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