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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/random.h>
  41 #include <linux/statfs.h>
  42 #include <linux/time.h>
  43 #include <linux/types.h>
  44 #include <libcfs/linux/linux-uuid.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51 #include <linux/security.h>
  52
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #ifdef HAVE_UAPI_LINUX_MOUNT_H
  55 #include <uapi/linux/mount.h>
  56 #endif
  57
  58 #include <lustre_ha.h>
  59 #include <lustre_dlm.h>
  60 #include <lprocfs_status.h>
  61 #include <lustre_disk.h>
  62 #include <uapi/linux/lustre/lustre_param.h>
  63 #include <lustre_log.h>
  64 #include <cl_object.h>
  65 #include <obd_cksum.h>
  66 #include "llite_internal.h"
  67
  68 struct kmem_cache *ll_file_data_slab;
  69
  70 #ifndef log2
  71 #define log2(n) ffz(~(n))
  72 #endif
  73
  74 static struct ll_sb_info *ll_init_sbi(void)
  75 {
  76         struct ll_sb_info *sbi = NULL;
  77         unsigned long pages;
  78         unsigned long lru_page_max;
  79         struct sysinfo si;
  80         int rc;
  81         int i;
  82
  83         ENTRY;
  84
  85         OBD_ALLOC_PTR(sbi);
  86         if (sbi == NULL)
  87                 RETURN(ERR_PTR(-ENOMEM));
  88
  89         rc = pcc_super_init(&sbi->ll_pcc_super);
  90         if (rc < 0)
  91                 GOTO(out_sbi, rc);
  92
  93         spin_lock_init(&sbi->ll_lock);
  94         mutex_init(&sbi->ll_lco.lco_lock);
  95         spin_lock_init(&sbi->ll_pp_extent_lock);
  96         spin_lock_init(&sbi->ll_process_lock);
  97         sbi->ll_rw_stats_on = 0;
  98         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
  99
 100         si_meminfo(&si);
 101         pages = si.totalram - si.totalhigh;
 102         lru_page_max = pages / 2;
 103
 104         sbi->ll_ra_info.ra_async_max_active = 0;
 105         sbi->ll_ra_info.ll_readahead_wq =
 106                 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
 107                                 sbi->ll_ra_info.ra_async_max_active);
 108         if (!sbi->ll_ra_info.ll_readahead_wq)
 109                 GOTO(out_pcc, rc = -ENOMEM);
 110
 111         /* initialize ll_cache data */
 112         sbi->ll_cache = cl_cache_init(lru_page_max);
 113         if (sbi->ll_cache == NULL)
 114                 GOTO(out_destroy_ra, rc = -ENOMEM);
 115
 116         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 117                                            SBI_DEFAULT_READAHEAD_MAX);
 118         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 119                                 sbi->ll_ra_info.ra_max_pages_per_file;
 120         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 121         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 122
 123         sbi->ll_flags |= LL_SBI_VERBOSE;
 124 #ifdef ENABLE_CHECKSUM
 125         sbi->ll_flags |= LL_SBI_CHECKSUM;
 126 #endif
 127 #ifdef ENABLE_FLOCK
 128         sbi->ll_flags |= LL_SBI_FLOCK;
 129 #endif
 130
 131 #ifdef HAVE_LRU_RESIZE_SUPPORT
 132         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 133 #endif
 134         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 135
 136         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 137                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 138                                pp_r_hist.oh_lock);
 139                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 140                                pp_w_hist.oh_lock);
 141         }
 142
 143         /* metadata statahead is enabled by default */
 144         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 145         sbi->ll_sa_max = LL_SA_RPC_DEF;
 146         atomic_set(&sbi->ll_sa_total, 0);
 147         atomic_set(&sbi->ll_sa_wrong, 0);
 148         atomic_set(&sbi->ll_sa_running, 0);
 149         atomic_set(&sbi->ll_agl_total, 0);
 150         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 151         sbi->ll_flags |= LL_SBI_FAST_READ;
 152         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 153
 154         /* root squash */
 155         sbi->ll_squash.rsi_uid = 0;
 156         sbi->ll_squash.rsi_gid = 0;
 157         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 158         spin_lock_init(&sbi->ll_squash.rsi_lock);
 159
 160         /* Per-filesystem file heat */
 161         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 162         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 163         RETURN(sbi);
 164 out_destroy_ra:
 165         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 166 out_pcc:
 167         pcc_super_fini(&sbi->ll_pcc_super);
 168 out_sbi:
 169         OBD_FREE_PTR(sbi);
 170         RETURN(ERR_PTR(rc));
 171 }
 172
 173 static void ll_free_sbi(struct super_block *sb)
 174 {
 175         struct ll_sb_info *sbi = ll_s2sbi(sb);
 176         ENTRY;
 177
 178         if (sbi != NULL) {
 179                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 180                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 181                 if (sbi->ll_ra_info.ll_readahead_wq)
 182                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 183                 if (sbi->ll_cache != NULL) {
 184                         cl_cache_decref(sbi->ll_cache);
 185                         sbi->ll_cache = NULL;
 186                 }
 187                 pcc_super_fini(&sbi->ll_pcc_super);
 188                 OBD_FREE(sbi, sizeof(*sbi));
 189         }
 190         EXIT;
 191 }
 192
 193 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 194                                     struct vfsmount *mnt)
 195 {
 196         struct inode *root = NULL;
 197         struct ll_sb_info *sbi = ll_s2sbi(sb);
 198         struct obd_statfs *osfs = NULL;
 199         struct ptlrpc_request *request = NULL;
 200         struct obd_connect_data *data = NULL;
 201         struct obd_uuid *uuid;
 202         struct md_op_data *op_data;
 203         struct lustre_md lmd;
 204         u64 valid;
 205         int size, err, checksum;
 206
 207         ENTRY;
 208         sbi->ll_md_obd = class_name2obd(md);
 209         if (!sbi->ll_md_obd) {
 210                 CERROR("MD %s: not setup or attached\n", md);
 211                 RETURN(-EINVAL);
 212         }
 213
 214         OBD_ALLOC_PTR(data);
 215         if (data == NULL)
 216                 RETURN(-ENOMEM);
 217
 218         OBD_ALLOC_PTR(osfs);
 219         if (osfs == NULL) {
 220                 OBD_FREE_PTR(data);
 221                 RETURN(-ENOMEM);
 222         }
 223
 224         /* pass client page size via ocd_grant_blkbits, the server should report
 225          * back its backend blocksize for grant calculation purpose */
 226         data->ocd_grant_blkbits = PAGE_SHIFT;
 227
 228         /* indicate MDT features supported by this client */
 229         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 230                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 231                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 232                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 233                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 234                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 235                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 236                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 237                                   OBD_CONNECT_64BITHASH |
 238                                   OBD_CONNECT_EINPROGRESS |
 239                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 240                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 241                                   OBD_CONNECT_MAX_EASIZE |
 242                                   OBD_CONNECT_FLOCK_DEAD |
 243                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 244                                   OBD_CONNECT_OPEN_BY_FID |
 245                                   OBD_CONNECT_DIR_STRIPE |
 246                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 247                                   OBD_CONNECT_SUBTREE |
 248                                   OBD_CONNECT_MULTIMODRPCS |
 249                                   OBD_CONNECT_GRANT_PARAM |
 250                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 251
 252         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 253                                    OBD_CONNECT2_SUM_STATFS |
 254                                    OBD_CONNECT2_OVERSTRIPING |
 255                                    OBD_CONNECT2_FLR |
 256                                    OBD_CONNECT2_LOCK_CONVERT |
 257                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 258                                    OBD_CONNECT2_LSOM |
 259                                    OBD_CONNECT2_ASYNC_DISCARD |
 260                                    OBD_CONNECT2_PCC;
 261
 262 #ifdef HAVE_LRU_RESIZE_SUPPORT
 263         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 264                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 265 #endif
 266 #ifdef CONFIG_FS_POSIX_ACL
 267         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 268                                    OBD_CONNECT_LARGE_ACL;
 269 #endif
 270
 271         data->ocd_cksum_types = obd_cksum_types_supported_client();
 272
 273         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 274                 /* flag mdc connection as lightweight, only used for test
 275                  * purpose, use with care */
 276                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 277
 278         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 279         data->ocd_version = LUSTRE_VERSION_CODE;
 280
 281         if (sb->s_flags & SB_RDONLY)
 282                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 283         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 284                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 285
 286 #ifdef SB_NOSEC
 287         /* Setting this indicates we correctly support S_NOSEC (See kernel
 288          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 289          */
 290         sb->s_flags |= SB_NOSEC;
 291 #endif
 292
 293         if (sbi->ll_flags & LL_SBI_FLOCK)
 294                 sbi->ll_fop = &ll_file_operations_flock;
 295         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 296                 sbi->ll_fop = &ll_file_operations;
 297         else
 298                 sbi->ll_fop = &ll_file_operations_noflock;
 299
 300         /* always ping even if server suppress_pings */
 301         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 302                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 303
 304         obd_connect_set_secctx(data);
 305
 306 #if defined(CONFIG_SECURITY)
 307         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 308 #endif
 309
 310         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 311
 312         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 313                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 314         if (err == -EBUSY) {
 315                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 316                                    "recovery, of which this client is not a "
 317                                    "part. Please wait for recovery to complete,"
 318                                    " abort, or time out.\n", md);
 319                 GOTO(out, err);
 320         } else if (err) {
 321                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 322                 GOTO(out, err);
 323         }
 324
 325         sbi->ll_md_exp->exp_connect_data = *data;
 326
 327         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 328                            LUSTRE_SEQ_METADATA);
 329         if (err) {
 330                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 331                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 332                 GOTO(out_md, err);
 333         }
 334
 335         /* For mount, we only need fs info from MDT0, and also in DNE, it
 336          * can make sure the client can be mounted as long as MDT0 is
 337          * avaible */
 338         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 339                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 340                         OBD_STATFS_FOR_MDT0);
 341         if (err)
 342                 GOTO(out_md_fid, err);
 343
 344         /* This needs to be after statfs to ensure connect has finished.
 345          * Note that "data" does NOT contain the valid connect reply.
 346          * If connecting to a 1.8 server there will be no LMV device, so
 347          * we can access the MDC export directly and exp_connect_flags will
 348          * be non-zero, but if accessing an upgraded 2.1 server it will
 349          * have the correct flags filled in.
 350          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 351         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 352         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 353             valid != CLIENT_CONNECT_MDT_REQD) {
 354                 char *buf;
 355
 356                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 357                 obd_connect_flags2str(buf, PAGE_SIZE,
 358                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 359                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 360                                    "feature(s) needed for correct operation "
 361                                    "of this client (%s). Please upgrade "
 362                                    "server or downgrade client.\n",
 363                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 364                 OBD_FREE(buf, PAGE_SIZE);
 365                 GOTO(out_md_fid, err = -EPROTO);
 366         }
 367
 368         size = sizeof(*data);
 369         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 370                            KEY_CONN_DATA,  &size, data);
 371         if (err) {
 372                 CERROR("%s: Get connect data failed: rc = %d\n",
 373                        sbi->ll_md_exp->exp_obd->obd_name, err);
 374                 GOTO(out_md_fid, err);
 375         }
 376
 377         LASSERT(osfs->os_bsize);
 378         sb->s_blocksize = osfs->os_bsize;
 379         sb->s_blocksize_bits = log2(osfs->os_bsize);
 380         sb->s_magic = LL_SUPER_MAGIC;
 381         sb->s_maxbytes = MAX_LFS_FILESIZE;
 382         sbi->ll_namelen = osfs->os_namelen;
 383         sbi->ll_mnt.mnt = current->fs->root.mnt;
 384
 385         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 386             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 387                 LCONSOLE_INFO("Disabling user_xattr feature because "
 388                               "it is not supported on the server\n");
 389                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 390         }
 391
 392         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 393 #ifdef SB_POSIXACL
 394                 sb->s_flags |= SB_POSIXACL;
 395 #endif
 396                 sbi->ll_flags |= LL_SBI_ACL;
 397         } else {
 398                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 399 #ifdef SB_POSIXACL
 400                 sb->s_flags &= ~SB_POSIXACL;
 401 #endif
 402                 sbi->ll_flags &= ~LL_SBI_ACL;
 403         }
 404
 405         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 406                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 407
 408         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 409                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 410
 411         if (obd_connect_has_secctx(data))
 412                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 413
 414         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 415                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 416                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 417                                       "unknown maximum xattr size.\n", dt);
 418                 } else if (!sbi->ll_xattr_cache_set) {
 419                         /* If xattr_cache is already set (no matter 0 or 1)
 420                          * during processing llog, it won't be enabled here. */
 421                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 422                         sbi->ll_xattr_cache_enabled = 1;
 423                 }
 424         }
 425
 426         sbi->ll_dt_obd = class_name2obd(dt);
 427         if (!sbi->ll_dt_obd) {
 428                 CERROR("DT %s: not setup or attached\n", dt);
 429                 GOTO(out_md_fid, err = -ENODEV);
 430         }
 431
 432         /* pass client page size via ocd_grant_blkbits, the server should report
 433          * back its backend blocksize for grant calculation purpose */
 434         data->ocd_grant_blkbits = PAGE_SHIFT;
 435
 436         /* indicate OST features supported by this client */
 437         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 438                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 439                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 440                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 441                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 442                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 443                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 444                                   OBD_CONNECT_EINPROGRESS |
 445                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 446                                   OBD_CONNECT_LAYOUTLOCK |
 447                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 448                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 449                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 450
 451 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 452  * can interoperate with an older version of lockahead which was released prior
 453  * to landing in master. This support will be dropped when 2.13 development
 454  * starts.  At the point, we should not just drop the connect flag (below), we
 455  * should also remove the support in the code.
 456  *
 457  * Removing it means a few things:
 458  * 1. Remove this section here
 459  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 460  * 3. Remove function exp_connect_lockahead_old
 461  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 462  * */
 463 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 464         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 465 #endif
 466
 467         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 468
 469         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 470                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 471
 472         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 473          * disabled by default, because it can still be enabled on the
 474          * fly via /sys. As a consequence, we still need to come to an
 475          * agreement on the supported algorithms at connect time
 476          */
 477         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 478
 479         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 480                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 481         else
 482                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 483
 484 #ifdef HAVE_LRU_RESIZE_SUPPORT
 485         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 486 #endif
 487         /* always ping even if server suppress_pings */
 488         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 489                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 490
 491         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 492                "ocd_grant: %d\n", data->ocd_connect_flags,
 493                data->ocd_version, data->ocd_grant);
 494
 495         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 496         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 497
 498         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 499
 500         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 501                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 502         if (err == -EBUSY) {
 503                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 504                                    "recovery, of which this client is not a "
 505                                    "part.  Please wait for recovery to "
 506                                    "complete, abort, or time out.\n", dt);
 507                 GOTO(out_md, err);
 508         } else if (err) {
 509                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 510                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 511                 GOTO(out_md, err);
 512         }
 513
 514         sbi->ll_dt_exp->exp_connect_data = *data;
 515
 516         /* Don't change value if it was specified in the config log */
 517         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 518                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 519                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 520                               (data->ocd_brw_size >> PAGE_SHIFT));
 521                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 522                     sbi->ll_ra_info.ra_max_pages_per_file)
 523                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 524                                 sbi->ll_ra_info.ra_max_pages_per_file;
 525         }
 526
 527         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 528                            LUSTRE_SEQ_METADATA);
 529         if (err) {
 530                 CERROR("%s: Can't init data layer FID infrastructure, "
 531                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 532                 GOTO(out_dt, err);
 533         }
 534
 535         mutex_lock(&sbi->ll_lco.lco_lock);
 536         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 537         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 538         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 539         mutex_unlock(&sbi->ll_lco.lco_lock);
 540
 541         fid_zero(&sbi->ll_root_fid);
 542         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 543                            &sbi->ll_root_fid);
 544         if (err) {
 545                 CERROR("cannot mds_connect: rc = %d\n", err);
 546                 GOTO(out_lock_cn_cb, err);
 547         }
 548         if (!fid_is_sane(&sbi->ll_root_fid)) {
 549                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 550                        sbi->ll_md_exp->exp_obd->obd_name,
 551                        PFID(&sbi->ll_root_fid));
 552                 GOTO(out_lock_cn_cb, err = -EINVAL);
 553         }
 554         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 555
 556         sb->s_op = &lustre_super_operations;
 557 #ifdef HAVE_XATTR_HANDLER_FLAGS
 558         sb->s_xattr = ll_xattr_handlers;
 559 #endif
 560 #if THREAD_SIZE >= 8192 /*b=17630*/
 561         sb->s_export_op = &lustre_export_operations;
 562 #endif
 563
 564         /* make root inode
 565          * XXX: move this to after cbd setup? */
 566         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 567         if (sbi->ll_flags & LL_SBI_ACL)
 568                 valid |= OBD_MD_FLACL;
 569
 570         OBD_ALLOC_PTR(op_data);
 571         if (op_data == NULL)
 572                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 573
 574         op_data->op_fid1 = sbi->ll_root_fid;
 575         op_data->op_mode = 0;
 576         op_data->op_valid = valid;
 577
 578         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 579
 580         OBD_FREE_PTR(op_data);
 581         if (err) {
 582                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 583                        sbi->ll_md_exp->exp_obd->obd_name, err);
 584                 GOTO(out_lock_cn_cb, err);
 585         }
 586
 587         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 588                                sbi->ll_md_exp, &lmd);
 589         if (err) {
 590                 CERROR("failed to understand root inode md: rc = %d\n", err);
 591                 ptlrpc_req_finished(request);
 592                 GOTO(out_lock_cn_cb, err);
 593         }
 594
 595         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 596         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 597                                             sbi->ll_flags & LL_SBI_32BIT_API),
 598                        &lmd);
 599         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 600         ptlrpc_req_finished(request);
 601
 602         if (IS_ERR(root)) {
 603 #ifdef CONFIG_FS_POSIX_ACL
 604                 if (lmd.posix_acl) {
 605                         posix_acl_release(lmd.posix_acl);
 606                         lmd.posix_acl = NULL;
 607                 }
 608 #endif
 609                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 610                 root = NULL;
 611                 CERROR("lustre_lite: bad iget4 for root\n");
 612                 GOTO(out_root, err);
 613         }
 614
 615         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 616         if (sbi->ll_checksum_set) {
 617                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 618                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 619                                          sizeof(checksum), &checksum, NULL);
 620                 if (err) {
 621                         CERROR("%s: Set checksum failed: rc = %d\n",
 622                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 623                         GOTO(out_root, err);
 624                 }
 625         }
 626         cl_sb_init(sb);
 627
 628         sb->s_root = d_make_root(root);
 629         if (sb->s_root == NULL) {
 630                 err = -ENOMEM;
 631                 CERROR("%s: can't make root dentry: rc = %d\n",
 632                        sbi->ll_fsname, err);
 633                 GOTO(out_root, err);
 634         }
 635 #ifdef HAVE_DCACHE_LOCK
 636         sb->s_root->d_op = &ll_d_ops;
 637 #endif
 638
 639         sbi->ll_sdev_orig = sb->s_dev;
 640
 641         /* We set sb->s_dev equal on all lustre clients in order to support
 642          * NFS export clustering.  NFSD requires that the FSID be the same
 643          * on all clients. */
 644         /* s_dev is also used in lt_compare() to compare two fs, but that is
 645          * only a node-local comparison. */
 646         uuid = obd_get_uuid(sbi->ll_md_exp);
 647         if (uuid != NULL)
 648                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 649
 650         if (data != NULL)
 651                 OBD_FREE_PTR(data);
 652         if (osfs != NULL)
 653                 OBD_FREE_PTR(osfs);
 654
 655         if (sbi->ll_dt_obd) {
 656                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 657                                         &sbi->ll_dt_obd->obd_kset.kobj,
 658                                         sbi->ll_dt_obd->obd_type->typ_name);
 659                 if (err < 0) {
 660                         CERROR("%s: could not register %s in llite: rc = %d\n",
 661                                dt, sbi->ll_fsname, err);
 662                         err = 0;
 663                 }
 664         }
 665
 666         if (sbi->ll_md_obd) {
 667                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 668                                         &sbi->ll_md_obd->obd_kset.kobj,
 669                                         sbi->ll_md_obd->obd_type->typ_name);
 670                 if (err < 0) {
 671                         CERROR("%s: could not register %s in llite: rc = %d\n",
 672                                md, sbi->ll_fsname, err);
 673                         err = 0;
 674                 }
 675         }
 676
 677         RETURN(err);
 678 out_root:
 679         if (root)
 680                 iput(root);
 681 out_lock_cn_cb:
 682         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 683 out_dt:
 684         obd_disconnect(sbi->ll_dt_exp);
 685         sbi->ll_dt_exp = NULL;
 686         sbi->ll_dt_obd = NULL;
 687 out_md_fid:
 688         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 689 out_md:
 690         obd_disconnect(sbi->ll_md_exp);
 691         sbi->ll_md_exp = NULL;
 692         sbi->ll_md_obd = NULL;
 693 out:
 694         if (data != NULL)
 695                 OBD_FREE_PTR(data);
 696         if (osfs != NULL)
 697                 OBD_FREE_PTR(osfs);
 698         return err;
 699 }
 700
 701 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 702 {
 703         int size, rc;
 704
 705         size = sizeof(*lmmsize);
 706         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 707                           KEY_MAX_EASIZE, &size, lmmsize);
 708         if (rc != 0) {
 709                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 710                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 711                 RETURN(rc);
 712         }
 713
 714         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 715
 716         size = sizeof(int);
 717         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 718                           KEY_MAX_EASIZE, &size, lmmsize);
 719         if (rc)
 720                 CERROR("Get max mdsize error rc %d\n", rc);
 721
 722         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 723
 724         RETURN(rc);
 725 }
 726
 727 /**
 728  * Get the value of the default_easize parameter.
 729  *
 730  * \see client_obd::cl_default_mds_easize
 731  *
 732  * \param[in] sbi       superblock info for this filesystem
 733  * \param[out] lmmsize  pointer to storage location for value
 734  *
 735  * \retval 0            on success
 736  * \retval negative     negated errno on failure
 737  */
 738 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 739 {
 740         int size, rc;
 741
 742         size = sizeof(int);
 743         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 744                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 745         if (rc)
 746                 CERROR("Get default mdsize error rc %d\n", rc);
 747
 748         RETURN(rc);
 749 }
 750
 751 /**
 752  * Set the default_easize parameter to the given value.
 753  *
 754  * \see client_obd::cl_default_mds_easize
 755  *
 756  * \param[in] sbi       superblock info for this filesystem
 757  * \param[in] lmmsize   the size to set
 758  *
 759  * \retval 0            on success
 760  * \retval negative     negated errno on failure
 761  */
 762 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 763 {
 764         int rc;
 765
 766         if (lmmsize < sizeof(struct lov_mds_md) ||
 767             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 768                 return -EINVAL;
 769
 770         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 771                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 772                                 sizeof(int), &lmmsize, NULL);
 773
 774         RETURN(rc);
 775 }
 776
 777 static void client_common_put_super(struct super_block *sb)
 778 {
 779         struct ll_sb_info *sbi = ll_s2sbi(sb);
 780         ENTRY;
 781
 782         cl_sb_fini(sb);
 783
 784         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 785         obd_disconnect(sbi->ll_dt_exp);
 786         sbi->ll_dt_exp = NULL;
 787
 788         ll_debugfs_unregister_super(sb);
 789
 790         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 791         obd_disconnect(sbi->ll_md_exp);
 792         sbi->ll_md_exp = NULL;
 793
 794         EXIT;
 795 }
 796
 797 void ll_kill_super(struct super_block *sb)
 798 {
 799         struct ll_sb_info *sbi;
 800         ENTRY;
 801
 802         /* not init sb ?*/
 803         if (!(sb->s_flags & SB_ACTIVE))
 804                 return;
 805
 806         sbi = ll_s2sbi(sb);
 807         /* we need restore s_dev from changed for clustred NFS before put_super
 808          * because new kernels have cached s_dev and change sb->s_dev in
 809          * put_super not affected real removing devices */
 810         if (sbi) {
 811                 sb->s_dev = sbi->ll_sdev_orig;
 812                 sbi->ll_umounting = 1;
 813
 814                 /* wait running statahead threads to quit */
 815                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 816                         set_current_state(TASK_UNINTERRUPTIBLE);
 817                         schedule_timeout(cfs_time_seconds(1 >> 3));
 818                 }
 819         }
 820
 821         EXIT;
 822 }
 823
 824 static inline int ll_set_opt(const char *opt, char *data, int fl)
 825 {
 826         if (strncmp(opt, data, strlen(opt)) != 0)
 827                 return 0;
 828         else
 829                 return fl;
 830 }
 831
 832 /* non-client-specific mount options are parsed in lmd_parse */
 833 static int ll_options(char *options, struct ll_sb_info *sbi)
 834 {
 835         int tmp;
 836         char *s1 = options, *s2;
 837         int *flags = &sbi->ll_flags;
 838         ENTRY;
 839
 840         if (!options)
 841                 RETURN(0);
 842
 843         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 844
 845         while (*s1) {
 846                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 847                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 858                 if (tmp) {
 859                         *flags |= tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 863                 if (tmp) {
 864                         *flags &= ~tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 868                 if (tmp) {
 869                         *flags |= tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 873                 if (tmp) {
 874                         *flags &= ~tmp;
 875                         goto next;
 876                 }
 877                 tmp = ll_set_opt("context", s1, 1);
 878                 if (tmp)
 879                         goto next;
 880                 tmp = ll_set_opt("fscontext", s1, 1);
 881                 if (tmp)
 882                         goto next;
 883                 tmp = ll_set_opt("defcontext", s1, 1);
 884                 if (tmp)
 885                         goto next;
 886                 tmp = ll_set_opt("rootcontext", s1, 1);
 887                 if (tmp)
 888                         goto next;
 889                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 890                 if (tmp) {
 891                         *flags |= tmp;
 892                         goto next;
 893                 }
 894                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 895                 if (tmp) {
 896                         *flags &= ~tmp;
 897                         goto next;
 898                 }
 899
 900                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 901                 if (tmp) {
 902                         *flags |= tmp;
 903                         sbi->ll_checksum_set = 1;
 904                         goto next;
 905                 }
 906                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 907                 if (tmp) {
 908                         *flags &= ~tmp;
 909                         sbi->ll_checksum_set = 1;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 913                 if (tmp) {
 914                         *flags |= tmp;
 915                         goto next;
 916                 }
 917                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 918                 if (tmp) {
 919                         *flags &= ~tmp;
 920                         goto next;
 921                 }
 922                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 923                 if (tmp) {
 924                         *flags |= tmp;
 925                         goto next;
 926                 }
 927                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 928                 if (tmp) {
 929                         *flags &= ~tmp;
 930                         goto next;
 931                 }
 932                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 933                 if (tmp) {
 934                         *flags |= tmp;
 935                         goto next;
 936                 }
 937                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 938                 if (tmp) {
 939                         *flags |= tmp;
 940                         goto next;
 941                 }
 942                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 943                 if (tmp) {
 944                         *flags &= ~tmp;
 945                         goto next;
 946                 }
 947                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 948                 if (tmp) {
 949                         *flags |= tmp;
 950                         goto next;
 951                 }
 952                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 953                                    s1);
 954                 RETURN(-EINVAL);
 955
 956 next:
 957                 /* Find next opt */
 958                 s2 = strchr(s1, ',');
 959                 if (s2 == NULL)
 960                         break;
 961                 s1 = s2 + 1;
 962         }
 963         RETURN(0);
 964 }
 965
 966 void ll_lli_init(struct ll_inode_info *lli)
 967 {
 968         lli->lli_inode_magic = LLI_INODE_MAGIC;
 969         lli->lli_flags = 0;
 970         spin_lock_init(&lli->lli_lock);
 971         lli->lli_posix_acl = NULL;
 972         /* Do not set lli_fid, it has been initialized already. */
 973         fid_zero(&lli->lli_pfid);
 974         lli->lli_mds_read_och = NULL;
 975         lli->lli_mds_write_och = NULL;
 976         lli->lli_mds_exec_och = NULL;
 977         lli->lli_open_fd_read_count = 0;
 978         lli->lli_open_fd_write_count = 0;
 979         lli->lli_open_fd_exec_count = 0;
 980         mutex_init(&lli->lli_och_mutex);
 981         spin_lock_init(&lli->lli_agl_lock);
 982         spin_lock_init(&lli->lli_layout_lock);
 983         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 984         lli->lli_clob = NULL;
 985
 986         init_rwsem(&lli->lli_xattrs_list_rwsem);
 987         mutex_init(&lli->lli_xattrs_enq_lock);
 988
 989         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 990         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 991                 mutex_init(&lli->lli_readdir_mutex);
 992                 lli->lli_opendir_key = NULL;
 993                 lli->lli_sai = NULL;
 994                 spin_lock_init(&lli->lli_sa_lock);
 995                 lli->lli_opendir_pid = 0;
 996                 lli->lli_sa_enabled = 0;
 997                 init_rwsem(&lli->lli_lsm_sem);
 998         } else {
 999                 mutex_init(&lli->lli_size_mutex);
1000                 lli->lli_symlink_name = NULL;
1001                 init_rwsem(&lli->lli_trunc_sem);
1002                 range_lock_tree_init(&lli->lli_write_tree);
1003                 init_rwsem(&lli->lli_glimpse_sem);
1004                 lli->lli_glimpse_time = ktime_set(0, 0);
1005                 INIT_LIST_HEAD(&lli->lli_agl_list);
1006                 lli->lli_agl_index = 0;
1007                 lli->lli_async_rc = 0;
1008                 spin_lock_init(&lli->lli_heat_lock);
1009                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1010                 lli->lli_heat_flags = 0;
1011                 mutex_init(&lli->lli_pcc_lock);
1012                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1013                 lli->lli_pcc_inode = NULL;
1014         }
1015         mutex_init(&lli->lli_layout_mutex);
1016         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1017 }
1018
1019 #define MAX_STRING_SIZE 128
1020
1021 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1022
1023 #define LSI_BDI_INITIALIZED     0x00400000
1024
1025 #ifndef HAVE_BDI_CAP_MAP_COPY
1026 # define BDI_CAP_MAP_COPY       0
1027 #endif
1028
1029 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1030 {
1031         struct  lustre_sb_info *lsi = s2lsi(sb);
1032         char buf[MAX_STRING_SIZE];
1033         va_list args;
1034         int err;
1035
1036         err = bdi_init(&lsi->lsi_bdi);
1037         if (err)
1038                 return err;
1039
1040         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1041         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1042         lsi->lsi_bdi.name = "lustre";
1043         va_start(args, fmt);
1044         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1045         va_end(args);
1046         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1047         va_end(args);
1048         if (!err)
1049                 sb->s_bdi = &lsi->lsi_bdi;
1050
1051         return err;
1052 }
1053 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1054
1055 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1056 {
1057         struct  lustre_profile *lprof = NULL;
1058         struct  lustre_sb_info *lsi = s2lsi(sb);
1059         struct  ll_sb_info *sbi = NULL;
1060         char    *dt = NULL, *md = NULL;
1061         char    *profilenm = get_profile_name(sb);
1062         struct config_llog_instance *cfg;
1063         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1064         const int instlen = 16 + 2;
1065         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1066         char name[MAX_STRING_SIZE];
1067         int md_len = 0;
1068         int dt_len = 0;
1069         uuid_t uuid;
1070         char *ptr;
1071         int len;
1072         int err;
1073
1074         ENTRY;
1075         /* for ASLR, to map between cfg_instance and hashed ptr */
1076         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1077                profilenm, cfg_instance, sb);
1078
1079         try_module_get(THIS_MODULE);
1080
1081         OBD_ALLOC_PTR(cfg);
1082         if (cfg == NULL)
1083                 GOTO(out_free_cfg, err = -ENOMEM);
1084
1085         /* client additional sb info */
1086         lsi->lsi_llsbi = sbi = ll_init_sbi();
1087         if (IS_ERR(sbi))
1088                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1089
1090         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1091         if (err)
1092                 GOTO(out_free_cfg, err);
1093
1094 #ifndef HAVE_DCACHE_LOCK
1095         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1096         sb->s_d_op = &ll_d_ops;
1097 #endif
1098         /* UUID handling */
1099         generate_random_uuid(uuid.b);
1100         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1101
1102         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1103
1104         /* Get fsname */
1105         len = strlen(profilenm);
1106         ptr = strrchr(profilenm, '-');
1107         if (ptr && (strcmp(ptr, "-client") == 0))
1108                 len -= 7;
1109
1110         if (len > LUSTRE_MAXFSNAME) {
1111                 if (unlikely(len >= MAX_STRING_SIZE))
1112                         len = MAX_STRING_SIZE - 1;
1113                 strncpy(name, profilenm, len);
1114                 name[len] = '\0';
1115                 err = -ENAMETOOLONG;
1116                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1117                        name, LUSTRE_MAXFSNAME, err);
1118                 GOTO(out_free_cfg, err);
1119         }
1120         strncpy(sbi->ll_fsname, profilenm, len);
1121         sbi->ll_fsname[len] = '\0';
1122
1123         /* Mount info */
1124         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1125                  profilenm, cfg_instance);
1126
1127         err = super_setup_bdi_name(sb, "%s", name);
1128         if (err)
1129                 GOTO(out_free_cfg, err);
1130
1131         /* Call ll_debugfs_register_super() before lustre_process_log()
1132          * so that "llite.*.*" params can be processed correctly.
1133          */
1134         err = ll_debugfs_register_super(sb, name);
1135         if (err < 0) {
1136                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1137                        sbi->ll_fsname, err);
1138                 err = 0;
1139         }
1140
1141         /* The cfg_instance is a value unique to this super, in case some
1142          * joker tries to mount the same fs at two mount points.
1143          */
1144         cfg->cfg_instance = cfg_instance;
1145         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1146         cfg->cfg_callback = class_config_llog_handler;
1147         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1148         /* set up client obds */
1149         err = lustre_process_log(sb, profilenm, cfg);
1150         if (err < 0)
1151                 GOTO(out_debugfs, err);
1152
1153         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1154         lprof = class_get_profile(profilenm);
1155         if (lprof == NULL) {
1156                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1157                                    " read from the MGS.  Does that filesystem "
1158                                    "exist?\n", profilenm);
1159                 GOTO(out_debugfs, err = -EINVAL);
1160         }
1161         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1162                lprof->lp_md, lprof->lp_dt);
1163
1164         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1165         OBD_ALLOC(dt, dt_len);
1166         if (!dt)
1167                 GOTO(out_profile, err = -ENOMEM);
1168         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1169
1170         md_len = strlen(lprof->lp_md) + instlen + 2;
1171         OBD_ALLOC(md, md_len);
1172         if (!md)
1173                 GOTO(out_free_dt, err = -ENOMEM);
1174         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1175
1176         /* connections, registrations, sb setup */
1177         err = client_common_fill_super(sb, md, dt, mnt);
1178         if (err < 0)
1179                 GOTO(out_free_md, err);
1180
1181         sbi->ll_client_common_fill_super_succeeded = 1;
1182
1183 out_free_md:
1184         if (md)
1185                 OBD_FREE(md, md_len);
1186 out_free_dt:
1187         if (dt)
1188                 OBD_FREE(dt, dt_len);
1189 out_profile:
1190         if (lprof)
1191                 class_put_profile(lprof);
1192 out_debugfs:
1193         if (err < 0)
1194                 ll_debugfs_unregister_super(sb);
1195 out_free_cfg:
1196         if (cfg)
1197                 OBD_FREE_PTR(cfg);
1198
1199         if (err)
1200                 ll_put_super(sb);
1201         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1202                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1203         RETURN(err);
1204 } /* ll_fill_super */
1205
1206 void ll_put_super(struct super_block *sb)
1207 {
1208         struct config_llog_instance cfg, params_cfg;
1209         struct obd_device *obd;
1210         struct lustre_sb_info *lsi = s2lsi(sb);
1211         struct ll_sb_info *sbi = ll_s2sbi(sb);
1212         char *profilenm = get_profile_name(sb);
1213         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1214         long ccc_count;
1215         int next, force = 1, rc = 0;
1216         ENTRY;
1217
1218         if (IS_ERR(sbi))
1219                 GOTO(out_no_sbi, 0);
1220
1221         /* Should replace instance_id with something better for ASLR */
1222         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1223                profilenm, cfg_instance, sb);
1224
1225         cfg.cfg_instance = cfg_instance;
1226         lustre_end_log(sb, profilenm, &cfg);
1227
1228         params_cfg.cfg_instance = cfg_instance;
1229         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1230
1231         if (sbi->ll_md_exp) {
1232                 obd = class_exp2obd(sbi->ll_md_exp);
1233                 if (obd)
1234                         force = obd->obd_force;
1235         }
1236
1237         /* Wait for unstable pages to be committed to stable storage */
1238         if (force == 0) {
1239                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1240                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1241                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1242                         &lwi);
1243         }
1244
1245         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1246         if (force == 0 && rc != -EINTR)
1247                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1248
1249         /* We need to set force before the lov_disconnect in
1250            lustre_common_put_super, since l_d cleans up osc's as well. */
1251         if (force) {
1252                 next = 0;
1253                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1254                                                      &next)) != NULL) {
1255                         obd->obd_force = force;
1256                 }
1257         }
1258
1259         if (sbi->ll_client_common_fill_super_succeeded) {
1260                 /* Only if client_common_fill_super succeeded */
1261                 client_common_put_super(sb);
1262         }
1263
1264         next = 0;
1265         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1266                 class_manual_cleanup(obd);
1267         }
1268
1269         if (sbi->ll_flags & LL_SBI_VERBOSE)
1270                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1271
1272         if (profilenm)
1273                 class_del_profile(profilenm);
1274
1275 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1276         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1277                 bdi_destroy(&lsi->lsi_bdi);
1278                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1279         }
1280 #endif
1281
1282         ll_free_sbi(sb);
1283         lsi->lsi_llsbi = NULL;
1284 out_no_sbi:
1285         lustre_common_put_super(sb);
1286
1287         cl_env_cache_purge(~0);
1288
1289         module_put(THIS_MODULE);
1290
1291         EXIT;
1292 } /* client_put_super */
1293
1294 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1295 {
1296         struct inode *inode = NULL;
1297
1298         /* NOTE: we depend on atomic igrab() -bzzz */
1299         lock_res_and_lock(lock);
1300         if (lock->l_resource->lr_lvb_inode) {
1301                 struct ll_inode_info * lli;
1302                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1303                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1304                         inode = igrab(lock->l_resource->lr_lvb_inode);
1305                 } else {
1306                         inode = lock->l_resource->lr_lvb_inode;
1307                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1308                                          D_WARNING, lock, "lr_lvb_inode %p is "
1309                                          "bogus: magic %08x",
1310                                          lock->l_resource->lr_lvb_inode,
1311                                          lli->lli_inode_magic);
1312                         inode = NULL;
1313                 }
1314         }
1315         unlock_res_and_lock(lock);
1316         return inode;
1317 }
1318
1319 void ll_dir_clear_lsm_md(struct inode *inode)
1320 {
1321         struct ll_inode_info *lli = ll_i2info(inode);
1322
1323         LASSERT(S_ISDIR(inode->i_mode));
1324
1325         if (lli->lli_lsm_md) {
1326                 lmv_free_memmd(lli->lli_lsm_md);
1327                 lli->lli_lsm_md = NULL;
1328         }
1329
1330         if (lli->lli_default_lsm_md) {
1331                 lmv_free_memmd(lli->lli_default_lsm_md);
1332                 lli->lli_default_lsm_md = NULL;
1333         }
1334 }
1335
1336 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1337                                       const struct lu_fid *fid,
1338                                       struct lustre_md *md)
1339 {
1340         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1341         struct mdt_body         *body = md->body;
1342         struct inode            *inode;
1343         ino_t                   ino;
1344         ENTRY;
1345
1346         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1347         inode = iget_locked(sb, ino);
1348         if (inode == NULL) {
1349                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1350                        sbi->ll_fsname, PFID(fid));
1351                 RETURN(ERR_PTR(-ENOENT));
1352         }
1353
1354         if (inode->i_state & I_NEW) {
1355                 struct ll_inode_info *lli = ll_i2info(inode);
1356                 struct lmv_stripe_md *lsm = md->lmv;
1357
1358                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1359                                 (body->mbo_mode & S_IFMT);
1360                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1361                          PFID(fid));
1362
1363                 inode->i_mtime.tv_sec = 0;
1364                 inode->i_atime.tv_sec = 0;
1365                 inode->i_ctime.tv_sec = 0;
1366                 inode->i_rdev = 0;
1367
1368 #ifdef HAVE_BACKING_DEV_INFO
1369                 /* initializing backing dev info. */
1370                 inode->i_mapping->backing_dev_info =
1371                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1372 #endif
1373                 inode->i_op = &ll_dir_inode_operations;
1374                 inode->i_fop = &ll_dir_operations;
1375                 lli->lli_fid = *fid;
1376                 ll_lli_init(lli);
1377
1378                 LASSERT(lsm != NULL);
1379                 /* master object FID */
1380                 lli->lli_pfid = body->mbo_fid1;
1381                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1382                        lli, PFID(fid), PFID(&lli->lli_pfid));
1383                 unlock_new_inode(inode);
1384         }
1385
1386         RETURN(inode);
1387 }
1388
1389 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1390 {
1391         struct lu_fid *fid;
1392         struct lmv_stripe_md *lsm = md->lmv;
1393         struct ll_inode_info *lli = ll_i2info(inode);
1394         int i;
1395
1396         LASSERT(lsm != NULL);
1397
1398         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1399                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1400         lsm_md_dump(D_INODE, lsm);
1401
1402         if (!lmv_dir_striped(lsm))
1403                 goto out;
1404
1405         /* XXX sigh, this lsm_root initialization should be in
1406          * LMV layer, but it needs ll_iget right now, so we
1407          * put this here right now. */
1408         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1409                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1410                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1411
1412                 if (!fid_is_sane(fid))
1413                         continue;
1414
1415                 /* Unfortunately ll_iget will call ll_update_inode,
1416                  * where the initialization of slave inode is slightly
1417                  * different, so it reset lsm_md to NULL to avoid
1418                  * initializing lsm for slave inode. */
1419                 lsm->lsm_md_oinfo[i].lmo_root =
1420                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1421                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1422                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1423
1424                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1425                         while (i-- > 0) {
1426                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1427                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1428                         }
1429                         return rc;
1430                 }
1431         }
1432 out:
1433         lli->lli_lsm_md = lsm;
1434
1435         return 0;
1436 }
1437
1438 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1439 {
1440         struct ll_inode_info *lli = ll_i2info(inode);
1441
1442         if (!md->default_lmv) {
1443                 /* clear default lsm */
1444                 if (lli->lli_default_lsm_md) {
1445                         down_write(&lli->lli_lsm_sem);
1446                         if (lli->lli_default_lsm_md) {
1447                                 lmv_free_memmd(lli->lli_default_lsm_md);
1448                                 lli->lli_default_lsm_md = NULL;
1449                         }
1450                         up_write(&lli->lli_lsm_sem);
1451                 }
1452         } else if (lli->lli_default_lsm_md) {
1453                 /* update default lsm if it changes */
1454                 down_read(&lli->lli_lsm_sem);
1455                 if (lli->lli_default_lsm_md &&
1456                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1457                         up_read(&lli->lli_lsm_sem);
1458                         down_write(&lli->lli_lsm_sem);
1459                         if (lli->lli_default_lsm_md)
1460                                 lmv_free_memmd(lli->lli_default_lsm_md);
1461                         lli->lli_default_lsm_md = md->default_lmv;
1462                         lsm_md_dump(D_INODE, md->default_lmv);
1463                         md->default_lmv = NULL;
1464                         up_write(&lli->lli_lsm_sem);
1465                 } else {
1466                         up_read(&lli->lli_lsm_sem);
1467                 }
1468         } else {
1469                 /* init default lsm */
1470                 down_write(&lli->lli_lsm_sem);
1471                 lli->lli_default_lsm_md = md->default_lmv;
1472                 lsm_md_dump(D_INODE, md->default_lmv);
1473                 md->default_lmv = NULL;
1474                 up_write(&lli->lli_lsm_sem);
1475         }
1476 }
1477
1478 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1479 {
1480         struct ll_inode_info *lli = ll_i2info(inode);
1481         struct lmv_stripe_md *lsm = md->lmv;
1482         int rc = 0;
1483
1484         ENTRY;
1485
1486         LASSERT(S_ISDIR(inode->i_mode));
1487         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1488                PFID(ll_inode2fid(inode)));
1489
1490         /* update default LMV */
1491         if (md->default_lmv)
1492                 ll_update_default_lsm_md(inode, md);
1493
1494         /*
1495          * no striped information from request, lustre_md from req does not
1496          * include stripeEA, see ll_md_setattr()
1497          */
1498         if (!lsm)
1499                 RETURN(0);
1500
1501         /*
1502          * normally dir layout doesn't change, only take read lock to check
1503          * that to avoid blocking other MD operations.
1504          */
1505         if (lli->lli_lsm_md)
1506                 down_read(&lli->lli_lsm_sem);
1507         else
1508                 down_write(&lli->lli_lsm_sem);
1509
1510         /*
1511          * if dir layout mismatch, check whether version is increased, which
1512          * means layout is changed, this happens in dir migration and lfsck.
1513          *
1514          * foreign LMV should not change.
1515          */
1516         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1517                 if (lmv_dir_striped(lli->lli_lsm_md) &&
1518                     lsm->lsm_md_layout_version <=
1519                     lli->lli_lsm_md->lsm_md_layout_version) {
1520                         CERROR("%s: "DFID" dir layout mismatch:\n",
1521                                ll_i2sbi(inode)->ll_fsname,
1522                                PFID(&lli->lli_fid));
1523                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1524                         lsm_md_dump(D_ERROR, lsm);
1525                         GOTO(unlock, rc = -EINVAL);
1526                 }
1527
1528                 /* layout changed, switch to write lock */
1529                 up_read(&lli->lli_lsm_sem);
1530                 down_write(&lli->lli_lsm_sem);
1531                 ll_dir_clear_lsm_md(inode);
1532         }
1533
1534         /* set directory layout */
1535         if (!lli->lli_lsm_md) {
1536                 struct cl_attr  *attr;
1537
1538                 rc = ll_init_lsm_md(inode, md);
1539                 up_write(&lli->lli_lsm_sem);
1540                 if (rc != 0)
1541                         RETURN(rc);
1542
1543                 /* set md->lmv to NULL, so the following free lustre_md
1544                  * will not free this lsm */
1545                 md->lmv = NULL;
1546
1547                 /*
1548                  * md_merge_attr() may take long, since lsm is already set,
1549                  * switch to read lock.
1550                  */
1551                 down_read(&lli->lli_lsm_sem);
1552
1553                 if (!lmv_dir_striped(lli->lli_lsm_md))
1554                         GOTO(unlock, rc);
1555
1556                 OBD_ALLOC_PTR(attr);
1557                 if (attr == NULL)
1558                         GOTO(unlock, rc = -ENOMEM);
1559
1560                 /* validate the lsm */
1561                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1562                                    ll_md_blocking_ast);
1563                 if (rc != 0) {
1564                         OBD_FREE_PTR(attr);
1565                         GOTO(unlock, rc);
1566                 }
1567
1568                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1569                         md->body->mbo_nlink = attr->cat_nlink;
1570                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1571                         md->body->mbo_size = attr->cat_size;
1572                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1573                         md->body->mbo_atime = attr->cat_atime;
1574                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1575                         md->body->mbo_ctime = attr->cat_ctime;
1576                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1577                         md->body->mbo_mtime = attr->cat_mtime;
1578
1579                 OBD_FREE_PTR(attr);
1580         }
1581 unlock:
1582         up_read(&lli->lli_lsm_sem);
1583
1584         RETURN(rc);
1585 }
1586
1587 void ll_clear_inode(struct inode *inode)
1588 {
1589         struct ll_inode_info *lli = ll_i2info(inode);
1590         struct ll_sb_info *sbi = ll_i2sbi(inode);
1591
1592         ENTRY;
1593
1594         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1595                PFID(ll_inode2fid(inode)), inode);
1596
1597         if (S_ISDIR(inode->i_mode)) {
1598                 /* these should have been cleared in ll_file_release */
1599                 LASSERT(lli->lli_opendir_key == NULL);
1600                 LASSERT(lli->lli_sai == NULL);
1601                 LASSERT(lli->lli_opendir_pid == 0);
1602         } else {
1603                 pcc_inode_free(inode);
1604         }
1605
1606         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1607
1608         LASSERT(!lli->lli_open_fd_write_count);
1609         LASSERT(!lli->lli_open_fd_read_count);
1610         LASSERT(!lli->lli_open_fd_exec_count);
1611
1612         if (lli->lli_mds_write_och)
1613                 ll_md_real_close(inode, FMODE_WRITE);
1614         if (lli->lli_mds_exec_och)
1615                 ll_md_real_close(inode, FMODE_EXEC);
1616         if (lli->lli_mds_read_och)
1617                 ll_md_real_close(inode, FMODE_READ);
1618
1619         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1620                 OBD_FREE(lli->lli_symlink_name,
1621                          strlen(lli->lli_symlink_name) + 1);
1622                 lli->lli_symlink_name = NULL;
1623         }
1624
1625         ll_xattr_cache_destroy(inode);
1626
1627 #ifdef CONFIG_FS_POSIX_ACL
1628         forget_all_cached_acls(inode);
1629         if (lli->lli_posix_acl) {
1630                 posix_acl_release(lli->lli_posix_acl);
1631                 lli->lli_posix_acl = NULL;
1632         }
1633 #endif
1634         lli->lli_inode_magic = LLI_INODE_DEAD;
1635
1636         if (S_ISDIR(inode->i_mode))
1637                 ll_dir_clear_lsm_md(inode);
1638         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1639                 LASSERT(list_empty(&lli->lli_agl_list));
1640
1641         /*
1642          * XXX This has to be done before lsm is freed below, because
1643          * cl_object still uses inode lsm.
1644          */
1645         cl_inode_fini(inode);
1646
1647         EXIT;
1648 }
1649
1650 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1651 {
1652         struct lustre_md md;
1653         struct inode *inode = dentry->d_inode;
1654         struct ll_sb_info *sbi = ll_i2sbi(inode);
1655         struct ptlrpc_request *request = NULL;
1656         int rc, ia_valid;
1657         ENTRY;
1658
1659         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1660                                      LUSTRE_OPC_ANY, NULL);
1661         if (IS_ERR(op_data))
1662                 RETURN(PTR_ERR(op_data));
1663
1664         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1665         if (rc) {
1666                 ptlrpc_req_finished(request);
1667                 if (rc == -ENOENT) {
1668                         clear_nlink(inode);
1669                         /* Unlinked special device node? Or just a race?
1670                          * Pretend we done everything. */
1671                         if (!S_ISREG(inode->i_mode) &&
1672                             !S_ISDIR(inode->i_mode)) {
1673                                 ia_valid = op_data->op_attr.ia_valid;
1674                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1675                                 rc = simple_setattr(dentry, &op_data->op_attr);
1676                                 op_data->op_attr.ia_valid = ia_valid;
1677                         }
1678                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1679                         CERROR("md_setattr fails: rc = %d\n", rc);
1680                 }
1681                 RETURN(rc);
1682         }
1683
1684         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1685                               sbi->ll_md_exp, &md);
1686         if (rc) {
1687                 ptlrpc_req_finished(request);
1688                 RETURN(rc);
1689         }
1690
1691         ia_valid = op_data->op_attr.ia_valid;
1692         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1693          * cache is not cleared yet. */
1694         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1695         if (S_ISREG(inode->i_mode))
1696                 inode_lock(inode);
1697         rc = simple_setattr(dentry, &op_data->op_attr);
1698         if (S_ISREG(inode->i_mode))
1699                 inode_unlock(inode);
1700         op_data->op_attr.ia_valid = ia_valid;
1701
1702         rc = ll_update_inode(inode, &md);
1703         ptlrpc_req_finished(request);
1704
1705         RETURN(rc);
1706 }
1707
1708 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1709  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1710  * keep these values until such a time that objects are allocated for it.
1711  * We do the MDS operations first, as it is checking permissions for us.
1712  * We don't to the MDS RPC if there is nothing that we want to store there,
1713  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1714  * going to do an RPC anyways.
1715  *
1716  * If we are doing a truncate, we will send the mtime and ctime updates
1717  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1718  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1719  * at the same time.
1720  *
1721  * In case of HSMimport, we only set attr on MDS.
1722  */
1723 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1724                    enum op_xvalid xvalid, bool hsm_import)
1725 {
1726         struct inode *inode = dentry->d_inode;
1727         struct ll_inode_info *lli = ll_i2info(inode);
1728         struct md_op_data *op_data = NULL;
1729         int rc = 0;
1730
1731         ENTRY;
1732
1733         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1734                "valid %x, hsm_import %d\n",
1735                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1736                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1737                hsm_import);
1738
1739         if (attr->ia_valid & ATTR_SIZE) {
1740                 /* Check new size against VFS/VM file size limit and rlimit */
1741                 rc = inode_newsize_ok(inode, attr->ia_size);
1742                 if (rc)
1743                         RETURN(rc);
1744
1745                 /* The maximum Lustre file size is variable, based on the
1746                  * OST maximum object size and number of stripes.  This
1747                  * needs another check in addition to the VFS check above. */
1748                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1749                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1750                                PFID(&lli->lli_fid), attr->ia_size,
1751                                ll_file_maxbytes(inode));
1752                         RETURN(-EFBIG);
1753                 }
1754
1755                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1756         }
1757
1758         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1759         if (attr->ia_valid & TIMES_SET_FLAGS) {
1760                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1761                     !cfs_capable(CFS_CAP_FOWNER))
1762                         RETURN(-EPERM);
1763         }
1764
1765         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1766         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1767              (attr->ia_valid & ATTR_CTIME)) {
1768                 attr->ia_ctime = current_time(inode);
1769                 xvalid |= OP_XVALID_CTIME_SET;
1770         }
1771         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1772             (attr->ia_valid & ATTR_ATIME)) {
1773                 attr->ia_atime = current_time(inode);
1774                 attr->ia_valid |= ATTR_ATIME_SET;
1775         }
1776         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1777             (attr->ia_valid & ATTR_MTIME)) {
1778                 attr->ia_mtime = current_time(inode);
1779                 attr->ia_valid |= ATTR_MTIME_SET;
1780         }
1781
1782         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1783                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1784                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1785                        ktime_get_real_seconds());
1786
1787         if (S_ISREG(inode->i_mode)) {
1788                 if (attr->ia_valid & ATTR_SIZE)
1789                         inode_dio_write_done(inode);
1790                 inode_unlock(inode);
1791         }
1792
1793         /* We always do an MDS RPC, even if we're only changing the size;
1794          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1795
1796         OBD_ALLOC_PTR(op_data);
1797         if (op_data == NULL)
1798                 GOTO(out, rc = -ENOMEM);
1799
1800         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1801                 /* If we are changing file size, file content is
1802                  * modified, flag it.
1803                  */
1804                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1805                 op_data->op_bias |= MDS_DATA_MODIFIED;
1806                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1807         }
1808
1809         if (attr->ia_valid & ATTR_FILE) {
1810                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1811
1812                 if (fd->fd_lease_och)
1813                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1814         }
1815
1816         op_data->op_attr = *attr;
1817         op_data->op_xvalid = xvalid;
1818
1819         rc = ll_md_setattr(dentry, op_data);
1820         if (rc)
1821                 GOTO(out, rc);
1822
1823         if (!S_ISREG(inode->i_mode) || hsm_import)
1824                 GOTO(out, rc = 0);
1825
1826         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1827                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1828             xvalid & OP_XVALID_CTIME_SET) {
1829                 bool cached = false;
1830
1831                 rc = pcc_inode_setattr(inode, attr, &cached);
1832                 if (cached) {
1833                         if (rc) {
1834                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1835                                        "rc = %d\n",
1836                                        ll_i2sbi(inode)->ll_fsname,
1837                                        PFID(&lli->lli_fid), rc);
1838                                 GOTO(out, rc);
1839                         }
1840                 } else {
1841                         /* For truncate and utimes sending attributes to OSTs,
1842                          * setting mtime/atime to the past will be performed
1843                          * under PW [0:EOF] extent lock (new_size:EOF for
1844                          * truncate). It may seem excessive to send mtime/atime
1845                          * updates to OSTs when not setting times to past, but
1846                          * it is necessary due to possible time
1847                          * de-synchronization between MDT inode and OST objects
1848                          */
1849                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1850                 }
1851         }
1852
1853         /* If the file was restored, it needs to set dirty flag.
1854          *
1855          * We've already sent MDS_DATA_MODIFIED flag in
1856          * ll_md_setattr() for truncate. However, the MDT refuses to
1857          * set the HS_DIRTY flag on released files, so we have to set
1858          * it again if the file has been restored. Please check how
1859          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1860          *
1861          * Please notice that if the file is not released, the previous
1862          * MDS_DATA_MODIFIED has taken effect and usually
1863          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1864          * This way we can save an RPC for common open + trunc
1865          * operation. */
1866         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1867                 struct hsm_state_set hss = {
1868                         .hss_valid = HSS_SETMASK,
1869                         .hss_setmask = HS_DIRTY,
1870                 };
1871                 int rc2;
1872
1873                 rc2 = ll_hsm_state_set(inode, &hss);
1874                 /* truncate and write can happen at the same time, so that
1875                  * the file can be set modified even though the file is not
1876                  * restored from released state, and ll_hsm_state_set() is
1877                  * not applicable for the file, and rc2 < 0 is normal in this
1878                  * case. */
1879                 if (rc2 < 0)
1880                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1881                                PFID(ll_inode2fid(inode)), rc2);
1882         }
1883
1884         EXIT;
1885 out:
1886         if (op_data != NULL)
1887                 ll_finish_md_op_data(op_data);
1888
1889         if (S_ISREG(inode->i_mode)) {
1890                 inode_lock(inode);
1891                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1892                         inode_dio_wait(inode);
1893                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1894                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1895                  * inode flags, so there is a gap where S_NOSEC is not set.
1896                  * This can cause a writer to take the i_mutex unnecessarily,
1897                  * but this is safe to do and should be rare. */
1898                 inode_has_no_xattr(inode);
1899         }
1900
1901         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1902                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1903
1904         return rc;
1905 }
1906
1907 int ll_setattr(struct dentry *de, struct iattr *attr)
1908 {
1909         int mode = de->d_inode->i_mode;
1910         enum op_xvalid xvalid = 0;
1911
1912         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1913                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1914                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1915
1916         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1917                                (ATTR_SIZE|ATTR_MODE)) &&
1918             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1919              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1920               !(attr->ia_mode & S_ISGID))))
1921                 attr->ia_valid |= ATTR_FORCE;
1922
1923         if ((attr->ia_valid & ATTR_MODE) &&
1924             (mode & S_ISUID) &&
1925             !(attr->ia_mode & S_ISUID) &&
1926             !(attr->ia_valid & ATTR_KILL_SUID))
1927                 attr->ia_valid |= ATTR_KILL_SUID;
1928
1929         if ((attr->ia_valid & ATTR_MODE) &&
1930             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1931             !(attr->ia_mode & S_ISGID) &&
1932             !(attr->ia_valid & ATTR_KILL_SGID))
1933                 attr->ia_valid |= ATTR_KILL_SGID;
1934
1935         return ll_setattr_raw(de, attr, xvalid, false);
1936 }
1937
1938 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1939                        u32 flags)
1940 {
1941         struct obd_statfs obd_osfs = { 0 };
1942         time64_t max_age;
1943         int rc;
1944
1945         ENTRY;
1946         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
1947
1948         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1949         if (rc)
1950                 RETURN(rc);
1951
1952         osfs->os_type = LL_SUPER_MAGIC;
1953
1954         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1955               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1956
1957         if (osfs->os_state & OS_STATE_SUM)
1958                 GOTO(out, rc);
1959
1960         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1961                 flags |= OBD_STATFS_NODELAY;
1962
1963         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1964         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1965                 GOTO(out, rc = 0);
1966
1967         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1968                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1969                obd_osfs.os_files);
1970
1971         osfs->os_bsize = obd_osfs.os_bsize;
1972         osfs->os_blocks = obd_osfs.os_blocks;
1973         osfs->os_bfree = obd_osfs.os_bfree;
1974         osfs->os_bavail = obd_osfs.os_bavail;
1975
1976         /* If we have _some_ OSTs, but don't have as many free objects on the
1977          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1978          * to compensate, so that the "inodes in use" number is correct.
1979          * This should be kept in sync with lod_statfs() behaviour.
1980          */
1981         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1982                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1983                                  obd_osfs.os_ffree;
1984                 osfs->os_ffree = obd_osfs.os_ffree;
1985         }
1986
1987 out:
1988         RETURN(rc);
1989 }
1990 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1991 {
1992         struct super_block *sb = de->d_sb;
1993         struct obd_statfs osfs;
1994         __u64 fsid = huge_encode_dev(sb->s_dev);
1995         int rc;
1996
1997         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1998         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS, 1);
1999
2000         /* Some amount of caching on the client is allowed */
2001         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2002         if (rc)
2003                 return rc;
2004
2005         statfs_unpack(sfs, &osfs);
2006
2007         /* We need to downshift for all 32-bit kernels, because we can't
2008          * tell if the kernel is being called via sys_statfs64() or not.
2009          * Stop before overflowing f_bsize - in which case it is better
2010          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2011         if (sizeof(long) < 8) {
2012                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2013                         sfs->f_bsize <<= 1;
2014
2015                         osfs.os_blocks >>= 1;
2016                         osfs.os_bfree >>= 1;
2017                         osfs.os_bavail >>= 1;
2018                 }
2019         }
2020
2021         sfs->f_blocks = osfs.os_blocks;
2022         sfs->f_bfree = osfs.os_bfree;
2023         sfs->f_bavail = osfs.os_bavail;
2024         sfs->f_fsid.val[0] = (__u32)fsid;
2025         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2026         return 0;
2027 }
2028
2029 void ll_inode_size_lock(struct inode *inode)
2030 {
2031         struct ll_inode_info *lli;
2032
2033         LASSERT(!S_ISDIR(inode->i_mode));
2034
2035         lli = ll_i2info(inode);
2036         mutex_lock(&lli->lli_size_mutex);
2037 }
2038
2039 void ll_inode_size_unlock(struct inode *inode)
2040 {
2041         struct ll_inode_info *lli;
2042
2043         lli = ll_i2info(inode);
2044         mutex_unlock(&lli->lli_size_mutex);
2045 }
2046
2047 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2048 {
2049         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2050         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2051                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2052         else
2053                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2054 }
2055
2056 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2057 {
2058         struct ll_inode_info *lli = ll_i2info(inode);
2059         struct mdt_body *body = md->body;
2060         struct ll_sb_info *sbi = ll_i2sbi(inode);
2061         int rc = 0;
2062
2063         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2064                 rc = cl_file_inode_init(inode, md);
2065                 if (rc)
2066                         return rc;
2067         }
2068
2069         if (S_ISDIR(inode->i_mode)) {
2070                 rc = ll_update_lsm_md(inode, md);
2071                 if (rc != 0)
2072                         return rc;
2073         }
2074
2075 #ifdef CONFIG_FS_POSIX_ACL
2076         if (body->mbo_valid & OBD_MD_FLACL) {
2077                 spin_lock(&lli->lli_lock);
2078                 if (lli->lli_posix_acl)
2079                         posix_acl_release(lli->lli_posix_acl);
2080                 lli->lli_posix_acl = md->posix_acl;
2081                 spin_unlock(&lli->lli_lock);
2082         }
2083 #endif
2084         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2085                                         sbi->ll_flags & LL_SBI_32BIT_API);
2086         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2087
2088         if (body->mbo_valid & OBD_MD_FLATIME) {
2089                 if (body->mbo_atime > inode->i_atime.tv_sec)
2090                         inode->i_atime.tv_sec = body->mbo_atime;
2091                 lli->lli_atime = body->mbo_atime;
2092         }
2093
2094         if (body->mbo_valid & OBD_MD_FLMTIME) {
2095                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2096                         CDEBUG(D_INODE,
2097                                "setting ino %lu mtime from %lld to %llu\n",
2098                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2099                                body->mbo_mtime);
2100                         inode->i_mtime.tv_sec = body->mbo_mtime;
2101                 }
2102                 lli->lli_mtime = body->mbo_mtime;
2103         }
2104
2105         if (body->mbo_valid & OBD_MD_FLCTIME) {
2106                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2107                         inode->i_ctime.tv_sec = body->mbo_ctime;
2108                 lli->lli_ctime = body->mbo_ctime;
2109         }
2110
2111         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2112         if (body->mbo_valid & OBD_MD_FLFLAGS)
2113                 ll_update_inode_flags(inode, body->mbo_flags);
2114         if (body->mbo_valid & OBD_MD_FLMODE)
2115                 inode->i_mode = (inode->i_mode & S_IFMT) |
2116                                 (body->mbo_mode & ~S_IFMT);
2117
2118         if (body->mbo_valid & OBD_MD_FLTYPE)
2119                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2120                                 (body->mbo_mode & S_IFMT);
2121
2122         LASSERT(inode->i_mode != 0);
2123         if (S_ISREG(inode->i_mode))
2124                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2125                                        LL_MAX_BLKSIZE_BITS);
2126         else
2127                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2128
2129         if (body->mbo_valid & OBD_MD_FLUID)
2130                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2131         if (body->mbo_valid & OBD_MD_FLGID)
2132                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2133         if (body->mbo_valid & OBD_MD_FLPROJID)
2134                 lli->lli_projid = body->mbo_projid;
2135         if (body->mbo_valid & OBD_MD_FLNLINK)
2136                 set_nlink(inode, body->mbo_nlink);
2137         if (body->mbo_valid & OBD_MD_FLRDEV)
2138                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2139
2140         if (body->mbo_valid & OBD_MD_FLID) {
2141                 /* FID shouldn't be changed! */
2142                 if (fid_is_sane(&lli->lli_fid)) {
2143                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2144                                  "Trying to change FID "DFID
2145                                  " to the "DFID", inode "DFID"(%p)\n",
2146                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2147                                  PFID(ll_inode2fid(inode)), inode);
2148                 } else {
2149                         lli->lli_fid = body->mbo_fid1;
2150                 }
2151         }
2152
2153         LASSERT(fid_seq(&lli->lli_fid) != 0);
2154
2155         if (body->mbo_valid & OBD_MD_FLSIZE) {
2156                 i_size_write(inode, body->mbo_size);
2157
2158                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2159                        PFID(ll_inode2fid(inode)),
2160                        (unsigned long long)body->mbo_size);
2161
2162                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2163                         inode->i_blocks = body->mbo_blocks;
2164         }
2165
2166         if (body->mbo_valid & OBD_MD_TSTATE) {
2167                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2168                  * clear it when done to ensure to start again
2169                  * glimpsing updated attrs
2170                  */
2171                 if (body->mbo_t_state & MS_RESTORE)
2172                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2173                 else
2174                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2175         }
2176
2177         return 0;
2178 }
2179
2180 int ll_read_inode2(struct inode *inode, void *opaque)
2181 {
2182         struct lustre_md *md = opaque;
2183         struct ll_inode_info *lli = ll_i2info(inode);
2184         int     rc;
2185         ENTRY;
2186
2187         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2188                PFID(&lli->lli_fid), inode);
2189
2190         /* Core attributes from the MDS first.  This is a new inode, and
2191          * the VFS doesn't zero times in the core inode so we have to do
2192          * it ourselves.  They will be overwritten by either MDS or OST
2193          * attributes - we just need to make sure they aren't newer.
2194          */
2195         inode->i_mtime.tv_sec = 0;
2196         inode->i_atime.tv_sec = 0;
2197         inode->i_ctime.tv_sec = 0;
2198         inode->i_rdev = 0;
2199         rc = ll_update_inode(inode, md);
2200         if (rc != 0)
2201                 RETURN(rc);
2202
2203         /* OIDEBUG(inode); */
2204
2205 #ifdef HAVE_BACKING_DEV_INFO
2206         /* initializing backing dev info. */
2207         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2208 #endif
2209         if (S_ISREG(inode->i_mode)) {
2210                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2211                 inode->i_op = &ll_file_inode_operations;
2212                 inode->i_fop = sbi->ll_fop;
2213                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2214                 EXIT;
2215         } else if (S_ISDIR(inode->i_mode)) {
2216                 inode->i_op = &ll_dir_inode_operations;
2217                 inode->i_fop = &ll_dir_operations;
2218                 EXIT;
2219         } else if (S_ISLNK(inode->i_mode)) {
2220                 inode->i_op = &ll_fast_symlink_inode_operations;
2221                 EXIT;
2222         } else {
2223                 inode->i_op = &ll_special_inode_operations;
2224
2225                 init_special_inode(inode, inode->i_mode,
2226                                    inode->i_rdev);
2227
2228                 EXIT;
2229         }
2230
2231         return 0;
2232 }
2233
2234 void ll_delete_inode(struct inode *inode)
2235 {
2236         struct ll_inode_info *lli = ll_i2info(inode);
2237         struct address_space *mapping = &inode->i_data;
2238         unsigned long nrpages;
2239         ENTRY;
2240
2241         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2242                 /* It is last chance to write out dirty pages,
2243                  * otherwise we may lose data while umount.
2244                  *
2245                  * If i_nlink is 0 then just discard data. This is safe because
2246                  * local inode gets i_nlink 0 from server only for the last
2247                  * unlink, so that file is not opened somewhere else
2248                  */
2249                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2250                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2251         }
2252         truncate_inode_pages_final(mapping);
2253
2254         /* Workaround for LU-118: Note nrpages may not be totally updated when
2255          * truncate_inode_pages() returns, as there can be a page in the process
2256          * of deletion (inside __delete_from_page_cache()) in the specified
2257          * range. Thus mapping->nrpages can be non-zero when this function
2258          * returns even after truncation of the whole mapping.  Only do this if
2259          * npages isn't already zero.
2260          */
2261         nrpages = mapping->nrpages;
2262         if (nrpages) {
2263                 xa_lock_irq(&mapping->i_pages);
2264                 nrpages = mapping->nrpages;
2265                 xa_unlock_irq(&mapping->i_pages);
2266         } /* Workaround end */
2267
2268         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2269                  "see https://jira.whamcloud.com/browse/LU-118\n",
2270                  ll_i2sbi(inode)->ll_fsname,
2271                  PFID(ll_inode2fid(inode)), inode, nrpages);
2272
2273 #ifdef HAVE_SBOPS_EVICT_INODE
2274         ll_clear_inode(inode);
2275 #endif
2276         clear_inode(inode);
2277
2278         EXIT;
2279 }
2280
2281 int ll_iocontrol(struct inode *inode, struct file *file,
2282                  unsigned int cmd, unsigned long arg)
2283 {
2284         struct ll_sb_info *sbi = ll_i2sbi(inode);
2285         struct ptlrpc_request *req = NULL;
2286         int rc, flags = 0;
2287         ENTRY;
2288
2289         switch (cmd) {
2290         case FS_IOC_GETFLAGS: {
2291                 struct mdt_body *body;
2292                 struct md_op_data *op_data;
2293
2294                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2295                                              0, 0, LUSTRE_OPC_ANY,
2296                                              NULL);
2297                 if (IS_ERR(op_data))
2298                         RETURN(PTR_ERR(op_data));
2299
2300                 op_data->op_valid = OBD_MD_FLFLAGS;
2301                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2302                 ll_finish_md_op_data(op_data);
2303                 if (rc) {
2304                         CERROR("%s: failure inode "DFID": rc = %d\n",
2305                                sbi->ll_md_exp->exp_obd->obd_name,
2306                                PFID(ll_inode2fid(inode)), rc);
2307                         RETURN(-abs(rc));
2308                 }
2309
2310                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2311
2312                 flags = body->mbo_flags;
2313
2314                 ptlrpc_req_finished(req);
2315
2316                 RETURN(put_user(flags, (int __user *)arg));
2317         }
2318         case FS_IOC_SETFLAGS: {
2319                 struct iattr *attr;
2320                 struct md_op_data *op_data;
2321                 struct cl_object *obj;
2322                 struct fsxattr fa = { 0 };
2323
2324                 if (get_user(flags, (int __user *)arg))
2325                         RETURN(-EFAULT);
2326
2327                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2328                 if (flags & LUSTRE_PROJINHERIT_FL)
2329                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2330
2331                 rc = ll_ioctl_check_project(inode, &fa);
2332                 if (rc)
2333                         RETURN(rc);
2334
2335                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2336                                              LUSTRE_OPC_ANY, NULL);
2337                 if (IS_ERR(op_data))
2338                         RETURN(PTR_ERR(op_data));
2339
2340                 op_data->op_attr_flags = flags;
2341                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2342                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2343                 ll_finish_md_op_data(op_data);
2344                 ptlrpc_req_finished(req);
2345                 if (rc)
2346                         RETURN(rc);
2347
2348                 ll_update_inode_flags(inode, flags);
2349
2350                 obj = ll_i2info(inode)->lli_clob;
2351                 if (obj == NULL)
2352                         RETURN(0);
2353
2354                 OBD_ALLOC_PTR(attr);
2355                 if (attr == NULL)
2356                         RETURN(-ENOMEM);
2357
2358                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2359
2360                 OBD_FREE_PTR(attr);
2361                 RETURN(rc);
2362         }
2363         default:
2364                 RETURN(-ENOSYS);
2365         }
2366
2367         RETURN(0);
2368 }
2369
2370 int ll_flush_ctx(struct inode *inode)
2371 {
2372         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2373
2374         CDEBUG(D_SEC, "flush context for user %d\n",
2375                from_kuid(&init_user_ns, current_uid()));
2376
2377         obd_set_info_async(NULL, sbi->ll_md_exp,
2378                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2379                            0, NULL, NULL);
2380         obd_set_info_async(NULL, sbi->ll_dt_exp,
2381                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2382                            0, NULL, NULL);
2383         return 0;
2384 }
2385
2386 /* umount -f client means force down, don't save state */
2387 void ll_umount_begin(struct super_block *sb)
2388 {
2389         struct ll_sb_info *sbi = ll_s2sbi(sb);
2390         struct obd_device *obd;
2391         struct obd_ioctl_data *ioc_data;
2392         struct l_wait_info lwi;
2393         wait_queue_head_t waitq;
2394         ENTRY;
2395
2396         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2397                sb->s_count, atomic_read(&sb->s_active));
2398
2399         obd = class_exp2obd(sbi->ll_md_exp);
2400         if (obd == NULL) {
2401                 CERROR("Invalid MDC connection handle %#llx\n",
2402                        sbi->ll_md_exp->exp_handle.h_cookie);
2403                 EXIT;
2404                 return;
2405         }
2406         obd->obd_force = 1;
2407
2408         obd = class_exp2obd(sbi->ll_dt_exp);
2409         if (obd == NULL) {
2410                 CERROR("Invalid LOV connection handle %#llx\n",
2411                        sbi->ll_dt_exp->exp_handle.h_cookie);
2412                 EXIT;
2413                 return;
2414         }
2415         obd->obd_force = 1;
2416
2417         OBD_ALLOC_PTR(ioc_data);
2418         if (ioc_data) {
2419                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2420                               sizeof *ioc_data, ioc_data, NULL);
2421
2422                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2423                               sizeof *ioc_data, ioc_data, NULL);
2424
2425                 OBD_FREE_PTR(ioc_data);
2426         }
2427
2428         /* Really, we'd like to wait until there are no requests outstanding,
2429          * and then continue.  For now, we just periodically checking for vfs
2430          * to decrement mnt_cnt and hope to finish it within 10sec.
2431          */
2432         init_waitqueue_head(&waitq);
2433         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2434                                    cfs_time_seconds(1), NULL, NULL);
2435         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2436
2437         EXIT;
2438 }
2439
2440 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2441 {
2442         struct ll_sb_info *sbi = ll_s2sbi(sb);
2443         char *profilenm = get_profile_name(sb);
2444         int err;
2445         __u32 read_only;
2446
2447         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2448                 read_only = *flags & MS_RDONLY;
2449                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2450                                          sizeof(KEY_READ_ONLY),
2451                                          KEY_READ_ONLY, sizeof(read_only),
2452                                          &read_only, NULL);
2453                 if (err) {
2454                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2455                                       profilenm, read_only ?
2456                                       "read-only" : "read-write", err);
2457                         return err;
2458                 }
2459
2460                 if (read_only)
2461                         sb->s_flags |= SB_RDONLY;
2462                 else
2463                         sb->s_flags &= ~SB_RDONLY;
2464
2465                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2466                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2467                                       read_only ?  "read-only" : "read-write");
2468         }
2469         return 0;
2470 }
2471
2472 /**
2473  * Cleanup the open handle that is cached on MDT-side.
2474  *
2475  * For open case, the client side open handling thread may hit error
2476  * after the MDT grant the open. Under such case, the client should
2477  * send close RPC to the MDT as cleanup; otherwise, the open handle
2478  * on the MDT will be leaked there until the client umount or evicted.
2479  *
2480  * In further, if someone unlinked the file, because the open handle
2481  * holds the reference on such file/object, then it will block the
2482  * subsequent threads that want to locate such object via FID.
2483  *
2484  * \param[in] sb        super block for this file-system
2485  * \param[in] open_req  pointer to the original open request
2486  */
2487 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2488 {
2489         struct mdt_body                 *body;
2490         struct md_op_data               *op_data;
2491         struct ptlrpc_request           *close_req = NULL;
2492         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2493         ENTRY;
2494
2495         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2496         OBD_ALLOC_PTR(op_data);
2497         if (op_data == NULL) {
2498                 CWARN("%s: cannot allocate op_data to release open handle for "
2499                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2500
2501                 RETURN_EXIT;
2502         }
2503
2504         op_data->op_fid1 = body->mbo_fid1;
2505         op_data->op_open_handle = body->mbo_open_handle;
2506         op_data->op_mod_time = ktime_get_real_seconds();
2507         md_close(exp, op_data, NULL, &close_req);
2508         ptlrpc_req_finished(close_req);
2509         ll_finish_md_op_data(op_data);
2510
2511         EXIT;
2512 }
2513
2514 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2515                   struct super_block *sb, struct lookup_intent *it)
2516 {
2517         struct ll_sb_info *sbi = NULL;
2518         struct lustre_md md = { NULL };
2519         bool default_lmv_deleted = false;
2520         int rc;
2521
2522         ENTRY;
2523
2524         LASSERT(*inode || sb);
2525         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2526         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2527                               sbi->ll_md_exp, &md);
2528         if (rc != 0)
2529                 GOTO(out, rc);
2530
2531         /*
2532          * clear default_lmv only if intent_getattr reply doesn't contain it.
2533          * but it needs to be done after iget, check this early because
2534          * ll_update_lsm_md() may change md.
2535          */
2536         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2537             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2538                 default_lmv_deleted = true;
2539
2540         if (*inode) {
2541                 rc = ll_update_inode(*inode, &md);
2542                 if (rc != 0)
2543                         GOTO(out, rc);
2544         } else {
2545                 LASSERT(sb != NULL);
2546
2547                 /*
2548                  * At this point server returns to client's same fid as client
2549                  * generated for creating. So using ->fid1 is okay here.
2550                  */
2551                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2552                         CERROR("%s: Fid is insane "DFID"\n",
2553                                 sbi->ll_fsname,
2554                                 PFID(&md.body->mbo_fid1));
2555                         GOTO(out, rc = -EINVAL);
2556                 }
2557
2558                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2559                                              sbi->ll_flags & LL_SBI_32BIT_API),
2560                                  &md);
2561                 if (IS_ERR(*inode)) {
2562 #ifdef CONFIG_FS_POSIX_ACL
2563                         if (md.posix_acl) {
2564                                 posix_acl_release(md.posix_acl);
2565                                 md.posix_acl = NULL;
2566                         }
2567 #endif
2568                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2569                         *inode = NULL;
2570                         CERROR("new_inode -fatal: rc %d\n", rc);
2571                         GOTO(out, rc);
2572                 }
2573         }
2574
2575         /* Handling piggyback layout lock.
2576          * Layout lock can be piggybacked by getattr and open request.
2577          * The lsm can be applied to inode only if it comes with a layout lock
2578          * otherwise correct layout may be overwritten, for example:
2579          * 1. proc1: mdt returns a lsm but not granting layout
2580          * 2. layout was changed by another client
2581          * 3. proc2: refresh layout and layout lock granted
2582          * 4. proc1: to apply a stale layout */
2583         if (it != NULL && it->it_lock_mode != 0) {
2584                 struct lustre_handle lockh;
2585                 struct ldlm_lock *lock;
2586
2587                 lockh.cookie = it->it_lock_handle;
2588                 lock = ldlm_handle2lock(&lockh);
2589                 LASSERT(lock != NULL);
2590                 if (ldlm_has_layout(lock)) {
2591                         struct cl_object_conf conf;
2592
2593                         memset(&conf, 0, sizeof(conf));
2594                         conf.coc_opc = OBJECT_CONF_SET;
2595                         conf.coc_inode = *inode;
2596                         conf.coc_lock = lock;
2597                         conf.u.coc_layout = md.layout;
2598                         (void)ll_layout_conf(*inode, &conf);
2599                 }
2600                 LDLM_LOCK_PUT(lock);
2601         }
2602
2603         if (default_lmv_deleted)
2604                 ll_update_default_lsm_md(*inode, &md);
2605
2606         GOTO(out, rc = 0);
2607
2608 out:
2609         /* cleanup will be done if necessary */
2610         md_free_lustre_md(sbi->ll_md_exp, &md);
2611
2612         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2613                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2614
2615         return rc;
2616 }
2617
2618 int ll_obd_statfs(struct inode *inode, void __user *arg)
2619 {
2620         struct ll_sb_info *sbi = NULL;
2621         struct obd_export *exp;
2622         char *buf = NULL;
2623         struct obd_ioctl_data *data = NULL;
2624         __u32 type;
2625         int len = 0, rc;
2626
2627         if (!inode || !(sbi = ll_i2sbi(inode)))
2628                 GOTO(out_statfs, rc = -EINVAL);
2629
2630         rc = obd_ioctl_getdata(&buf, &len, arg);
2631         if (rc)
2632                 GOTO(out_statfs, rc);
2633
2634         data = (void*)buf;
2635         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2636             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2637                 GOTO(out_statfs, rc = -EINVAL);
2638
2639         if (data->ioc_inllen1 != sizeof(__u32) ||
2640             data->ioc_inllen2 != sizeof(__u32) ||
2641             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2642             data->ioc_plen2 != sizeof(struct obd_uuid))
2643                 GOTO(out_statfs, rc = -EINVAL);
2644
2645         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2646         if (type & LL_STATFS_LMV)
2647                 exp = sbi->ll_md_exp;
2648         else if (type & LL_STATFS_LOV)
2649                 exp = sbi->ll_dt_exp;
2650         else
2651                 GOTO(out_statfs, rc = -ENODEV);
2652
2653         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2654         if (rc)
2655                 GOTO(out_statfs, rc);
2656 out_statfs:
2657         OBD_FREE_LARGE(buf, len);
2658         return rc;
2659 }
2660
2661 /*
2662  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2663  * be called early to avoid deadlock.
2664  */
2665 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2666 {
2667         if (op_data->op_mea2_sem) {
2668                 up_read(op_data->op_mea2_sem);
2669                 op_data->op_mea2_sem = NULL;
2670         }
2671
2672         if (op_data->op_mea1_sem) {
2673                 up_read(op_data->op_mea1_sem);
2674                 op_data->op_mea1_sem = NULL;
2675         }
2676 }
2677
2678 /* this function prepares md_op_data hint for passing it down to MD stack. */
2679 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2680                                       struct inode *i1, struct inode *i2,
2681                                       const char *name, size_t namelen,
2682                                       __u32 mode, enum md_op_code opc,
2683                                       void *data)
2684 {
2685         LASSERT(i1 != NULL);
2686
2687         if (name == NULL) {
2688                 /* Do not reuse namelen for something else. */
2689                 if (namelen != 0)
2690                         return ERR_PTR(-EINVAL);
2691         } else {
2692                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2693                         return ERR_PTR(-ENAMETOOLONG);
2694
2695                 if (!lu_name_is_valid_2(name, namelen))
2696                         return ERR_PTR(-EINVAL);
2697         }
2698
2699         if (op_data == NULL)
2700                 OBD_ALLOC_PTR(op_data);
2701
2702         if (op_data == NULL)
2703                 return ERR_PTR(-ENOMEM);
2704
2705         ll_i2gids(op_data->op_suppgids, i1, i2);
2706         op_data->op_fid1 = *ll_inode2fid(i1);
2707         op_data->op_code = opc;
2708
2709         if (S_ISDIR(i1->i_mode)) {
2710                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2711                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2712                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2713                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2714         }
2715
2716         if (i2) {
2717                 op_data->op_fid2 = *ll_inode2fid(i2);
2718                 if (S_ISDIR(i2->i_mode)) {
2719                         if (i2 != i1) {
2720                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2721                                 op_data->op_mea2_sem =
2722                                                 &ll_i2info(i2)->lli_lsm_sem;
2723                         }
2724                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2725                 }
2726         } else {
2727                 fid_zero(&op_data->op_fid2);
2728         }
2729
2730         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2731                 op_data->op_cli_flags |= CLI_HASH64;
2732
2733         if (ll_need_32bit_api(ll_i2sbi(i1)))
2734                 op_data->op_cli_flags |= CLI_API32;
2735
2736         op_data->op_name = name;
2737         op_data->op_namelen = namelen;
2738         op_data->op_mode = mode;
2739         op_data->op_mod_time = ktime_get_real_seconds();
2740         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2741         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2742         op_data->op_cap = cfs_curproc_cap_pack();
2743         op_data->op_mds = 0;
2744         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2745              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2746                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2747         }
2748         op_data->op_data = data;
2749
2750         return op_data;
2751 }
2752
2753 void ll_finish_md_op_data(struct md_op_data *op_data)
2754 {
2755         ll_unlock_md_op_lsm(op_data);
2756         security_release_secctx(op_data->op_file_secctx,
2757                                 op_data->op_file_secctx_size);
2758         OBD_FREE_PTR(op_data);
2759 }
2760
2761 #ifdef HAVE_SUPEROPS_USE_DENTRY
2762 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2763 #else
2764 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2765 #endif
2766 {
2767         struct ll_sb_info *sbi;
2768
2769 #ifdef HAVE_SUPEROPS_USE_DENTRY
2770         LASSERT((seq != NULL) && (dentry != NULL));
2771         sbi = ll_s2sbi(dentry->d_sb);
2772 #else
2773         LASSERT((seq != NULL) && (vfs != NULL));
2774         sbi = ll_s2sbi(vfs->mnt_sb);
2775 #endif
2776
2777         if (sbi->ll_flags & LL_SBI_NOLCK)
2778                 seq_puts(seq, ",nolock");
2779
2780         if (sbi->ll_flags & LL_SBI_FLOCK)
2781                 seq_puts(seq, ",flock");
2782
2783         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2784                 seq_puts(seq, ",localflock");
2785
2786         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2787                 seq_puts(seq, ",user_xattr");
2788
2789         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2790                 seq_puts(seq, ",lazystatfs");
2791
2792         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2793                 seq_puts(seq, ",user_fid2path");
2794
2795         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2796                 seq_puts(seq, ",always_ping");
2797
2798         RETURN(0);
2799 }
2800
2801 /**
2802  * Get obd name by cmd, and copy out to user space
2803  */
2804 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2805 {
2806         struct ll_sb_info *sbi = ll_i2sbi(inode);
2807         struct obd_device *obd;
2808         ENTRY;
2809
2810         if (cmd == OBD_IOC_GETDTNAME)
2811                 obd = class_exp2obd(sbi->ll_dt_exp);
2812         else if (cmd == OBD_IOC_GETMDNAME)
2813                 obd = class_exp2obd(sbi->ll_md_exp);
2814         else
2815                 RETURN(-EINVAL);
2816
2817         if (!obd)
2818                 RETURN(-ENOENT);
2819
2820         if (copy_to_user((void __user *)arg, obd->obd_name,
2821                          strlen(obd->obd_name) + 1))
2822                 RETURN(-EFAULT);
2823
2824         RETURN(0);
2825 }
2826
2827 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2828 {
2829         char *path = NULL;
2830
2831         struct path p;
2832
2833         p.dentry = dentry;
2834         p.mnt = current->fs->root.mnt;
2835         path_get(&p);
2836         path = d_path(&p, buf, bufsize);
2837         path_put(&p);
2838         return path;
2839 }
2840
2841 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2842 {
2843         char *buf, *path = NULL;
2844         struct dentry *dentry = NULL;
2845         struct inode *inode = page->mapping->host;
2846
2847         /* this can be called inside spin lock so use GFP_ATOMIC. */
2848         buf = (char *)__get_free_page(GFP_ATOMIC);
2849         if (buf != NULL) {
2850                 dentry = d_find_alias(page->mapping->host);
2851                 if (dentry != NULL)
2852                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2853         }
2854
2855         CDEBUG(D_WARNING,
2856                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2857                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2858                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2859                PFID(ll_inode2fid(inode)),
2860                (path && !IS_ERR(path)) ? path : "", ioret);
2861
2862         if (dentry != NULL)
2863                 dput(dentry);
2864
2865         if (buf != NULL)
2866                 free_page((unsigned long)buf);
2867 }
2868
2869 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2870                         struct lov_user_md **kbuf)
2871 {
2872         struct lov_user_md      lum;
2873         ssize_t                 lum_size;
2874         ENTRY;
2875
2876         if (copy_from_user(&lum, md, sizeof(lum)))
2877                 RETURN(-EFAULT);
2878
2879         lum_size = ll_lov_user_md_size(&lum);
2880         if (lum_size < 0)
2881                 RETURN(lum_size);
2882
2883         OBD_ALLOC(*kbuf, lum_size);
2884         if (*kbuf == NULL)
2885                 RETURN(-ENOMEM);
2886
2887         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2888                 OBD_FREE(*kbuf, lum_size);
2889                 RETURN(-EFAULT);
2890         }
2891
2892         RETURN(lum_size);
2893 }
2894
2895 /*
2896  * Compute llite root squash state after a change of root squash
2897  * configuration setting or add/remove of a lnet nid
2898  */
2899 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2900 {
2901         struct root_squash_info *squash = &sbi->ll_squash;
2902         int i;
2903         bool matched;
2904         struct lnet_process_id id;
2905
2906         /* Update norootsquash flag */
2907         spin_lock(&squash->rsi_lock);
2908         if (list_empty(&squash->rsi_nosquash_nids))
2909                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2910         else {
2911                 /* Do not apply root squash as soon as one of our NIDs is
2912                  * in the nosquash_nids list */
2913                 matched = false;
2914                 i = 0;
2915                 while (LNetGetId(i++, &id) != -ENOENT) {
2916                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2917                                 continue;
2918                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2919                                 matched = true;
2920                                 break;
2921                         }
2922                 }
2923                 if (matched)
2924                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2925                 else
2926                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2927         }
2928         spin_unlock(&squash->rsi_lock);
2929 }
2930
2931 /**
2932  * Parse linkea content to extract information about a given hardlink
2933  *
2934  * \param[in]   ldata      - Initialized linkea data
2935  * \param[in]   linkno     - Link identifier
2936  * \param[out]  parent_fid - The entry's parent FID
2937  * \param[out]  ln         - Entry name destination buffer
2938  *
2939  * \retval 0 on success
2940  * \retval Appropriate negative error code on failure
2941  */
2942 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2943                             struct lu_fid *parent_fid, struct lu_name *ln)
2944 {
2945         unsigned int    idx;
2946         int             rc;
2947         ENTRY;
2948
2949         rc = linkea_init_with_rec(ldata);
2950         if (rc < 0)
2951                 RETURN(rc);
2952
2953         if (linkno >= ldata->ld_leh->leh_reccount)
2954                 /* beyond last link */
2955                 RETURN(-ENODATA);
2956
2957         linkea_first_entry(ldata);
2958         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2959                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2960                                     parent_fid);
2961                 if (idx == linkno)
2962                         break;
2963
2964                 linkea_next_entry(ldata);
2965         }
2966
2967         if (idx < linkno)
2968                 RETURN(-ENODATA);
2969
2970         RETURN(0);
2971 }
2972
2973 /**
2974  * Get parent FID and name of an identified link. Operation is performed for
2975  * a given link number, letting the caller iterate over linkno to list one or
2976  * all links of an entry.
2977  *
2978  * \param[in]     file - File descriptor against which to perform the operation
2979  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2980  *                       on and the available size. It is eventually filled with
2981  *                       the requested information or left untouched on error
2982  *
2983  * \retval - 0 on success
2984  * \retval - Appropriate negative error code on failure
2985  */
2986 int ll_getparent(struct file *file, struct getparent __user *arg)
2987 {
2988         struct inode            *inode = file_inode(file);
2989         struct linkea_data      *ldata;
2990         struct lu_buf            buf = LU_BUF_NULL;
2991         struct lu_name           ln;
2992         struct lu_fid            parent_fid;
2993         __u32                    linkno;
2994         __u32                    name_size;
2995         int                      rc;
2996
2997         ENTRY;
2998
2999         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
3000             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3001                 RETURN(-EPERM);
3002
3003         if (get_user(name_size, &arg->gp_name_size))
3004                 RETURN(-EFAULT);
3005
3006         if (get_user(linkno, &arg->gp_linkno))
3007                 RETURN(-EFAULT);
3008
3009         if (name_size > PATH_MAX)
3010                 RETURN(-EINVAL);
3011
3012         OBD_ALLOC(ldata, sizeof(*ldata));
3013         if (ldata == NULL)
3014                 RETURN(-ENOMEM);
3015
3016         rc = linkea_data_new(ldata, &buf);
3017         if (rc < 0)
3018                 GOTO(ldata_free, rc);
3019
3020 #ifdef HAVE_XATTR_HANDLER_FLAGS
3021         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3022                            buf.lb_len, OBD_MD_FLXATTR);
3023 #else
3024         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
3025                          buf.lb_len);
3026 #endif /* HAVE_XATTR_HANDLER_FLAGS */
3027         if (rc < 0)
3028                 GOTO(lb_free, rc);
3029
3030         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3031         if (rc < 0)
3032                 GOTO(lb_free, rc);
3033
3034         if (ln.ln_namelen >= name_size)
3035                 GOTO(lb_free, rc = -EOVERFLOW);
3036
3037         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3038                 GOTO(lb_free, rc = -EFAULT);
3039
3040         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3041                 GOTO(lb_free, rc = -EFAULT);
3042
3043         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3044                 GOTO(lb_free, rc = -EFAULT);
3045
3046 lb_free:
3047         lu_buf_free(&buf);
3048 ldata_free:
3049         OBD_FREE(ldata, sizeof(*ldata));
3050
3051         RETURN(rc);
3052 }