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