq(&inode->i_data.tree_lock);
2136                 spin_unlock_irq(&inode->i_data.tree_lock);
2137                 LASSERTF(inode->i_data.nrpages == 0,
2138                          "inode="DFID"(%p) nrpages=%lu, see "
2139                          "http://jira.whamcloud.com/browse/LU-118\n",
2140                          PFID(ll_inode2fid(inode)), inode,
2141                          inode->i_data.nrpages);
2142         }
2143         /* Workaround end */
2144
2145 #ifdef HAVE_SBOPS_EVICT_INODE
2146         ll_clear_inode(inode);
2147 #endif
2148         clear_inode(inode);
2149
2150         EXIT;
2151 }
2152
2153 int ll_iocontrol(struct inode *inode, struct file *file,
2154                  unsigned int cmd, unsigned long arg)
2155 {
2156         struct ll_sb_info *sbi = ll_i2sbi(inode);
2157         struct ptlrpc_request *req = NULL;
2158         int rc, flags = 0;
2159         ENTRY;
2160
2161         switch(cmd) {
2162         case FSFILT_IOC_GETFLAGS: {
2163                 struct mdt_body *body;
2164                 struct md_op_data *op_data;
2165
2166                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2167                                              0, 0, LUSTRE_OPC_ANY,
2168                                              NULL);
2169                 if (IS_ERR(op_data))
2170                         RETURN(PTR_ERR(op_data));
2171
2172                 op_data->op_valid = OBD_MD_FLFLAGS;
2173                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2174                 ll_finish_md_op_data(op_data);
2175                 if (rc) {
2176                         CERROR("%s: failure inode "DFID": rc = %d\n",
2177                                sbi->ll_md_exp->exp_obd->obd_name,
2178                                PFID(ll_inode2fid(inode)), rc);
2179                         RETURN(-abs(rc));
2180                 }
2181
2182                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2183
2184                 flags = body->flags;
2185
2186                 ptlrpc_req_finished(req);
2187
2188                 RETURN(put_user(flags, (int *)arg));
2189         }
2190         case FSFILT_IOC_SETFLAGS: {
2191                 struct lov_stripe_md *lsm;
2192                 struct obd_info oinfo = { { { 0 } } };
2193                 struct md_op_data *op_data;
2194
2195                 if (get_user(flags, (int *)arg))
2196                         RETURN(-EFAULT);
2197
2198                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2199                                              LUSTRE_OPC_ANY, NULL);
2200                 if (IS_ERR(op_data))
2201                         RETURN(PTR_ERR(op_data));
2202
2203                 ((struct ll_iattr *)&op_data->op_attr)->ia_attr_flags = flags;
2204                 op_data->op_attr.ia_valid |= ATTR_ATTR_FLAG;
2205                 rc = md_setattr(sbi->ll_md_exp, op_data,
2206                                 NULL, 0, NULL, 0, &req, NULL);
2207                 ll_finish_md_op_data(op_data);
2208                 ptlrpc_req_finished(req);
2209                 if (rc)
2210                         RETURN(rc);
2211
2212                 inode->i_flags = ll_ext_to_inode_flags(flags);
2213
2214                 lsm = ccc_inode_lsm_get(inode);
2215                 if (!lsm_has_objects(lsm)) {
2216                         ccc_inode_lsm_put(inode, lsm);
2217                         RETURN(0);
2218                 }
2219
2220                 OBDO_ALLOC(oinfo.oi_oa);
2221                 if (!oinfo.oi_oa) {
2222                         ccc_inode_lsm_put(inode, lsm);
2223                         RETURN(-ENOMEM);
2224                 }
2225                 oinfo.oi_md = lsm;
2226                 oinfo.oi_oa->o_oi = lsm->lsm_oi;
2227                 oinfo.oi_oa->o_flags = flags;
2228                 oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS |
2229                                        OBD_MD_FLGROUP;
2230                 oinfo.oi_capa = ll_mdscapa_get(inode);
2231                 obdo_set_parent_fid(oinfo.oi_oa, &ll_i2info(inode)->lli_fid);
2232                 rc = obd_setattr_rqset(sbi->ll_dt_exp, &oinfo, NULL);
2233                 capa_put(oinfo.oi_capa);
2234                 OBDO_FREE(oinfo.oi_oa);
2235                 ccc_inode_lsm_put(inode, lsm);
2236
2237                 if (rc && rc != -EPERM && rc != -EACCES)
2238                         CERROR("osc_setattr_async fails: rc = %d\n", rc);
2239
2240                 RETURN(rc);
2241         }
2242         default:
2243                 RETURN(-ENOSYS);
2244         }
2245
2246         RETURN(0);
2247 }
2248
2249 int ll_flush_ctx(struct inode *inode)
2250 {
2251         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2252
2253         CDEBUG(D_SEC, "flush context for user %d\n",
2254                from_kuid(&init_user_ns, current_uid()));
2255
2256         obd_set_info_async(NULL, sbi->ll_md_exp,
2257                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2258                            0, NULL, NULL);
2259         obd_set_info_async(NULL, sbi->ll_dt_exp,
2260                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2261                            0, NULL, NULL);
2262         return 0;
2263 }
2264
2265 /* umount -f client means force down, don't save state */
2266 void ll_umount_begin(struct super_block *sb)
2267 {
2268         struct ll_sb_info *sbi = ll_s2sbi(sb);
2269         struct obd_device *obd;
2270         struct obd_ioctl_data *ioc_data;
2271         ENTRY;
2272
2273         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2274                sb->s_count, atomic_read(&sb->s_active));
2275
2276         obd = class_exp2obd(sbi->ll_md_exp);
2277         if (obd == NULL) {
2278                 CERROR("Invalid MDC connection handle "LPX64"\n",
2279                        sbi->ll_md_exp->exp_handle.h_cookie);
2280                 EXIT;
2281                 return;
2282         }
2283         obd->obd_force = 1;
2284
2285         obd = class_exp2obd(sbi->ll_dt_exp);
2286         if (obd == NULL) {
2287                 CERROR("Invalid LOV connection handle "LPX64"\n",
2288                        sbi->ll_dt_exp->exp_handle.h_cookie);
2289                 EXIT;
2290                 return;
2291         }
2292         obd->obd_force = 1;
2293
2294         OBD_ALLOC_PTR(ioc_data);
2295         if (ioc_data) {
2296                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2297                               sizeof *ioc_data, ioc_data, NULL);
2298
2299                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2300                               sizeof *ioc_data, ioc_data, NULL);
2301
2302                 OBD_FREE_PTR(ioc_data);
2303         }
2304
2305         /* Really, we'd like to wait until there are no requests outstanding,
2306          * and then continue.  For now, we just invalidate the requests,
2307          * schedule() and sleep one second if needed, and hope.
2308          */
2309         schedule();
2310         EXIT;
2311 }
2312
2313 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2314 {
2315         struct ll_sb_info *sbi = ll_s2sbi(sb);
2316         char *profilenm = get_profile_name(sb);
2317         int err;
2318         __u32 read_only;
2319
2320         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2321                 read_only = *flags & MS_RDONLY;
2322                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2323                                          sizeof(KEY_READ_ONLY),
2324                                          KEY_READ_ONLY, sizeof(read_only),
2325                                          &read_only, NULL);
2326                 if (err) {
2327                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2328                                       profilenm, read_only ?
2329                                       "read-only" : "read-write", err);
2330                         return err;
2331                 }
2332
2333                 if (read_only)
2334                         sb->s_flags |= MS_RDONLY;
2335                 else
2336                         sb->s_flags &= ~MS_RDONLY;
2337
2338                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2339                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2340                                       read_only ?  "read-only" : "read-write");
2341         }
2342         return 0;
2343 }
2344
2345 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2346                   struct super_block *sb, struct lookup_intent *it)
2347 {
2348         struct ll_sb_info *sbi = NULL;
2349         struct lustre_md md = { 0 };
2350         int rc;
2351         ENTRY;
2352
2353         LASSERT(*inode || sb);
2354         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2355         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2356                               sbi->ll_md_exp, &md);
2357         if (rc)
2358                 RETURN(rc);
2359
2360         if (*inode) {
2361                 ll_update_inode(*inode, &md);
2362         } else {
2363                 LASSERT(sb != NULL);
2364
2365                 /*
2366                  * At this point server returns to client's same fid as client
2367                  * generated for creating. So using ->fid1 is okay here.
2368                  */
2369                 LASSERT(fid_is_sane(&md.body->fid1));
2370
2371                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->fid1,
2372                                              sbi->ll_flags & LL_SBI_32BIT_API),
2373                                  &md);
2374                 if (*inode == NULL || IS_ERR(*inode)) {
2375 #ifdef CONFIG_FS_POSIX_ACL
2376                         if (md.posix_acl) {
2377                                 posix_acl_release(md.posix_acl);
2378                                 md.posix_acl = NULL;
2379                         }
2380 #endif
2381                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2382                         *inode = NULL;
2383                         CERROR("new_inode -fatal: rc %d\n", rc);
2384                         GOTO(out, rc);
2385                 }
2386         }
2387
2388         /* Handling piggyback layout lock.
2389          * Layout lock can be piggybacked by getattr and open request.
2390          * The lsm can be applied to inode only if it comes with a layout lock
2391          * otherwise correct layout may be overwritten, for example:
2392          * 1. proc1: mdt returns a lsm but not granting layout
2393          * 2. layout was changed by another client
2394          * 3. proc2: refresh layout and layout lock granted
2395          * 4. proc1: to apply a stale layout */
2396         if (it != NULL && it->d.lustre.it_lock_mode != 0) {
2397                 struct lustre_handle lockh;
2398                 struct ldlm_lock *lock;
2399
2400                 lockh.cookie = it->d.lustre.it_lock_handle;
2401                 lock = ldlm_handle2lock(&lockh);
2402                 LASSERT(lock != NULL);
2403                 if (ldlm_has_layout(lock)) {
2404                         struct cl_object_conf conf;
2405
2406                         memset(&conf, 0, sizeof(conf));
2407                         conf.coc_opc = OBJECT_CONF_SET;
2408                         conf.coc_inode = *inode;
2409                         conf.coc_lock = lock;
2410                         conf.u.coc_md = &md;
2411                         (void)ll_layout_conf(*inode, &conf);
2412                 }
2413                 LDLM_LOCK_PUT(lock);
2414         }
2415
2416 out:
2417         if (md.lsm != NULL)
2418                 obd_free_memmd(sbi->ll_dt_exp, &md.lsm);
2419         md_free_lustre_md(sbi->ll_md_exp, &md);
2420         RETURN(rc);
2421 }
2422
2423 int ll_obd_statfs(struct inode *inode, void *arg)
2424 {
2425         struct ll_sb_info *sbi = NULL;
2426         struct obd_export *exp;
2427         char *buf = NULL;
2428         struct obd_ioctl_data *data = NULL;
2429         __u32 type;
2430         __u32 flags;
2431         int len = 0, rc;
2432
2433         if (!inode || !(sbi = ll_i2sbi(inode)))
2434                 GOTO(out_statfs, rc = -EINVAL);
2435
2436         rc = obd_ioctl_getdata(&buf, &len, arg);
2437         if (rc)
2438                 GOTO(out_statfs, rc);
2439
2440         data = (void*)buf;
2441         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2442             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2443                 GOTO(out_statfs, rc = -EINVAL);
2444
2445         if (data->ioc_inllen1 != sizeof(__u32) ||
2446             data->ioc_inllen2 != sizeof(__u32) ||
2447             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2448             data->ioc_plen2 != sizeof(struct obd_uuid))
2449                 GOTO(out_statfs, rc = -EINVAL);
2450
2451         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2452         if (type & LL_STATFS_LMV)
2453                 exp = sbi->ll_md_exp;
2454         else if (type & LL_STATFS_LOV)
2455                 exp = sbi->ll_dt_exp;
2456         else
2457                 GOTO(out_statfs, rc = -ENODEV);
2458
2459         flags = (type & LL_STATFS_NODELAY) ? OBD_STATFS_NODELAY : 0;
2460         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, &flags);
2461         if (rc)
2462                 GOTO(out_statfs, rc);
2463 out_statfs:
2464         if (buf)
2465                 obd_ioctl_freedata(buf, len);
2466         return rc;
2467 }
2468
2469 int ll_process_config(struct lustre_cfg *lcfg)
2470 {
2471         struct super_block *sb;
2472         unsigned long x;
2473         int rc = 0;
2474         char *ptr;
2475
2476         /* The instance name contains the sb: lustre-client-aacfe000 */
2477         ptr = strrchr(lustre_cfg_string(lcfg, 0), '-');
2478         if (!ptr || !*(++ptr))
2479                 return -EINVAL;
2480         if (sscanf(ptr, "%lx", &x) != 1)
2481                 return -EINVAL;
2482         sb = (struct super_block *)x;
2483         /* This better be a real Lustre superblock! */
2484         LASSERT(s2lsi(sb)->lsi_lmd->lmd_magic == LMD_MAGIC);
2485
2486         /* Note we have not called client_common_fill_super yet, so
2487            proc fns must be able to handle that! */
2488         rc = class_process_proc_seq_param(PARAM_LLITE, lprocfs_llite_obd_vars,
2489                                           lcfg, sb);
2490         if (rc > 0)
2491                 rc = 0;
2492         return rc;
2493 }
2494
2495 /* this function prepares md_op_data hint for passing ot down to MD stack. */
2496 struct md_op_data * ll_prep_md_op_data(struct md_op_data *op_data,
2497                                        struct inode *i1, struct inode *i2,
2498                                        const char *name, int namelen,
2499                                        int mode, __u32 opc, void *data)
2500 {
2501         LASSERT(i1 != NULL);
2502
2503         if (namelen > ll_i2sbi(i1)->ll_namelen)
2504                 return ERR_PTR(-ENAMETOOLONG);
2505
2506         if (op_data == NULL)
2507                 OBD_ALLOC_PTR(op_data);
2508
2509         if (op_data == NULL)
2510                 return ERR_PTR(-ENOMEM);
2511
2512         ll_i2gids(op_data->op_suppgids, i1, i2);
2513         op_data->op_fid1 = *ll_inode2fid(i1);
2514         op_data->op_capa1 = ll_mdscapa_get(i1);
2515         if (S_ISDIR(i1->i_mode))
2516                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2517
2518         if (i2) {
2519                 op_data->op_fid2 = *ll_inode2fid(i2);
2520                 op_data->op_capa2 = ll_mdscapa_get(i2);
2521                 if (S_ISDIR(i2->i_mode))
2522                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2523         } else {
2524                 fid_zero(&op_data->op_fid2);
2525                 op_data->op_capa2 = NULL;
2526         }
2527
2528         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2529                 op_data->op_cli_flags |= CLI_HASH64;
2530
2531         if (ll_need_32bit_api(ll_i2sbi(i1)))
2532                 op_data->op_cli_flags |= CLI_API32;
2533
2534         op_data->op_name = name;
2535         op_data->op_namelen = namelen;
2536         op_data->op_mode = mode;
2537         op_data->op_mod_time = cfs_time_current_sec();
2538         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2539         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2540         op_data->op_cap = cfs_curproc_cap_pack();
2541         op_data->op_bias = 0;
2542         op_data->op_cli_flags = 0;
2543         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2544              filename_is_volatile(name, namelen, NULL))
2545                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2546         op_data->op_mds = 0;
2547         op_data->op_data = data;
2548
2549         /* If the file is being opened after mknod() (normally due to NFS)
2550          * try to use the default stripe data from parent directory for
2551          * allocating OST objects.  Try to pass the parent FID to MDS. */
2552         if (opc == LUSTRE_OPC_CREATE && i1 == i2 && S_ISREG(i2->i_mode) &&
2553             !ll_i2info(i2)->lli_has_smd) {
2554                 struct ll_inode_info *lli = ll_i2info(i2);
2555
2556                 spin_lock(&lli->lli_lock);
2557                 if (likely(!lli->lli_has_smd && !fid_is_zero(&lli->lli_pfid)))
2558                         op_data->op_fid1 = lli->lli_pfid;
2559                 spin_unlock(&lli->lli_lock);
2560                 /** We ignore parent's capability temporary. */
2561         }
2562
2563         /* When called by ll_setattr_raw, file is i1. */
2564         if (LLIF_DATA_MODIFIED & ll_i2info(i1)->lli_flags)
2565                 op_data->op_bias |= MDS_DATA_MODIFIED;
2566
2567         return op_data;
2568 }
2569
2570 void ll_finish_md_op_data(struct md_op_data *op_data)
2571 {
2572         capa_put(op_data->op_capa1);
2573         capa_put(op_data->op_capa2);
2574         OBD_FREE_PTR(op_data);
2575 }
2576
2577 #ifdef HAVE_SUPEROPS_USE_DENTRY
2578 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2579 #else
2580 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2581 #endif
2582 {
2583         struct ll_sb_info *sbi;
2584
2585 #ifdef HAVE_SUPEROPS_USE_DENTRY
2586         LASSERT((seq != NULL) && (dentry != NULL));
2587         sbi = ll_s2sbi(dentry->d_sb);
2588 #else
2589         LASSERT((seq != NULL) && (vfs != NULL));
2590         sbi = ll_s2sbi(vfs->mnt_sb);
2591 #endif
2592
2593         if (sbi->ll_flags & LL_SBI_NOLCK)
2594                 seq_puts(seq, ",nolock");
2595
2596         if (sbi->ll_flags & LL_SBI_FLOCK)
2597                 seq_puts(seq, ",flock");
2598
2599         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2600                 seq_puts(seq, ",localflock");
2601
2602         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2603                 seq_puts(seq, ",user_xattr");
2604
2605         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2606                 seq_puts(seq, ",lazystatfs");
2607
2608         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2609                 seq_puts(seq, ",user_fid2path");
2610
2611         RETURN(0);
2612 }
2613
2614 /**
2615  * Get obd name by cmd, and copy out to user space
2616  */
2617 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2618 {
2619         struct ll_sb_info *sbi = ll_i2sbi(inode);
2620         struct obd_device *obd;
2621         ENTRY;
2622
2623         if (cmd == OBD_IOC_GETDTNAME)
2624                 obd = class_exp2obd(sbi->ll_dt_exp);
2625         else if (cmd == OBD_IOC_GETMDNAME)
2626                 obd = class_exp2obd(sbi->ll_md_exp);
2627         else
2628                 RETURN(-EINVAL);
2629
2630         if (!obd)
2631                 RETURN(-ENOENT);
2632
2633         if (copy_to_user((void *)arg, obd->obd_name,
2634                          strlen(obd->obd_name) + 1))
2635                 RETURN(-EFAULT);
2636
2637         RETURN(0);
2638 }
2639
2640 /**
2641  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2642  * fsname will be returned in this buffer; otherwise, a static buffer will be
2643  * used to store the fsname and returned to caller.
2644  */
2645 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2646 {
2647         static char fsname_static[MTI_NAME_MAXLEN];
2648         struct lustre_sb_info *lsi = s2lsi(sb);
2649         char *ptr;
2650         int len;
2651
2652         if (buf == NULL) {
2653                 /* this means the caller wants to use static buffer
2654                  * and it doesn't care about race. Usually this is
2655                  * in error reporting path */
2656                 buf = fsname_static;
2657                 buflen = sizeof(fsname_static);
2658         }
2659
2660         len = strlen(lsi->lsi_lmd->lmd_profile);
2661         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2662         if (ptr && (strcmp(ptr, "-client") == 0))
2663                 len -= 7;
2664
2665         if (unlikely(len >= buflen))
2666                 len = buflen - 1;
2667         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2668         buf[len] = '\0';
2669
2670         return buf;
2671 }
2672
2673 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2674 {
2675         char *path = NULL;
2676
2677         struct path p;
2678
2679         p.dentry = dentry;
2680         p.mnt = current->fs->root.mnt;
2681         path_get(&p);
2682         path = d_path(&p, buf, bufsize);
2683         path_put(&p);
2684         return path;
2685 }
2686
2687 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2688 {
2689         char *buf, *path = NULL;
2690         struct dentry *dentry = NULL;
2691         struct ccc_object *obj = cl_inode2ccc(page->mapping->host);
2692
2693         /* this can be called inside spin lock so use GFP_ATOMIC. */
2694         buf = (char *)__get_free_page(GFP_ATOMIC);
2695         if (buf != NULL) {
2696                 dentry = d_find_alias(page->mapping->host);
2697                 if (dentry != NULL)
2698                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2699         }
2700
2701         CDEBUG(D_WARNING,
2702                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2703                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2704                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2705                PFID(&obj->cob_header.coh_lu.loh_fid),
2706                (path && !IS_ERR(path)) ? path : "", ioret);
2707
2708         if (dentry != NULL)
2709                 dput(dentry);
2710
2711         if (buf != NULL)
2712                 free_page((unsigned long)buf);
2713 }