lustre/llite/llite_lib.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/llite_lib.c
  33  *
  34  * Lustre Light Super operations
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_LLITE
  38
  39 #include <linux/module.h>
  40 #include <linux/random.h>
  41 #include <linux/statfs.h>
  42 #include <linux/time.h>
  43 #include <linux/types.h>
  44 #include <libcfs/linux/linux-uuid.h>
  45 #include <linux/version.h>
  46 #include <linux/mm.h>
  47 #include <linux/user_namespace.h>
  48 #ifdef HAVE_UIDGID_HEADER
  49 # include <linux/uidgid.h>
  50 #endif
  51 #include <linux/security.h>
  52
  53 #include <uapi/linux/lustre/lustre_ioctl.h>
  54 #include <lustre_ha.h>
  55 #include <lustre_dlm.h>
  56 #include <lprocfs_status.h>
  57 #include <lustre_disk.h>
  58 #include <uapi/linux/lustre/lustre_param.h>
  59 #include <lustre_log.h>
  60 #include <cl_object.h>
  61 #include <obd_cksum.h>
  62 #include "llite_internal.h"
  63
  64 struct kmem_cache *ll_file_data_slab;
  65
  66 #ifndef log2
  67 #define log2(n) ffz(~(n))
  68 #endif
  69
  70 static struct ll_sb_info *ll_init_sbi(void)
  71 {
  72         struct ll_sb_info *sbi = NULL;
  73         unsigned long pages;
  74         unsigned long lru_page_max;
  75         struct sysinfo si;
  76         int rc;
  77         int i;
  78
  79         ENTRY;
  80
  81         OBD_ALLOC_PTR(sbi);
  82         if (sbi == NULL)
  83                 RETURN(ERR_PTR(-ENOMEM));
  84
  85         rc = pcc_super_init(&sbi->ll_pcc_super);
  86         if (rc < 0)
  87                 GOTO(out_sbi, rc);
  88
  89         spin_lock_init(&sbi->ll_lock);
  90         mutex_init(&sbi->ll_lco.lco_lock);
  91         spin_lock_init(&sbi->ll_pp_extent_lock);
  92         spin_lock_init(&sbi->ll_process_lock);
  93         sbi->ll_rw_stats_on = 0;
  94
  95         si_meminfo(&si);
  96         pages = si.totalram - si.totalhigh;
  97         lru_page_max = pages / 2;
  98
  99         sbi->ll_ra_info.ra_async_max_active = 0;
 100         sbi->ll_ra_info.ll_readahead_wq =
 101                 alloc_workqueue("ll-readahead-wq", WQ_UNBOUND,
 102                                 sbi->ll_ra_info.ra_async_max_active);
 103         if (!sbi->ll_ra_info.ll_readahead_wq)
 104                 GOTO(out_pcc, rc = -ENOMEM);
 105
 106         /* initialize ll_cache data */
 107         sbi->ll_cache = cl_cache_init(lru_page_max);
 108         if (sbi->ll_cache == NULL)
 109                 GOTO(out_destroy_ra, rc = -ENOMEM);
 110
 111         sbi->ll_ra_info.ra_max_pages_per_file = min(pages / 32,
 112                                            SBI_DEFAULT_READAHEAD_MAX);
 113         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 114                                 sbi->ll_ra_info.ra_max_pages_per_file;
 115         sbi->ll_ra_info.ra_max_pages = sbi->ll_ra_info.ra_max_pages_per_file;
 116         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 117
 118         sbi->ll_flags |= LL_SBI_VERBOSE;
 119 #ifdef ENABLE_CHECKSUM
 120         sbi->ll_flags |= LL_SBI_CHECKSUM;
 121 #endif
 122 #ifdef ENABLE_FLOCK
 123         sbi->ll_flags |= LL_SBI_FLOCK;
 124 #endif
 125
 126 #ifdef HAVE_LRU_RESIZE_SUPPORT
 127         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 128 #endif
 129         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 130
 131         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 132                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 133                                pp_r_hist.oh_lock);
 134                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 135                                pp_w_hist.oh_lock);
 136         }
 137
 138         /* metadata statahead is enabled by default */
 139         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 140         sbi->ll_sa_max = LL_SA_RPC_DEF;
 141         atomic_set(&sbi->ll_sa_total, 0);
 142         atomic_set(&sbi->ll_sa_wrong, 0);
 143         atomic_set(&sbi->ll_sa_running, 0);
 144         atomic_set(&sbi->ll_agl_total, 0);
 145         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 146         sbi->ll_flags |= LL_SBI_FAST_READ;
 147         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 148
 149         /* root squash */
 150         sbi->ll_squash.rsi_uid = 0;
 151         sbi->ll_squash.rsi_gid = 0;
 152         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 153         init_rwsem(&sbi->ll_squash.rsi_sem);
 154
 155         /* Per-filesystem file heat */
 156         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 157         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 158         RETURN(sbi);
 159 out_destroy_ra:
 160         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 161 out_pcc:
 162         pcc_super_fini(&sbi->ll_pcc_super);
 163 out_sbi:
 164         OBD_FREE_PTR(sbi);
 165         RETURN(ERR_PTR(rc));
 166 }
 167
 168 static void ll_free_sbi(struct super_block *sb)
 169 {
 170         struct ll_sb_info *sbi = ll_s2sbi(sb);
 171         ENTRY;
 172
 173         if (sbi != NULL) {
 174                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 175                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 176                 if (sbi->ll_ra_info.ll_readahead_wq)
 177                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 178                 if (sbi->ll_cache != NULL) {
 179                         cl_cache_decref(sbi->ll_cache);
 180                         sbi->ll_cache = NULL;
 181                 }
 182                 pcc_super_fini(&sbi->ll_pcc_super);
 183                 OBD_FREE(sbi, sizeof(*sbi));
 184         }
 185         EXIT;
 186 }
 187
 188 static int client_common_fill_super(struct super_block *sb, char *md, char *dt,
 189                                     struct vfsmount *mnt)
 190 {
 191         struct inode *root = NULL;
 192         struct ll_sb_info *sbi = ll_s2sbi(sb);
 193         struct obd_statfs *osfs = NULL;
 194         struct ptlrpc_request *request = NULL;
 195         struct obd_connect_data *data = NULL;
 196         struct obd_uuid *uuid;
 197         struct md_op_data *op_data;
 198         struct lustre_md lmd;
 199         u64 valid;
 200         int size, err, checksum;
 201
 202         ENTRY;
 203         sbi->ll_md_obd = class_name2obd(md);
 204         if (!sbi->ll_md_obd) {
 205                 CERROR("MD %s: not setup or attached\n", md);
 206                 RETURN(-EINVAL);
 207         }
 208
 209         OBD_ALLOC_PTR(data);
 210         if (data == NULL)
 211                 RETURN(-ENOMEM);
 212
 213         OBD_ALLOC_PTR(osfs);
 214         if (osfs == NULL) {
 215                 OBD_FREE_PTR(data);
 216                 RETURN(-ENOMEM);
 217         }
 218
 219         /* pass client page size via ocd_grant_blkbits, the server should report
 220          * back its backend blocksize for grant calculation purpose */
 221         data->ocd_grant_blkbits = PAGE_SHIFT;
 222
 223         /* indicate MDT features supported by this client */
 224         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 225                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 226                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 227                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 228                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 229                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 230                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 231                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 232                                   OBD_CONNECT_64BITHASH |
 233                                   OBD_CONNECT_EINPROGRESS |
 234                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 235                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 236                                   OBD_CONNECT_MAX_EASIZE |
 237                                   OBD_CONNECT_FLOCK_DEAD |
 238                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 239                                   OBD_CONNECT_OPEN_BY_FID |
 240                                   OBD_CONNECT_DIR_STRIPE |
 241                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 242                                   OBD_CONNECT_SUBTREE |
 243                                   OBD_CONNECT_MULTIMODRPCS |
 244                                   OBD_CONNECT_GRANT_PARAM |
 245                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 246
 247         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 248                                    OBD_CONNECT2_SUM_STATFS |
 249                                    OBD_CONNECT2_OVERSTRIPING |
 250                                    OBD_CONNECT2_FLR |
 251                                    OBD_CONNECT2_LOCK_CONVERT |
 252                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 253                                    OBD_CONNECT2_LSOM |
 254                                    OBD_CONNECT2_ASYNC_DISCARD |
 255                                    OBD_CONNECT2_PCC;
 256
 257 #ifdef HAVE_LRU_RESIZE_SUPPORT
 258         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 259                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 260 #endif
 261 #ifdef CONFIG_FS_POSIX_ACL
 262         data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_UMASK |
 263                                    OBD_CONNECT_LARGE_ACL;
 264 #endif
 265
 266         data->ocd_cksum_types = obd_cksum_types_supported_client();
 267
 268         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 269                 /* flag mdc connection as lightweight, only used for test
 270                  * purpose, use with care */
 271                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 272
 273         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 274         data->ocd_version = LUSTRE_VERSION_CODE;
 275
 276         if (sb->s_flags & MS_RDONLY)
 277                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 278         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 279                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 280
 281 #ifdef MS_NOSEC
 282         /* Setting this indicates we correctly support S_NOSEC (See kernel
 283          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 284          */
 285         sb->s_flags |= MS_NOSEC;
 286 #endif
 287
 288         if (sbi->ll_flags & LL_SBI_FLOCK)
 289                 sbi->ll_fop = &ll_file_operations_flock;
 290         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 291                 sbi->ll_fop = &ll_file_operations;
 292         else
 293                 sbi->ll_fop = &ll_file_operations_noflock;
 294
 295         /* always ping even if server suppress_pings */
 296         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 297                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 298
 299         obd_connect_set_secctx(data);
 300
 301 #if defined(CONFIG_SECURITY)
 302         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 303 #endif
 304
 305         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 306
 307         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 308                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 309         if (err == -EBUSY) {
 310                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 311                                    "recovery, of which this client is not a "
 312                                    "part. Please wait for recovery to complete,"
 313                                    " abort, or time out.\n", md);
 314                 GOTO(out, err);
 315         } else if (err) {
 316                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 317                 GOTO(out, err);
 318         }
 319
 320         sbi->ll_md_exp->exp_connect_data = *data;
 321
 322         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 323                            LUSTRE_SEQ_METADATA);
 324         if (err) {
 325                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 326                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 327                 GOTO(out_md, err);
 328         }
 329
 330         /* For mount, we only need fs info from MDT0, and also in DNE, it
 331          * can make sure the client can be mounted as long as MDT0 is
 332          * avaible */
 333         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 334                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 335                         OBD_STATFS_FOR_MDT0);
 336         if (err)
 337                 GOTO(out_md_fid, err);
 338
 339         /* This needs to be after statfs to ensure connect has finished.
 340          * Note that "data" does NOT contain the valid connect reply.
 341          * If connecting to a 1.8 server there will be no LMV device, so
 342          * we can access the MDC export directly and exp_connect_flags will
 343          * be non-zero, but if accessing an upgraded 2.1 server it will
 344          * have the correct flags filled in.
 345          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 346         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 347         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 348             valid != CLIENT_CONNECT_MDT_REQD) {
 349                 char *buf;
 350
 351                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 352                 obd_connect_flags2str(buf, PAGE_SIZE,
 353                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 354                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 355                                    "feature(s) needed for correct operation "
 356                                    "of this client (%s). Please upgrade "
 357                                    "server or downgrade client.\n",
 358                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 359                 OBD_FREE(buf, PAGE_SIZE);
 360                 GOTO(out_md_fid, err = -EPROTO);
 361         }
 362
 363         size = sizeof(*data);
 364         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 365                            KEY_CONN_DATA,  &size, data);
 366         if (err) {
 367                 CERROR("%s: Get connect data failed: rc = %d\n",
 368                        sbi->ll_md_exp->exp_obd->obd_name, err);
 369                 GOTO(out_md_fid, err);
 370         }
 371
 372         LASSERT(osfs->os_bsize);
 373         sb->s_blocksize = osfs->os_bsize;
 374         sb->s_blocksize_bits = log2(osfs->os_bsize);
 375         sb->s_magic = LL_SUPER_MAGIC;
 376         sb->s_maxbytes = MAX_LFS_FILESIZE;
 377         sbi->ll_namelen = osfs->os_namelen;
 378         sbi->ll_mnt.mnt = current->fs->root.mnt;
 379
 380         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 381             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 382                 LCONSOLE_INFO("Disabling user_xattr feature because "
 383                               "it is not supported on the server\n");
 384                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 385         }
 386
 387         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 388 #ifdef MS_POSIXACL
 389                 sb->s_flags |= MS_POSIXACL;
 390 #endif
 391                 sbi->ll_flags |= LL_SBI_ACL;
 392         } else {
 393                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 394 #ifdef MS_POSIXACL
 395                 sb->s_flags &= ~MS_POSIXACL;
 396 #endif
 397                 sbi->ll_flags &= ~LL_SBI_ACL;
 398         }
 399
 400         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 401                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 402
 403         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 404                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 405
 406         if (obd_connect_has_secctx(data))
 407                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 408
 409         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 410                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 411                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 412                                       "unknown maximum xattr size.\n", dt);
 413                 } else if (!sbi->ll_xattr_cache_set) {
 414                         /* If xattr_cache is already set (no matter 0 or 1)
 415                          * during processing llog, it won't be enabled here. */
 416                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 417                         sbi->ll_xattr_cache_enabled = 1;
 418                 }
 419         }
 420
 421         sbi->ll_dt_obd = class_name2obd(dt);
 422         if (!sbi->ll_dt_obd) {
 423                 CERROR("DT %s: not setup or attached\n", dt);
 424                 GOTO(out_md_fid, err = -ENODEV);
 425         }
 426
 427         /* pass client page size via ocd_grant_blkbits, the server should report
 428          * back its backend blocksize for grant calculation purpose */
 429         data->ocd_grant_blkbits = PAGE_SHIFT;
 430
 431         /* indicate OST features supported by this client */
 432         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 433                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 434                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 435                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 436                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 437                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 438                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 439                                   OBD_CONNECT_EINPROGRESS |
 440                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 441                                   OBD_CONNECT_LAYOUTLOCK |
 442                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 443                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 444                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 445
 446 /* The client currently advertises support for OBD_CONNECT_LOCKAHEAD_OLD so it
 447  * can interoperate with an older version of lockahead which was released prior
 448  * to landing in master. This support will be dropped when 2.13 development
 449  * starts.  At the point, we should not just drop the connect flag (below), we
 450  * should also remove the support in the code.
 451  *
 452  * Removing it means a few things:
 453  * 1. Remove this section here
 454  * 2. Remove CEF_NONBLOCK in ll_file_lockahead()
 455  * 3. Remove function exp_connect_lockahead_old
 456  * 4. Remove LDLM_FL_LOCKAHEAD_OLD_RESERVED in lustre_dlm_flags.h
 457  * */
 458 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 12, 50, 0)
 459         data->ocd_connect_flags |= OBD_CONNECT_LOCKAHEAD_OLD;
 460 #endif
 461
 462         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD;
 463
 464         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 465                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 466
 467         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 468          * disabled by default, because it can still be enabled on the
 469          * fly via /sys. As a consequence, we still need to come to an
 470          * agreement on the supported algorithms at connect time
 471          */
 472         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 473
 474         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 475                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 476         else
 477                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 478
 479 #ifdef HAVE_LRU_RESIZE_SUPPORT
 480         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 481 #endif
 482         /* always ping even if server suppress_pings */
 483         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 484                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 485
 486         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 487                "ocd_grant: %d\n", data->ocd_connect_flags,
 488                data->ocd_version, data->ocd_grant);
 489
 490         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 491         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 492
 493         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 494
 495         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 496                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 497         if (err == -EBUSY) {
 498                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 499                                    "recovery, of which this client is not a "
 500                                    "part.  Please wait for recovery to "
 501                                    "complete, abort, or time out.\n", dt);
 502                 GOTO(out_md, err);
 503         } else if (err) {
 504                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 505                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 506                 GOTO(out_md, err);
 507         }
 508
 509         sbi->ll_dt_exp->exp_connect_data = *data;
 510
 511         /* Don't change value if it was specified in the config log */
 512         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 513                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 514                         max_t(unsigned long, SBI_DEFAULT_READAHEAD_WHOLE_MAX,
 515                               (data->ocd_brw_size >> PAGE_SHIFT));
 516                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 517                     sbi->ll_ra_info.ra_max_pages_per_file)
 518                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 519                                 sbi->ll_ra_info.ra_max_pages_per_file;
 520         }
 521
 522         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 523                            LUSTRE_SEQ_METADATA);
 524         if (err) {
 525                 CERROR("%s: Can't init data layer FID infrastructure, "
 526                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 527                 GOTO(out_dt, err);
 528         }
 529
 530         mutex_lock(&sbi->ll_lco.lco_lock);
 531         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 532         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 533         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 534         mutex_unlock(&sbi->ll_lco.lco_lock);
 535
 536         fid_zero(&sbi->ll_root_fid);
 537         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 538                            &sbi->ll_root_fid);
 539         if (err) {
 540                 CERROR("cannot mds_connect: rc = %d\n", err);
 541                 GOTO(out_lock_cn_cb, err);
 542         }
 543         if (!fid_is_sane(&sbi->ll_root_fid)) {
 544                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 545                        sbi->ll_md_exp->exp_obd->obd_name,
 546                        PFID(&sbi->ll_root_fid));
 547                 GOTO(out_lock_cn_cb, err = -EINVAL);
 548         }
 549         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 550
 551         sb->s_op = &lustre_super_operations;
 552 #ifdef HAVE_XATTR_HANDLER_FLAGS
 553         sb->s_xattr = ll_xattr_handlers;
 554 #endif
 555 #if THREAD_SIZE >= 8192 /*b=17630*/
 556         sb->s_export_op = &lustre_export_operations;
 557 #endif
 558
 559         /* make root inode
 560          * XXX: move this to after cbd setup? */
 561         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 562         if (sbi->ll_flags & LL_SBI_ACL)
 563                 valid |= OBD_MD_FLACL;
 564
 565         OBD_ALLOC_PTR(op_data);
 566         if (op_data == NULL)
 567                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 568
 569         op_data->op_fid1 = sbi->ll_root_fid;
 570         op_data->op_mode = 0;
 571         op_data->op_valid = valid;
 572
 573         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 574
 575         OBD_FREE_PTR(op_data);
 576         if (err) {
 577                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 578                        sbi->ll_md_exp->exp_obd->obd_name, err);
 579                 GOTO(out_lock_cn_cb, err);
 580         }
 581
 582         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 583                                sbi->ll_md_exp, &lmd);
 584         if (err) {
 585                 CERROR("failed to understand root inode md: rc = %d\n", err);
 586                 ptlrpc_req_finished(request);
 587                 GOTO(out_lock_cn_cb, err);
 588         }
 589
 590         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 591         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 592                                             sbi->ll_flags & LL_SBI_32BIT_API),
 593                        &lmd);
 594         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 595         ptlrpc_req_finished(request);
 596
 597         if (IS_ERR(root)) {
 598 #ifdef CONFIG_FS_POSIX_ACL
 599                 if (lmd.posix_acl) {
 600                         posix_acl_release(lmd.posix_acl);
 601                         lmd.posix_acl = NULL;
 602                 }
 603 #endif
 604                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 605                 root = NULL;
 606                 CERROR("lustre_lite: bad iget4 for root\n");
 607                 GOTO(out_root, err);
 608         }
 609
 610         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 611         if (sbi->ll_checksum_set) {
 612                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 613                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 614                                          sizeof(checksum), &checksum, NULL);
 615                 if (err) {
 616                         CERROR("%s: Set checksum failed: rc = %d\n",
 617                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 618                         GOTO(out_root, err);
 619                 }
 620         }
 621         cl_sb_init(sb);
 622
 623         sb->s_root = d_make_root(root);
 624         if (sb->s_root == NULL) {
 625                 err = -ENOMEM;
 626                 CERROR("%s: can't make root dentry: rc = %d\n",
 627                        sbi->ll_fsname, err);
 628                 GOTO(out_root, err);
 629         }
 630 #ifdef HAVE_DCACHE_LOCK
 631         sb->s_root->d_op = &ll_d_ops;
 632 #endif
 633
 634         sbi->ll_sdev_orig = sb->s_dev;
 635
 636         /* We set sb->s_dev equal on all lustre clients in order to support
 637          * NFS export clustering.  NFSD requires that the FSID be the same
 638          * on all clients. */
 639         /* s_dev is also used in lt_compare() to compare two fs, but that is
 640          * only a node-local comparison. */
 641         uuid = obd_get_uuid(sbi->ll_md_exp);
 642         if (uuid != NULL)
 643                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 644
 645         if (data != NULL)
 646                 OBD_FREE_PTR(data);
 647         if (osfs != NULL)
 648                 OBD_FREE_PTR(osfs);
 649
 650         if (sbi->ll_dt_obd) {
 651                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 652                                         &sbi->ll_dt_obd->obd_kset.kobj,
 653                                         sbi->ll_dt_obd->obd_type->typ_name);
 654                 if (err < 0) {
 655                         CERROR("%s: could not register %s in llite: rc = %d\n",
 656                                dt, sbi->ll_fsname, err);
 657                         err = 0;
 658                 }
 659         }
 660
 661         if (sbi->ll_md_obd) {
 662                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 663                                         &sbi->ll_md_obd->obd_kset.kobj,
 664                                         sbi->ll_md_obd->obd_type->typ_name);
 665                 if (err < 0) {
 666                         CERROR("%s: could not register %s in llite: rc = %d\n",
 667                                md, sbi->ll_fsname, err);
 668                         err = 0;
 669                 }
 670         }
 671
 672         RETURN(err);
 673 out_root:
 674         if (root)
 675                 iput(root);
 676 out_lock_cn_cb:
 677         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 678 out_dt:
 679         obd_disconnect(sbi->ll_dt_exp);
 680         sbi->ll_dt_exp = NULL;
 681         sbi->ll_dt_obd = NULL;
 682 out_md_fid:
 683         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 684 out_md:
 685         obd_disconnect(sbi->ll_md_exp);
 686         sbi->ll_md_exp = NULL;
 687         sbi->ll_md_obd = NULL;
 688 out:
 689         if (data != NULL)
 690                 OBD_FREE_PTR(data);
 691         if (osfs != NULL)
 692                 OBD_FREE_PTR(osfs);
 693         return err;
 694 }
 695
 696 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 697 {
 698         int size, rc;
 699
 700         size = sizeof(*lmmsize);
 701         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 702                           KEY_MAX_EASIZE, &size, lmmsize);
 703         if (rc != 0) {
 704                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 705                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 706                 RETURN(rc);
 707         }
 708
 709         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 710
 711         size = sizeof(int);
 712         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 713                           KEY_MAX_EASIZE, &size, lmmsize);
 714         if (rc)
 715                 CERROR("Get max mdsize error rc %d\n", rc);
 716
 717         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 718
 719         RETURN(rc);
 720 }
 721
 722 /**
 723  * Get the value of the default_easize parameter.
 724  *
 725  * \see client_obd::cl_default_mds_easize
 726  *
 727  * \param[in] sbi       superblock info for this filesystem
 728  * \param[out] lmmsize  pointer to storage location for value
 729  *
 730  * \retval 0            on success
 731  * \retval negative     negated errno on failure
 732  */
 733 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 734 {
 735         int size, rc;
 736
 737         size = sizeof(int);
 738         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 739                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 740         if (rc)
 741                 CERROR("Get default mdsize error rc %d\n", rc);
 742
 743         RETURN(rc);
 744 }
 745
 746 /**
 747  * Set the default_easize parameter to the given value.
 748  *
 749  * \see client_obd::cl_default_mds_easize
 750  *
 751  * \param[in] sbi       superblock info for this filesystem
 752  * \param[in] lmmsize   the size to set
 753  *
 754  * \retval 0            on success
 755  * \retval negative     negated errno on failure
 756  */
 757 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 758 {
 759         int rc;
 760
 761         if (lmmsize < sizeof(struct lov_mds_md) ||
 762             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 763                 return -EINVAL;
 764
 765         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 766                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 767                                 sizeof(int), &lmmsize, NULL);
 768
 769         RETURN(rc);
 770 }
 771
 772 static void client_common_put_super(struct super_block *sb)
 773 {
 774         struct ll_sb_info *sbi = ll_s2sbi(sb);
 775         ENTRY;
 776
 777         cl_sb_fini(sb);
 778
 779         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 780         obd_disconnect(sbi->ll_dt_exp);
 781         sbi->ll_dt_exp = NULL;
 782
 783         ll_debugfs_unregister_super(sb);
 784
 785         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 786         obd_disconnect(sbi->ll_md_exp);
 787         sbi->ll_md_exp = NULL;
 788
 789         EXIT;
 790 }
 791
 792 void ll_kill_super(struct super_block *sb)
 793 {
 794         struct ll_sb_info *sbi;
 795         ENTRY;
 796
 797         /* not init sb ?*/
 798         if (!(sb->s_flags & MS_ACTIVE))
 799                 return;
 800
 801         sbi = ll_s2sbi(sb);
 802         /* we need restore s_dev from changed for clustred NFS before put_super
 803          * because new kernels have cached s_dev and change sb->s_dev in
 804          * put_super not affected real removing devices */
 805         if (sbi) {
 806                 sb->s_dev = sbi->ll_sdev_orig;
 807                 sbi->ll_umounting = 1;
 808
 809                 /* wait running statahead threads to quit */
 810                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 811                         set_current_state(TASK_UNINTERRUPTIBLE);
 812                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 813                 }
 814         }
 815
 816         EXIT;
 817 }
 818
 819 static inline int ll_set_opt(const char *opt, char *data, int fl)
 820 {
 821         if (strncmp(opt, data, strlen(opt)) != 0)
 822                 return 0;
 823         else
 824                 return fl;
 825 }
 826
 827 /* non-client-specific mount options are parsed in lmd_parse */
 828 static int ll_options(char *options, struct ll_sb_info *sbi)
 829 {
 830         int tmp;
 831         char *s1 = options, *s2;
 832         int *flags = &sbi->ll_flags;
 833         ENTRY;
 834
 835         if (!options)
 836                 RETURN(0);
 837
 838         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 839
 840         while (*s1) {
 841                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 842                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 843                 if (tmp) {
 844                         *flags |= tmp;
 845                         goto next;
 846                 }
 847                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 848                 if (tmp) {
 849                         *flags |= tmp;
 850                         goto next;
 851                 }
 852                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 853                 if (tmp) {
 854                         *flags |= tmp;
 855                         goto next;
 856                 }
 857                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 858                 if (tmp) {
 859                         *flags &= ~tmp;
 860                         goto next;
 861                 }
 862                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 863                 if (tmp) {
 864                         *flags |= tmp;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 868                 if (tmp) {
 869                         *flags &= ~tmp;
 870                         goto next;
 871                 }
 872                 tmp = ll_set_opt("context", s1, 1);
 873                 if (tmp)
 874                         goto next;
 875                 tmp = ll_set_opt("fscontext", s1, 1);
 876                 if (tmp)
 877                         goto next;
 878                 tmp = ll_set_opt("defcontext", s1, 1);
 879                 if (tmp)
 880                         goto next;
 881                 tmp = ll_set_opt("rootcontext", s1, 1);
 882                 if (tmp)
 883                         goto next;
 884                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 885                 if (tmp) {
 886                         *flags |= tmp;
 887                         goto next;
 888                 }
 889                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 890                 if (tmp) {
 891                         *flags &= ~tmp;
 892                         goto next;
 893                 }
 894
 895                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 896                 if (tmp) {
 897                         *flags |= tmp;
 898                         sbi->ll_checksum_set = 1;
 899                         goto next;
 900                 }
 901                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 902                 if (tmp) {
 903                         *flags &= ~tmp;
 904                         sbi->ll_checksum_set = 1;
 905                         goto next;
 906                 }
 907                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 908                 if (tmp) {
 909                         *flags |= tmp;
 910                         goto next;
 911                 }
 912                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 913                 if (tmp) {
 914                         *flags &= ~tmp;
 915                         goto next;
 916                 }
 917                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 918                 if (tmp) {
 919                         *flags |= tmp;
 920                         goto next;
 921                 }
 922                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 923                 if (tmp) {
 924                         *flags &= ~tmp;
 925                         goto next;
 926                 }
 927                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 928                 if (tmp) {
 929                         *flags |= tmp;
 930                         goto next;
 931                 }
 932                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 933                 if (tmp) {
 934                         *flags |= tmp;
 935                         goto next;
 936                 }
 937                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 938                 if (tmp) {
 939                         *flags &= ~tmp;
 940                         goto next;
 941                 }
 942                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 943                 if (tmp) {
 944                         *flags |= tmp;
 945                         goto next;
 946                 }
 947                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 948                                    s1);
 949                 RETURN(-EINVAL);
 950
 951 next:
 952                 /* Find next opt */
 953                 s2 = strchr(s1, ',');
 954                 if (s2 == NULL)
 955                         break;
 956                 s1 = s2 + 1;
 957         }
 958         RETURN(0);
 959 }
 960
 961 void ll_lli_init(struct ll_inode_info *lli)
 962 {
 963         lli->lli_inode_magic = LLI_INODE_MAGIC;
 964         lli->lli_flags = 0;
 965         spin_lock_init(&lli->lli_lock);
 966         lli->lli_posix_acl = NULL;
 967         /* Do not set lli_fid, it has been initialized already. */
 968         fid_zero(&lli->lli_pfid);
 969         lli->lli_mds_read_och = NULL;
 970         lli->lli_mds_write_och = NULL;
 971         lli->lli_mds_exec_och = NULL;
 972         lli->lli_open_fd_read_count = 0;
 973         lli->lli_open_fd_write_count = 0;
 974         lli->lli_open_fd_exec_count = 0;
 975         mutex_init(&lli->lli_och_mutex);
 976         spin_lock_init(&lli->lli_agl_lock);
 977         spin_lock_init(&lli->lli_layout_lock);
 978         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 979         lli->lli_clob = NULL;
 980
 981         init_rwsem(&lli->lli_xattrs_list_rwsem);
 982         mutex_init(&lli->lli_xattrs_enq_lock);
 983
 984         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 985         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 986                 mutex_init(&lli->lli_readdir_mutex);
 987                 lli->lli_opendir_key = NULL;
 988                 lli->lli_sai = NULL;
 989                 spin_lock_init(&lli->lli_sa_lock);
 990                 lli->lli_opendir_pid = 0;
 991                 lli->lli_sa_enabled = 0;
 992                 init_rwsem(&lli->lli_lsm_sem);
 993         } else {
 994                 mutex_init(&lli->lli_size_mutex);
 995                 lli->lli_symlink_name = NULL;
 996                 init_rwsem(&lli->lli_trunc_sem);
 997                 range_lock_tree_init(&lli->lli_write_tree);
 998                 init_rwsem(&lli->lli_glimpse_sem);
 999                 lli->lli_glimpse_time = ktime_set(0, 0);
1000                 INIT_LIST_HEAD(&lli->lli_agl_list);
1001                 lli->lli_agl_index = 0;
1002                 lli->lli_async_rc = 0;
1003                 spin_lock_init(&lli->lli_heat_lock);
1004                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1005                 lli->lli_heat_flags = 0;
1006                 mutex_init(&lli->lli_pcc_lock);
1007                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1008                 lli->lli_pcc_inode = NULL;
1009         }
1010         mutex_init(&lli->lli_layout_mutex);
1011         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1012 }
1013
1014 #define MAX_STRING_SIZE 128
1015
1016 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1017
1018 #define LSI_BDI_INITIALIZED     0x00400000
1019
1020 #ifndef HAVE_BDI_CAP_MAP_COPY
1021 # define BDI_CAP_MAP_COPY       0
1022 #endif
1023
1024 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1025 {
1026         struct  lustre_sb_info *lsi = s2lsi(sb);
1027         char buf[MAX_STRING_SIZE];
1028         va_list args;
1029         int err;
1030
1031         err = bdi_init(&lsi->lsi_bdi);
1032         if (err)
1033                 return err;
1034
1035         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1036         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1037         lsi->lsi_bdi.name = "lustre";
1038         va_start(args, fmt);
1039         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1040         va_end(args);
1041         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1042         va_end(args);
1043         if (!err)
1044                 sb->s_bdi = &lsi->lsi_bdi;
1045
1046         return err;
1047 }
1048 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1049
1050 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1051 {
1052         struct  lustre_profile *lprof = NULL;
1053         struct  lustre_sb_info *lsi = s2lsi(sb);
1054         struct  ll_sb_info *sbi = NULL;
1055         char    *dt = NULL, *md = NULL;
1056         char    *profilenm = get_profile_name(sb);
1057         struct config_llog_instance *cfg;
1058         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1059         const int instlen = 16 + 2;
1060         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1061         char name[MAX_STRING_SIZE];
1062         int md_len = 0;
1063         int dt_len = 0;
1064         uuid_t uuid;
1065         char *ptr;
1066         int len;
1067         int err;
1068
1069         ENTRY;
1070         /* for ASLR, to map between cfg_instance and hashed ptr */
1071         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1072                profilenm, cfg_instance, sb);
1073
1074         try_module_get(THIS_MODULE);
1075
1076         OBD_ALLOC_PTR(cfg);
1077         if (cfg == NULL)
1078                 GOTO(out_free_cfg, err = -ENOMEM);
1079
1080         /* client additional sb info */
1081         lsi->lsi_llsbi = sbi = ll_init_sbi();
1082         if (IS_ERR(sbi))
1083                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1084
1085         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1086         if (err)
1087                 GOTO(out_free_cfg, err);
1088
1089 #ifndef HAVE_DCACHE_LOCK
1090         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1091         sb->s_d_op = &ll_d_ops;
1092 #endif
1093         /* UUID handling */
1094         generate_random_uuid(uuid.b);
1095         snprintf(sbi->ll_sb_uuid.uuid, UUID_SIZE, "%pU", uuid.b);
1096
1097         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1098
1099         /* Get fsname */
1100         len = strlen(profilenm);
1101         ptr = strrchr(profilenm, '-');
1102         if (ptr && (strcmp(ptr, "-client") == 0))
1103                 len -= 7;
1104
1105         if (len > LUSTRE_MAXFSNAME) {
1106                 if (unlikely(len >= MAX_STRING_SIZE))
1107                         len = MAX_STRING_SIZE - 1;
1108                 strncpy(name, profilenm, len);
1109                 name[len] = '\0';
1110                 err = -ENAMETOOLONG;
1111                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1112                        name, LUSTRE_MAXFSNAME, err);
1113                 GOTO(out_free_cfg, err);
1114         }
1115         strncpy(sbi->ll_fsname, profilenm, len);
1116         sbi->ll_fsname[len] = '\0';
1117
1118         /* Mount info */
1119         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1120                  profilenm, cfg_instance);
1121
1122         err = super_setup_bdi_name(sb, "%s", name);
1123         if (err)
1124                 GOTO(out_free_cfg, err);
1125
1126         /* Call ll_debugfs_register_super() before lustre_process_log()
1127          * so that "llite.*.*" params can be processed correctly.
1128          */
1129         err = ll_debugfs_register_super(sb, name);
1130         if (err < 0) {
1131                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1132                        sbi->ll_fsname, err);
1133                 err = 0;
1134         }
1135
1136         /* The cfg_instance is a value unique to this super, in case some
1137          * joker tries to mount the same fs at two mount points.
1138          */
1139         cfg->cfg_instance = cfg_instance;
1140         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1141         cfg->cfg_callback = class_config_llog_handler;
1142         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1143         /* set up client obds */
1144         err = lustre_process_log(sb, profilenm, cfg);
1145         if (err < 0)
1146                 GOTO(out_debugfs, err);
1147
1148         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1149         lprof = class_get_profile(profilenm);
1150         if (lprof == NULL) {
1151                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1152                                    " read from the MGS.  Does that filesystem "
1153                                    "exist?\n", profilenm);
1154                 GOTO(out_debugfs, err = -EINVAL);
1155         }
1156         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1157                lprof->lp_md, lprof->lp_dt);
1158
1159         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1160         OBD_ALLOC(dt, dt_len);
1161         if (!dt)
1162                 GOTO(out_profile, err = -ENOMEM);
1163         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1164
1165         md_len = strlen(lprof->lp_md) + instlen + 2;
1166         OBD_ALLOC(md, md_len);
1167         if (!md)
1168                 GOTO(out_free_dt, err = -ENOMEM);
1169         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1170
1171         /* connections, registrations, sb setup */
1172         err = client_common_fill_super(sb, md, dt, mnt);
1173         if (err < 0)
1174                 GOTO(out_free_md, err);
1175
1176         sbi->ll_client_common_fill_super_succeeded = 1;
1177
1178 out_free_md:
1179         if (md)
1180                 OBD_FREE(md, md_len);
1181 out_free_dt:
1182         if (dt)
1183                 OBD_FREE(dt, dt_len);
1184 out_profile:
1185         if (lprof)
1186                 class_put_profile(lprof);
1187 out_debugfs:
1188         if (err < 0)
1189                 ll_debugfs_unregister_super(sb);
1190 out_free_cfg:
1191         if (cfg)
1192                 OBD_FREE_PTR(cfg);
1193
1194         if (err)
1195                 ll_put_super(sb);
1196         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1197                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1198         RETURN(err);
1199 } /* ll_fill_super */
1200
1201 void ll_put_super(struct super_block *sb)
1202 {
1203         struct config_llog_instance cfg, params_cfg;
1204         struct obd_device *obd;
1205         struct lustre_sb_info *lsi = s2lsi(sb);
1206         struct ll_sb_info *sbi = ll_s2sbi(sb);
1207         char *profilenm = get_profile_name(sb);
1208         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1209         long ccc_count;
1210         int next, force = 1, rc = 0;
1211         ENTRY;
1212
1213         if (IS_ERR(sbi))
1214                 GOTO(out_no_sbi, 0);
1215
1216         /* Should replace instance_id with something better for ASLR */
1217         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1218                profilenm, cfg_instance, sb);
1219
1220         cfg.cfg_instance = cfg_instance;
1221         lustre_end_log(sb, profilenm, &cfg);
1222
1223         params_cfg.cfg_instance = cfg_instance;
1224         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1225
1226         if (sbi->ll_md_exp) {
1227                 obd = class_exp2obd(sbi->ll_md_exp);
1228                 if (obd)
1229                         force = obd->obd_force;
1230         }
1231
1232         /* Wait for unstable pages to be committed to stable storage */
1233         if (force == 0) {
1234                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1235                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1236                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1237                         &lwi);
1238         }
1239
1240         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1241         if (force == 0 && rc != -EINTR)
1242                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1243
1244         /* We need to set force before the lov_disconnect in
1245            lustre_common_put_super, since l_d cleans up osc's as well. */
1246         if (force) {
1247                 next = 0;
1248                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1249                                                      &next)) != NULL) {
1250                         obd->obd_force = force;
1251                 }
1252         }
1253
1254         if (sbi->ll_client_common_fill_super_succeeded) {
1255                 /* Only if client_common_fill_super succeeded */
1256                 client_common_put_super(sb);
1257         }
1258
1259         next = 0;
1260         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1261                 class_manual_cleanup(obd);
1262         }
1263
1264         if (sbi->ll_flags & LL_SBI_VERBOSE)
1265                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1266
1267         if (profilenm)
1268                 class_del_profile(profilenm);
1269
1270 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1271         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1272                 bdi_destroy(&lsi->lsi_bdi);
1273                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1274         }
1275 #endif
1276
1277         ll_free_sbi(sb);
1278         lsi->lsi_llsbi = NULL;
1279 out_no_sbi:
1280         lustre_common_put_super(sb);
1281
1282         cl_env_cache_purge(~0);
1283
1284         module_put(THIS_MODULE);
1285
1286         EXIT;
1287 } /* client_put_super */
1288
1289 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1290 {
1291         struct inode *inode = NULL;
1292
1293         /* NOTE: we depend on atomic igrab() -bzzz */
1294         lock_res_and_lock(lock);
1295         if (lock->l_resource->lr_lvb_inode) {
1296                 struct ll_inode_info * lli;
1297                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1298                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1299                         inode = igrab(lock->l_resource->lr_lvb_inode);
1300                 } else {
1301                         inode = lock->l_resource->lr_lvb_inode;
1302                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1303                                          D_WARNING, lock, "lr_lvb_inode %p is "
1304                                          "bogus: magic %08x",
1305                                          lock->l_resource->lr_lvb_inode,
1306                                          lli->lli_inode_magic);
1307                         inode = NULL;
1308                 }
1309         }
1310         unlock_res_and_lock(lock);
1311         return inode;
1312 }
1313
1314 void ll_dir_clear_lsm_md(struct inode *inode)
1315 {
1316         struct ll_inode_info *lli = ll_i2info(inode);
1317
1318         LASSERT(S_ISDIR(inode->i_mode));
1319
1320         if (lli->lli_lsm_md) {
1321                 lmv_free_memmd(lli->lli_lsm_md);
1322                 lli->lli_lsm_md = NULL;
1323         }
1324
1325         if (lli->lli_default_lsm_md) {
1326                 lmv_free_memmd(lli->lli_default_lsm_md);
1327                 lli->lli_default_lsm_md = NULL;
1328         }
1329 }
1330
1331 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1332                                       const struct lu_fid *fid,
1333                                       struct lustre_md *md)
1334 {
1335         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1336         struct mdt_body         *body = md->body;
1337         struct inode            *inode;
1338         ino_t                   ino;
1339         ENTRY;
1340
1341         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1342         inode = iget_locked(sb, ino);
1343         if (inode == NULL) {
1344                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1345                        sbi->ll_fsname, PFID(fid));
1346                 RETURN(ERR_PTR(-ENOENT));
1347         }
1348
1349         if (inode->i_state & I_NEW) {
1350                 struct ll_inode_info *lli = ll_i2info(inode);
1351                 struct lmv_stripe_md *lsm = md->lmv;
1352
1353                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1354                                 (body->mbo_mode & S_IFMT);
1355                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1356                          PFID(fid));
1357
1358                 inode->i_mtime.tv_sec = 0;
1359                 inode->i_atime.tv_sec = 0;
1360                 inode->i_ctime.tv_sec = 0;
1361                 inode->i_rdev = 0;
1362
1363 #ifdef HAVE_BACKING_DEV_INFO
1364                 /* initializing backing dev info. */
1365                 inode->i_mapping->backing_dev_info =
1366                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1367 #endif
1368                 inode->i_op = &ll_dir_inode_operations;
1369                 inode->i_fop = &ll_dir_operations;
1370                 lli->lli_fid = *fid;
1371                 ll_lli_init(lli);
1372
1373                 LASSERT(lsm != NULL);
1374                 /* master object FID */
1375                 lli->lli_pfid = body->mbo_fid1;
1376                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1377                        lli, PFID(fid), PFID(&lli->lli_pfid));
1378                 unlock_new_inode(inode);
1379         }
1380
1381         RETURN(inode);
1382 }
1383
1384 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1385 {
1386         struct lu_fid *fid;
1387         struct lmv_stripe_md *lsm = md->lmv;
1388         struct ll_inode_info *lli = ll_i2info(inode);
1389         int i;
1390
1391         LASSERT(lsm != NULL);
1392
1393         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1394                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1395         lsm_md_dump(D_INODE, lsm);
1396
1397         if (!lmv_dir_striped(lsm))
1398                 goto out;
1399
1400         /* XXX sigh, this lsm_root initialization should be in
1401          * LMV layer, but it needs ll_iget right now, so we
1402          * put this here right now. */
1403         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1404                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1405                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1406
1407                 if (!fid_is_sane(fid))
1408                         continue;
1409
1410                 /* Unfortunately ll_iget will call ll_update_inode,
1411                  * where the initialization of slave inode is slightly
1412                  * different, so it reset lsm_md to NULL to avoid
1413                  * initializing lsm for slave inode. */
1414                 lsm->lsm_md_oinfo[i].lmo_root =
1415                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1416                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1417                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1418
1419                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1420                         while (i-- > 0) {
1421                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1422                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1423                         }
1424                         return rc;
1425                 }
1426         }
1427 out:
1428         lli->lli_lsm_md = lsm;
1429
1430         return 0;
1431 }
1432
1433 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1434 {
1435         struct ll_inode_info *lli = ll_i2info(inode);
1436
1437         if (!md->default_lmv) {
1438                 /* clear default lsm */
1439                 if (lli->lli_default_lsm_md) {
1440                         down_write(&lli->lli_lsm_sem);
1441                         if (lli->lli_default_lsm_md) {
1442                                 lmv_free_memmd(lli->lli_default_lsm_md);
1443                                 lli->lli_default_lsm_md = NULL;
1444                         }
1445                         up_write(&lli->lli_lsm_sem);
1446                 }
1447         } else if (lli->lli_default_lsm_md) {
1448                 /* update default lsm if it changes */
1449                 down_read(&lli->lli_lsm_sem);
1450                 if (lli->lli_default_lsm_md &&
1451                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1452                         up_read(&lli->lli_lsm_sem);
1453                         down_write(&lli->lli_lsm_sem);
1454                         if (lli->lli_default_lsm_md)
1455                                 lmv_free_memmd(lli->lli_default_lsm_md);
1456                         lli->lli_default_lsm_md = md->default_lmv;
1457                         lsm_md_dump(D_INODE, md->default_lmv);
1458                         md->default_lmv = NULL;
1459                         up_write(&lli->lli_lsm_sem);
1460                 } else {
1461                         up_read(&lli->lli_lsm_sem);
1462                 }
1463         } else {
1464                 /* init default lsm */
1465                 down_write(&lli->lli_lsm_sem);
1466                 lli->lli_default_lsm_md = md->default_lmv;
1467                 lsm_md_dump(D_INODE, md->default_lmv);
1468                 md->default_lmv = NULL;
1469                 up_write(&lli->lli_lsm_sem);
1470         }
1471 }
1472
1473 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1474 {
1475         struct ll_inode_info *lli = ll_i2info(inode);
1476         struct lmv_stripe_md *lsm = md->lmv;
1477         int rc = 0;
1478
1479         ENTRY;
1480
1481         LASSERT(S_ISDIR(inode->i_mode));
1482         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1483                PFID(ll_inode2fid(inode)));
1484
1485         /* update default LMV */
1486         if (md->default_lmv)
1487                 ll_update_default_lsm_md(inode, md);
1488
1489         /*
1490          * no striped information from request, lustre_md from req does not
1491          * include stripeEA, see ll_md_setattr()
1492          */
1493         if (!lsm)
1494                 RETURN(0);
1495
1496         /*
1497          * normally dir layout doesn't change, only take read lock to check
1498          * that to avoid blocking other MD operations.
1499          */
1500         if (lli->lli_lsm_md)
1501                 down_read(&lli->lli_lsm_sem);
1502         else
1503                 down_write(&lli->lli_lsm_sem);
1504
1505         /*
1506          * if dir layout mismatch, check whether version is increased, which
1507          * means layout is changed, this happens in dir migration and lfsck.
1508          *
1509          * foreign LMV should not change.
1510          */
1511         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1512                 if (lmv_dir_striped(lli->lli_lsm_md) &&
1513                     lsm->lsm_md_layout_version <=
1514                     lli->lli_lsm_md->lsm_md_layout_version) {
1515                         CERROR("%s: "DFID" dir layout mismatch:\n",
1516                                ll_i2sbi(inode)->ll_fsname,
1517                                PFID(&lli->lli_fid));
1518                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1519                         lsm_md_dump(D_ERROR, lsm);
1520                         GOTO(unlock, rc = -EINVAL);
1521                 }
1522
1523                 /* layout changed, switch to write lock */
1524                 up_read(&lli->lli_lsm_sem);
1525                 down_write(&lli->lli_lsm_sem);
1526                 ll_dir_clear_lsm_md(inode);
1527         }
1528
1529         /* set directory layout */
1530         if (!lli->lli_lsm_md) {
1531                 struct cl_attr  *attr;
1532
1533                 rc = ll_init_lsm_md(inode, md);
1534                 up_write(&lli->lli_lsm_sem);
1535                 if (rc != 0)
1536                         RETURN(rc);
1537
1538                 /* set md->lmv to NULL, so the following free lustre_md
1539                  * will not free this lsm */
1540                 md->lmv = NULL;
1541
1542                 /*
1543                  * md_merge_attr() may take long, since lsm is already set,
1544                  * switch to read lock.
1545                  */
1546                 down_read(&lli->lli_lsm_sem);
1547
1548                 if (!lmv_dir_striped(lli->lli_lsm_md))
1549                         GOTO(unlock, rc);
1550
1551                 OBD_ALLOC_PTR(attr);
1552                 if (attr == NULL)
1553                         GOTO(unlock, rc = -ENOMEM);
1554
1555                 /* validate the lsm */
1556                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1557                                    ll_md_blocking_ast);
1558                 if (rc != 0) {
1559                         OBD_FREE_PTR(attr);
1560                         GOTO(unlock, rc);
1561                 }
1562
1563                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1564                         md->body->mbo_nlink = attr->cat_nlink;
1565                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1566                         md->body->mbo_size = attr->cat_size;
1567                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1568                         md->body->mbo_atime = attr->cat_atime;
1569                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1570                         md->body->mbo_ctime = attr->cat_ctime;
1571                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1572                         md->body->mbo_mtime = attr->cat_mtime;
1573
1574                 OBD_FREE_PTR(attr);
1575         }
1576 unlock:
1577         up_read(&lli->lli_lsm_sem);
1578
1579         RETURN(rc);
1580 }
1581
1582 void ll_clear_inode(struct inode *inode)
1583 {
1584         struct ll_inode_info *lli = ll_i2info(inode);
1585         struct ll_sb_info *sbi = ll_i2sbi(inode);
1586
1587         ENTRY;
1588
1589         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1590                PFID(ll_inode2fid(inode)), inode);
1591
1592         if (S_ISDIR(inode->i_mode)) {
1593                 /* these should have been cleared in ll_file_release */
1594                 LASSERT(lli->lli_opendir_key == NULL);
1595                 LASSERT(lli->lli_sai == NULL);
1596                 LASSERT(lli->lli_opendir_pid == 0);
1597         } else {
1598                 pcc_inode_free(inode);
1599         }
1600
1601         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1602
1603         LASSERT(!lli->lli_open_fd_write_count);
1604         LASSERT(!lli->lli_open_fd_read_count);
1605         LASSERT(!lli->lli_open_fd_exec_count);
1606
1607         if (lli->lli_mds_write_och)
1608                 ll_md_real_close(inode, FMODE_WRITE);
1609         if (lli->lli_mds_exec_och)
1610                 ll_md_real_close(inode, FMODE_EXEC);
1611         if (lli->lli_mds_read_och)
1612                 ll_md_real_close(inode, FMODE_READ);
1613
1614         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1615                 OBD_FREE(lli->lli_symlink_name,
1616                          strlen(lli->lli_symlink_name) + 1);
1617                 lli->lli_symlink_name = NULL;
1618         }
1619
1620         ll_xattr_cache_destroy(inode);
1621
1622 #ifdef CONFIG_FS_POSIX_ACL
1623         forget_all_cached_acls(inode);
1624         if (lli->lli_posix_acl) {
1625                 posix_acl_release(lli->lli_posix_acl);
1626                 lli->lli_posix_acl = NULL;
1627         }
1628 #endif
1629         lli->lli_inode_magic = LLI_INODE_DEAD;
1630
1631         if (S_ISDIR(inode->i_mode))
1632                 ll_dir_clear_lsm_md(inode);
1633         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1634                 LASSERT(list_empty(&lli->lli_agl_list));
1635
1636         /*
1637          * XXX This has to be done before lsm is freed below, because
1638          * cl_object still uses inode lsm.
1639          */
1640         cl_inode_fini(inode);
1641
1642         EXIT;
1643 }
1644
1645 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1646 {
1647         struct lustre_md md;
1648         struct inode *inode = dentry->d_inode;
1649         struct ll_sb_info *sbi = ll_i2sbi(inode);
1650         struct ptlrpc_request *request = NULL;
1651         int rc, ia_valid;
1652         ENTRY;
1653
1654         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1655                                      LUSTRE_OPC_ANY, NULL);
1656         if (IS_ERR(op_data))
1657                 RETURN(PTR_ERR(op_data));
1658
1659         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1660         if (rc) {
1661                 ptlrpc_req_finished(request);
1662                 if (rc == -ENOENT) {
1663                         clear_nlink(inode);
1664                         /* Unlinked special device node? Or just a race?
1665                          * Pretend we done everything. */
1666                         if (!S_ISREG(inode->i_mode) &&
1667                             !S_ISDIR(inode->i_mode)) {
1668                                 ia_valid = op_data->op_attr.ia_valid;
1669                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1670                                 rc = simple_setattr(dentry, &op_data->op_attr);
1671                                 op_data->op_attr.ia_valid = ia_valid;
1672                         }
1673                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1674                         CERROR("md_setattr fails: rc = %d\n", rc);
1675                 }
1676                 RETURN(rc);
1677         }
1678
1679         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1680                               sbi->ll_md_exp, &md);
1681         if (rc) {
1682                 ptlrpc_req_finished(request);
1683                 RETURN(rc);
1684         }
1685
1686         ia_valid = op_data->op_attr.ia_valid;
1687         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1688          * cache is not cleared yet. */
1689         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1690         if (S_ISREG(inode->i_mode))
1691                 inode_lock(inode);
1692         rc = simple_setattr(dentry, &op_data->op_attr);
1693         if (S_ISREG(inode->i_mode))
1694                 inode_unlock(inode);
1695         op_data->op_attr.ia_valid = ia_valid;
1696
1697         rc = ll_update_inode(inode, &md);
1698         ptlrpc_req_finished(request);
1699
1700         RETURN(rc);
1701 }
1702
1703 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1704  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1705  * keep these values until such a time that objects are allocated for it.
1706  * We do the MDS operations first, as it is checking permissions for us.
1707  * We don't to the MDS RPC if there is nothing that we want to store there,
1708  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1709  * going to do an RPC anyways.
1710  *
1711  * If we are doing a truncate, we will send the mtime and ctime updates
1712  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1713  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1714  * at the same time.
1715  *
1716  * In case of HSMimport, we only set attr on MDS.
1717  */
1718 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1719                    enum op_xvalid xvalid, bool hsm_import)
1720 {
1721         struct inode *inode = dentry->d_inode;
1722         struct ll_inode_info *lli = ll_i2info(inode);
1723         struct md_op_data *op_data = NULL;
1724         int rc = 0;
1725
1726         ENTRY;
1727
1728         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1729                "valid %x, hsm_import %d\n",
1730                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
1731                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1732                hsm_import);
1733
1734         if (attr->ia_valid & ATTR_SIZE) {
1735                 /* Check new size against VFS/VM file size limit and rlimit */
1736                 rc = inode_newsize_ok(inode, attr->ia_size);
1737                 if (rc)
1738                         RETURN(rc);
1739
1740                 /* The maximum Lustre file size is variable, based on the
1741                  * OST maximum object size and number of stripes.  This
1742                  * needs another check in addition to the VFS check above. */
1743                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1744                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1745                                PFID(&lli->lli_fid), attr->ia_size,
1746                                ll_file_maxbytes(inode));
1747                         RETURN(-EFBIG);
1748                 }
1749
1750                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1751         }
1752
1753         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1754         if (attr->ia_valid & TIMES_SET_FLAGS) {
1755                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1756                     !cfs_capable(CFS_CAP_FOWNER))
1757                         RETURN(-EPERM);
1758         }
1759
1760         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1761         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1762              (attr->ia_valid & ATTR_CTIME)) {
1763                 attr->ia_ctime = current_time(inode);
1764                 xvalid |= OP_XVALID_CTIME_SET;
1765         }
1766         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1767             (attr->ia_valid & ATTR_ATIME)) {
1768                 attr->ia_atime = current_time(inode);
1769                 attr->ia_valid |= ATTR_ATIME_SET;
1770         }
1771         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1772             (attr->ia_valid & ATTR_MTIME)) {
1773                 attr->ia_mtime = current_time(inode);
1774                 attr->ia_valid |= ATTR_MTIME_SET;
1775         }
1776
1777         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1778                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1779                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1780                        ktime_get_real_seconds());
1781
1782         if (S_ISREG(inode->i_mode)) {
1783                 if (attr->ia_valid & ATTR_SIZE)
1784                         inode_dio_write_done(inode);
1785                 inode_unlock(inode);
1786         }
1787
1788         /* We always do an MDS RPC, even if we're only changing the size;
1789          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1790
1791         OBD_ALLOC_PTR(op_data);
1792         if (op_data == NULL)
1793                 GOTO(out, rc = -ENOMEM);
1794
1795         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1796                 /* If we are changing file size, file content is
1797                  * modified, flag it.
1798                  */
1799                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1800                 op_data->op_bias |= MDS_DATA_MODIFIED;
1801                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1802         }
1803
1804         if (attr->ia_valid & ATTR_FILE) {
1805                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1806
1807                 if (fd->fd_lease_och)
1808                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1809         }
1810
1811         op_data->op_attr = *attr;
1812         op_data->op_xvalid = xvalid;
1813
1814         rc = ll_md_setattr(dentry, op_data);
1815         if (rc)
1816                 GOTO(out, rc);
1817
1818         if (!S_ISREG(inode->i_mode) || hsm_import)
1819                 GOTO(out, rc = 0);
1820
1821         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1822                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1823             xvalid & OP_XVALID_CTIME_SET) {
1824                 bool cached = false;
1825
1826                 rc = pcc_inode_setattr(inode, attr, &cached);
1827                 if (cached) {
1828                         if (rc) {
1829                                 CERROR("%s: PCC inode "DFID" setattr failed: "
1830                                        "rc = %d\n",
1831                                        ll_i2sbi(inode)->ll_fsname,
1832                                        PFID(&lli->lli_fid), rc);
1833                                 GOTO(out, rc);
1834                         }
1835                 } else {
1836                         /* For truncate and utimes sending attributes to OSTs,
1837                          * setting mtime/atime to the past will be performed
1838                          * under PW [0:EOF] extent lock (new_size:EOF for
1839                          * truncate). It may seem excessive to send mtime/atime
1840                          * updates to OSTs when not setting times to past, but
1841                          * it is necessary due to possible time
1842                          * de-synchronization between MDT inode and OST objects
1843                          */
1844                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1845                 }
1846         }
1847
1848         /* If the file was restored, it needs to set dirty flag.
1849          *
1850          * We've already sent MDS_DATA_MODIFIED flag in
1851          * ll_md_setattr() for truncate. However, the MDT refuses to
1852          * set the HS_DIRTY flag on released files, so we have to set
1853          * it again if the file has been restored. Please check how
1854          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1855          *
1856          * Please notice that if the file is not released, the previous
1857          * MDS_DATA_MODIFIED has taken effect and usually
1858          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1859          * This way we can save an RPC for common open + trunc
1860          * operation. */
1861         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1862                 struct hsm_state_set hss = {
1863                         .hss_valid = HSS_SETMASK,
1864                         .hss_setmask = HS_DIRTY,
1865                 };
1866                 int rc2;
1867
1868                 rc2 = ll_hsm_state_set(inode, &hss);
1869                 /* truncate and write can happen at the same time, so that
1870                  * the file can be set modified even though the file is not
1871                  * restored from released state, and ll_hsm_state_set() is
1872                  * not applicable for the file, and rc2 < 0 is normal in this
1873                  * case. */
1874                 if (rc2 < 0)
1875                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1876                                PFID(ll_inode2fid(inode)), rc2);
1877         }
1878
1879         EXIT;
1880 out:
1881         if (op_data != NULL)
1882                 ll_finish_md_op_data(op_data);
1883
1884         if (S_ISREG(inode->i_mode)) {
1885                 inode_lock(inode);
1886                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1887                         inode_dio_wait(inode);
1888                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1889                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1890                  * inode flags, so there is a gap where S_NOSEC is not set.
1891                  * This can cause a writer to take the i_mutex unnecessarily,
1892                  * but this is safe to do and should be rare. */
1893                 inode_has_no_xattr(inode);
1894         }
1895
1896         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1897                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1898
1899         return rc;
1900 }
1901
1902 int ll_setattr(struct dentry *de, struct iattr *attr)
1903 {
1904         int mode = de->d_inode->i_mode;
1905         enum op_xvalid xvalid = 0;
1906
1907         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1908                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1909                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1910
1911         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1912                                (ATTR_SIZE|ATTR_MODE)) &&
1913             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1914              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1915               !(attr->ia_mode & S_ISGID))))
1916                 attr->ia_valid |= ATTR_FORCE;
1917
1918         if ((attr->ia_valid & ATTR_MODE) &&
1919             (mode & S_ISUID) &&
1920             !(attr->ia_mode & S_ISUID) &&
1921             !(attr->ia_valid & ATTR_KILL_SUID))
1922                 attr->ia_valid |= ATTR_KILL_SUID;
1923
1924         if ((attr->ia_valid & ATTR_MODE) &&
1925             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1926             !(attr->ia_mode & S_ISGID) &&
1927             !(attr->ia_valid & ATTR_KILL_SGID))
1928                 attr->ia_valid |= ATTR_KILL_SGID;
1929
1930         return ll_setattr_raw(de, attr, xvalid, false);
1931 }
1932
1933 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1934                        u32 flags)
1935 {
1936         struct obd_statfs obd_osfs = { 0 };
1937         time64_t max_age;
1938         int rc;
1939
1940         ENTRY;
1941         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1942
1943         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1944         if (rc)
1945                 RETURN(rc);
1946
1947         osfs->os_type = LL_SUPER_MAGIC;
1948
1949         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1950               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1951
1952         if (osfs->os_state & OS_STATE_SUM)
1953                 GOTO(out, rc);
1954
1955         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1956                 flags |= OBD_STATFS_NODELAY;
1957
1958         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1959         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1960                 GOTO(out, rc = 0);
1961
1962         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1963                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1964                obd_osfs.os_files);
1965
1966         osfs->os_bsize = obd_osfs.os_bsize;
1967         osfs->os_blocks = obd_osfs.os_blocks;
1968         osfs->os_bfree = obd_osfs.os_bfree;
1969         osfs->os_bavail = obd_osfs.os_bavail;
1970
1971         /* If we have _some_ OSTs, but don't have as many free objects on the
1972          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1973          * to compensate, so that the "inodes in use" number is correct.
1974          * This should be kept in sync with lod_statfs() behaviour.
1975          */
1976         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1977                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1978                                  obd_osfs.os_ffree;
1979                 osfs->os_ffree = obd_osfs.os_ffree;
1980         }
1981
1982 out:
1983         RETURN(rc);
1984 }
1985 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1986 {
1987         struct super_block *sb = de->d_sb;
1988         struct obd_statfs osfs;
1989         __u64 fsid = huge_encode_dev(sb->s_dev);
1990         int rc;
1991
1992         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1993         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1994
1995         /* Some amount of caching on the client is allowed */
1996         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1997         if (rc)
1998                 return rc;
1999
2000         statfs_unpack(sfs, &osfs);
2001
2002         /* We need to downshift for all 32-bit kernels, because we can't
2003          * tell if the kernel is being called via sys_statfs64() or not.
2004          * Stop before overflowing f_bsize - in which case it is better
2005          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2006         if (sizeof(long) < 8) {
2007                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2008                         sfs->f_bsize <<= 1;
2009
2010                         osfs.os_blocks >>= 1;
2011                         osfs.os_bfree >>= 1;
2012                         osfs.os_bavail >>= 1;
2013                 }
2014         }
2015
2016         sfs->f_blocks = osfs.os_blocks;
2017         sfs->f_bfree = osfs.os_bfree;
2018         sfs->f_bavail = osfs.os_bavail;
2019         sfs->f_fsid.val[0] = (__u32)fsid;
2020         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2021         return 0;
2022 }
2023
2024 void ll_inode_size_lock(struct inode *inode)
2025 {
2026         struct ll_inode_info *lli;
2027
2028         LASSERT(!S_ISDIR(inode->i_mode));
2029
2030         lli = ll_i2info(inode);
2031         mutex_lock(&lli->lli_size_mutex);
2032 }
2033
2034 void ll_inode_size_unlock(struct inode *inode)
2035 {
2036         struct ll_inode_info *lli;
2037
2038         lli = ll_i2info(inode);
2039         mutex_unlock(&lli->lli_size_mutex);
2040 }
2041
2042 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2043 {
2044         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2045         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2046                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2047         else
2048                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2049 }
2050
2051 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2052 {
2053         struct ll_inode_info *lli = ll_i2info(inode);
2054         struct mdt_body *body = md->body;
2055         struct ll_sb_info *sbi = ll_i2sbi(inode);
2056         int rc = 0;
2057
2058         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2059                 rc = cl_file_inode_init(inode, md);
2060                 if (rc)
2061                         return rc;
2062         }
2063
2064         if (S_ISDIR(inode->i_mode)) {
2065                 rc = ll_update_lsm_md(inode, md);
2066                 if (rc != 0)
2067                         return rc;
2068         }
2069
2070 #ifdef CONFIG_FS_POSIX_ACL
2071         if (body->mbo_valid & OBD_MD_FLACL) {
2072                 spin_lock(&lli->lli_lock);
2073                 if (lli->lli_posix_acl)
2074                         posix_acl_release(lli->lli_posix_acl);
2075                 lli->lli_posix_acl = md->posix_acl;
2076                 spin_unlock(&lli->lli_lock);
2077         }
2078 #endif
2079         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2080                                         sbi->ll_flags & LL_SBI_32BIT_API);
2081         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2082
2083         if (body->mbo_valid & OBD_MD_FLATIME) {
2084                 if (body->mbo_atime > inode->i_atime.tv_sec)
2085                         inode->i_atime.tv_sec = body->mbo_atime;
2086                 lli->lli_atime = body->mbo_atime;
2087         }
2088
2089         if (body->mbo_valid & OBD_MD_FLMTIME) {
2090                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2091                         CDEBUG(D_INODE,
2092                                "setting ino %lu mtime from %lld to %llu\n",
2093                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2094                                body->mbo_mtime);
2095                         inode->i_mtime.tv_sec = body->mbo_mtime;
2096                 }
2097                 lli->lli_mtime = body->mbo_mtime;
2098         }
2099
2100         if (body->mbo_valid & OBD_MD_FLCTIME) {
2101                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2102                         inode->i_ctime.tv_sec = body->mbo_ctime;
2103                 lli->lli_ctime = body->mbo_ctime;
2104         }
2105
2106         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2107         if (body->mbo_valid & OBD_MD_FLFLAGS)
2108                 ll_update_inode_flags(inode, body->mbo_flags);
2109         if (body->mbo_valid & OBD_MD_FLMODE)
2110                 inode->i_mode = (inode->i_mode & S_IFMT) |
2111                                 (body->mbo_mode & ~S_IFMT);
2112
2113         if (body->mbo_valid & OBD_MD_FLTYPE)
2114                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2115                                 (body->mbo_mode & S_IFMT);
2116
2117         LASSERT(inode->i_mode != 0);
2118         if (S_ISREG(inode->i_mode))
2119                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
2120                                        LL_MAX_BLKSIZE_BITS);
2121         else
2122                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
2123
2124         if (body->mbo_valid & OBD_MD_FLUID)
2125                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2126         if (body->mbo_valid & OBD_MD_FLGID)
2127                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2128         if (body->mbo_valid & OBD_MD_FLPROJID)
2129                 lli->lli_projid = body->mbo_projid;
2130         if (body->mbo_valid & OBD_MD_FLNLINK)
2131                 set_nlink(inode, body->mbo_nlink);
2132         if (body->mbo_valid & OBD_MD_FLRDEV)
2133                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2134
2135         if (body->mbo_valid & OBD_MD_FLID) {
2136                 /* FID shouldn't be changed! */
2137                 if (fid_is_sane(&lli->lli_fid)) {
2138                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2139                                  "Trying to change FID "DFID
2140                                  " to the "DFID", inode "DFID"(%p)\n",
2141                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2142                                  PFID(ll_inode2fid(inode)), inode);
2143                 } else {
2144                         lli->lli_fid = body->mbo_fid1;
2145                 }
2146         }
2147
2148         LASSERT(fid_seq(&lli->lli_fid) != 0);
2149
2150         if (body->mbo_valid & OBD_MD_FLSIZE) {
2151                 i_size_write(inode, body->mbo_size);
2152
2153                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2154                        PFID(ll_inode2fid(inode)),
2155                        (unsigned long long)body->mbo_size);
2156
2157                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2158                         inode->i_blocks = body->mbo_blocks;
2159         }
2160
2161         if (body->mbo_valid & OBD_MD_TSTATE) {
2162                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2163                  * clear it when done to ensure to start again
2164                  * glimpsing updated attrs
2165                  */
2166                 if (body->mbo_t_state & MS_RESTORE)
2167                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2168                 else
2169                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2170         }
2171
2172         return 0;
2173 }
2174
2175 int ll_read_inode2(struct inode *inode, void *opaque)
2176 {
2177         struct lustre_md *md = opaque;
2178         struct ll_inode_info *lli = ll_i2info(inode);
2179         int     rc;
2180         ENTRY;
2181
2182         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2183                PFID(&lli->lli_fid), inode);
2184
2185         /* Core attributes from the MDS first.  This is a new inode, and
2186          * the VFS doesn't zero times in the core inode so we have to do
2187          * it ourselves.  They will be overwritten by either MDS or OST
2188          * attributes - we just need to make sure they aren't newer.
2189          */
2190         inode->i_mtime.tv_sec = 0;
2191         inode->i_atime.tv_sec = 0;
2192         inode->i_ctime.tv_sec = 0;
2193         inode->i_rdev = 0;
2194         rc = ll_update_inode(inode, md);
2195         if (rc != 0)
2196                 RETURN(rc);
2197
2198         /* OIDEBUG(inode); */
2199
2200 #ifdef HAVE_BACKING_DEV_INFO
2201         /* initializing backing dev info. */
2202         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2203 #endif
2204         if (S_ISREG(inode->i_mode)) {
2205                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2206                 inode->i_op = &ll_file_inode_operations;
2207                 inode->i_fop = sbi->ll_fop;
2208                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2209                 EXIT;
2210         } else if (S_ISDIR(inode->i_mode)) {
2211                 inode->i_op = &ll_dir_inode_operations;
2212                 inode->i_fop = &ll_dir_operations;
2213                 EXIT;
2214         } else if (S_ISLNK(inode->i_mode)) {
2215                 inode->i_op = &ll_fast_symlink_inode_operations;
2216                 EXIT;
2217         } else {
2218                 inode->i_op = &ll_special_inode_operations;
2219
2220                 init_special_inode(inode, inode->i_mode,
2221                                    inode->i_rdev);
2222
2223                 EXIT;
2224         }
2225
2226         return 0;
2227 }
2228
2229 void ll_delete_inode(struct inode *inode)
2230 {
2231         struct ll_inode_info *lli = ll_i2info(inode);
2232         struct address_space *mapping = &inode->i_data;
2233         unsigned long nrpages;
2234         ENTRY;
2235
2236         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2237                 /* It is last chance to write out dirty pages,
2238                  * otherwise we may lose data while umount.
2239                  *
2240                  * If i_nlink is 0 then just discard data. This is safe because
2241                  * local inode gets i_nlink 0 from server only for the last
2242                  * unlink, so that file is not opened somewhere else
2243                  */
2244                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2245                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2246         }
2247         truncate_inode_pages_final(mapping);
2248
2249         /* Workaround for LU-118: Note nrpages may not be totally updated when
2250          * truncate_inode_pages() returns, as there can be a page in the process
2251          * of deletion (inside __delete_from_page_cache()) in the specified
2252          * range. Thus mapping->nrpages can be non-zero when this function
2253          * returns even after truncation of the whole mapping.  Only do this if
2254          * npages isn't already zero.
2255          */
2256         nrpages = mapping->nrpages;
2257         if (nrpages) {
2258                 xa_lock_irq(&mapping->i_pages);
2259                 nrpages = mapping->nrpages;
2260                 xa_unlock_irq(&mapping->i_pages);
2261         } /* Workaround end */
2262
2263         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2264                  "see https://jira.whamcloud.com/browse/LU-118\n",
2265                  ll_i2sbi(inode)->ll_fsname,
2266                  PFID(ll_inode2fid(inode)), inode, nrpages);
2267
2268 #ifdef HAVE_SBOPS_EVICT_INODE
2269         ll_clear_inode(inode);
2270 #endif
2271         clear_inode(inode);
2272
2273         EXIT;
2274 }
2275
2276 int ll_iocontrol(struct inode *inode, struct file *file,
2277                  unsigned int cmd, unsigned long arg)
2278 {
2279         struct ll_sb_info *sbi = ll_i2sbi(inode);
2280         struct ptlrpc_request *req = NULL;
2281         int rc, flags = 0;
2282         ENTRY;
2283
2284         switch (cmd) {
2285         case FS_IOC_GETFLAGS: {
2286                 struct mdt_body *body;
2287                 struct md_op_data *op_data;
2288
2289                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2290                                              0, 0, LUSTRE_OPC_ANY,
2291                                              NULL);
2292                 if (IS_ERR(op_data))
2293                         RETURN(PTR_ERR(op_data));
2294
2295                 op_data->op_valid = OBD_MD_FLFLAGS;
2296                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2297                 ll_finish_md_op_data(op_data);
2298                 if (rc) {
2299                         CERROR("%s: failure inode "DFID": rc = %d\n",
2300                                sbi->ll_md_exp->exp_obd->obd_name,
2301                                PFID(ll_inode2fid(inode)), rc);
2302                         RETURN(-abs(rc));
2303                 }
2304
2305                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2306
2307                 flags = body->mbo_flags;
2308
2309                 ptlrpc_req_finished(req);
2310
2311                 RETURN(put_user(flags, (int __user *)arg));
2312         }
2313         case FS_IOC_SETFLAGS: {
2314                 struct iattr *attr;
2315                 struct md_op_data *op_data;
2316                 struct cl_object *obj;
2317                 struct fsxattr fa = { 0 };
2318
2319                 if (get_user(flags, (int __user *)arg))
2320                         RETURN(-EFAULT);
2321
2322                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2323                 if (flags & LUSTRE_PROJINHERIT_FL)
2324                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2325
2326                 rc = ll_ioctl_check_project(inode, &fa);
2327                 if (rc)
2328                         RETURN(rc);
2329
2330                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2331                                              LUSTRE_OPC_ANY, NULL);
2332                 if (IS_ERR(op_data))
2333                         RETURN(PTR_ERR(op_data));
2334
2335                 op_data->op_attr_flags = flags;
2336                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2337                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2338                 ll_finish_md_op_data(op_data);
2339                 ptlrpc_req_finished(req);
2340                 if (rc)
2341                         RETURN(rc);
2342
2343                 ll_update_inode_flags(inode, flags);
2344
2345                 obj = ll_i2info(inode)->lli_clob;
2346                 if (obj == NULL)
2347                         RETURN(0);
2348
2349                 OBD_ALLOC_PTR(attr);
2350                 if (attr == NULL)
2351                         RETURN(-ENOMEM);
2352
2353                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2354
2355                 OBD_FREE_PTR(attr);
2356                 RETURN(rc);
2357         }
2358         default:
2359                 RETURN(-ENOSYS);
2360         }
2361
2362         RETURN(0);
2363 }
2364
2365 int ll_flush_ctx(struct inode *inode)
2366 {
2367         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2368
2369         CDEBUG(D_SEC, "flush context for user %d\n",
2370                from_kuid(&init_user_ns, current_uid()));
2371
2372         obd_set_info_async(NULL, sbi->ll_md_exp,
2373                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2374                            0, NULL, NULL);
2375         obd_set_info_async(NULL, sbi->ll_dt_exp,
2376                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2377                            0, NULL, NULL);
2378         return 0;
2379 }
2380
2381 /* umount -f client means force down, don't save state */
2382 void ll_umount_begin(struct super_block *sb)
2383 {
2384         struct ll_sb_info *sbi = ll_s2sbi(sb);
2385         struct obd_device *obd;
2386         struct obd_ioctl_data *ioc_data;
2387         struct l_wait_info lwi;
2388         wait_queue_head_t waitq;
2389         ENTRY;
2390
2391         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2392                sb->s_count, atomic_read(&sb->s_active));
2393
2394         obd = class_exp2obd(sbi->ll_md_exp);
2395         if (obd == NULL) {
2396                 CERROR("Invalid MDC connection handle %#llx\n",
2397                        sbi->ll_md_exp->exp_handle.h_cookie);
2398                 EXIT;
2399                 return;
2400         }
2401         obd->obd_force = 1;
2402
2403         obd = class_exp2obd(sbi->ll_dt_exp);
2404         if (obd == NULL) {
2405                 CERROR("Invalid LOV connection handle %#llx\n",
2406                        sbi->ll_dt_exp->exp_handle.h_cookie);
2407                 EXIT;
2408                 return;
2409         }
2410         obd->obd_force = 1;
2411
2412         OBD_ALLOC_PTR(ioc_data);
2413         if (ioc_data) {
2414                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2415                               sizeof *ioc_data, ioc_data, NULL);
2416
2417                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2418                               sizeof *ioc_data, ioc_data, NULL);
2419
2420                 OBD_FREE_PTR(ioc_data);
2421         }
2422
2423         /* Really, we'd like to wait until there are no requests outstanding,
2424          * and then continue.  For now, we just periodically checking for vfs
2425          * to decrement mnt_cnt and hope to finish it within 10sec.
2426          */
2427         init_waitqueue_head(&waitq);
2428         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2429                                    cfs_time_seconds(1), NULL, NULL);
2430         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2431
2432         EXIT;
2433 }
2434
2435 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2436 {
2437         struct ll_sb_info *sbi = ll_s2sbi(sb);
2438         char *profilenm = get_profile_name(sb);
2439         int err;
2440         __u32 read_only;
2441
2442         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2443                 read_only = *flags & MS_RDONLY;
2444                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2445                                          sizeof(KEY_READ_ONLY),
2446                                          KEY_READ_ONLY, sizeof(read_only),
2447                                          &read_only, NULL);
2448                 if (err) {
2449                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2450                                       profilenm, read_only ?
2451                                       "read-only" : "read-write", err);
2452                         return err;
2453                 }
2454
2455                 if (read_only)
2456                         sb->s_flags |= MS_RDONLY;
2457                 else
2458                         sb->s_flags &= ~MS_RDONLY;
2459
2460                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2461                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2462                                       read_only ?  "read-only" : "read-write");
2463         }
2464         return 0;
2465 }
2466
2467 /**
2468  * Cleanup the open handle that is cached on MDT-side.
2469  *
2470  * For open case, the client side open handling thread may hit error
2471  * after the MDT grant the open. Under such case, the client should
2472  * send close RPC to the MDT as cleanup; otherwise, the open handle
2473  * on the MDT will be leaked there until the client umount or evicted.
2474  *
2475  * In further, if someone unlinked the file, because the open handle
2476  * holds the reference on such file/object, then it will block the
2477  * subsequent threads that want to locate such object via FID.
2478  *
2479  * \param[in] sb        super block for this file-system
2480  * \param[in] open_req  pointer to the original open request
2481  */
2482 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2483 {
2484         struct mdt_body                 *body;
2485         struct md_op_data               *op_data;
2486         struct ptlrpc_request           *close_req = NULL;
2487         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2488         ENTRY;
2489
2490         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2491         OBD_ALLOC_PTR(op_data);
2492         if (op_data == NULL) {
2493                 CWARN("%s: cannot allocate op_data to release open handle for "
2494                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2495
2496                 RETURN_EXIT;
2497         }
2498
2499         op_data->op_fid1 = body->mbo_fid1;
2500         op_data->op_open_handle = body->mbo_open_handle;
2501         op_data->op_mod_time = ktime_get_real_seconds();
2502         md_close(exp, op_data, NULL, &close_req);
2503         ptlrpc_req_finished(close_req);
2504         ll_finish_md_op_data(op_data);
2505
2506         EXIT;
2507 }
2508
2509 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2510                   struct super_block *sb, struct lookup_intent *it)
2511 {
2512         struct ll_sb_info *sbi = NULL;
2513         struct lustre_md md = { NULL };
2514         bool default_lmv_deleted = false;
2515         int rc;
2516
2517         ENTRY;
2518
2519         LASSERT(*inode || sb);
2520         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2521         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2522                               sbi->ll_md_exp, &md);
2523         if (rc != 0)
2524                 GOTO(out, rc);
2525
2526         /*
2527          * clear default_lmv only if intent_getattr reply doesn't contain it.
2528          * but it needs to be done after iget, check this early because
2529          * ll_update_lsm_md() may change md.
2530          */
2531         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2532             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2533                 default_lmv_deleted = true;
2534
2535         if (*inode) {
2536                 rc = ll_update_inode(*inode, &md);
2537                 if (rc != 0)
2538                         GOTO(out, rc);
2539         } else {
2540                 LASSERT(sb != NULL);
2541
2542                 /*
2543                  * At this point server returns to client's same fid as client
2544                  * generated for creating. So using ->fid1 is okay here.
2545                  */
2546                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2547                         CERROR("%s: Fid is insane "DFID"\n",
2548                                 sbi->ll_fsname,
2549                                 PFID(&md.body->mbo_fid1));
2550                         GOTO(out, rc = -EINVAL);
2551                 }
2552
2553                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2554                                              sbi->ll_flags & LL_SBI_32BIT_API),
2555                                  &md);
2556                 if (IS_ERR(*inode)) {
2557 #ifdef CONFIG_FS_POSIX_ACL
2558                         if (md.posix_acl) {
2559                                 posix_acl_release(md.posix_acl);
2560                                 md.posix_acl = NULL;
2561                         }
2562 #endif
2563                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2564                         *inode = NULL;
2565                         CERROR("new_inode -fatal: rc %d\n", rc);
2566                         GOTO(out, rc);
2567                 }
2568         }
2569
2570         /* Handling piggyback layout lock.
2571          * Layout lock can be piggybacked by getattr and open request.
2572          * The lsm can be applied to inode only if it comes with a layout lock
2573          * otherwise correct layout may be overwritten, for example:
2574          * 1. proc1: mdt returns a lsm but not granting layout
2575          * 2. layout was changed by another client
2576          * 3. proc2: refresh layout and layout lock granted
2577          * 4. proc1: to apply a stale layout */
2578         if (it != NULL && it->it_lock_mode != 0) {
2579                 struct lustre_handle lockh;
2580                 struct ldlm_lock *lock;
2581
2582                 lockh.cookie = it->it_lock_handle;
2583                 lock = ldlm_handle2lock(&lockh);
2584                 LASSERT(lock != NULL);
2585                 if (ldlm_has_layout(lock)) {
2586                         struct cl_object_conf conf;
2587
2588                         memset(&conf, 0, sizeof(conf));
2589                         conf.coc_opc = OBJECT_CONF_SET;
2590                         conf.coc_inode = *inode;
2591                         conf.coc_lock = lock;
2592                         conf.u.coc_layout = md.layout;
2593                         (void)ll_layout_conf(*inode, &conf);
2594                 }
2595                 LDLM_LOCK_PUT(lock);
2596         }
2597
2598         if (default_lmv_deleted)
2599                 ll_update_default_lsm_md(*inode, &md);
2600
2601         GOTO(out, rc = 0);
2602
2603 out:
2604         /* cleanup will be done if necessary */
2605         md_free_lustre_md(sbi->ll_md_exp, &md);
2606
2607         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2608                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2609
2610         return rc;
2611 }
2612
2613 int ll_obd_statfs(struct inode *inode, void __user *arg)
2614 {
2615         struct ll_sb_info *sbi = NULL;
2616         struct obd_export *exp;
2617         char *buf = NULL;
2618         struct obd_ioctl_data *data = NULL;
2619         __u32 type;
2620         int len = 0, rc;
2621
2622         if (!inode || !(sbi = ll_i2sbi(inode)))
2623                 GOTO(out_statfs, rc = -EINVAL);
2624
2625         rc = obd_ioctl_getdata(&buf, &len, arg);
2626         if (rc)
2627                 GOTO(out_statfs, rc);
2628
2629         data = (void*)buf;
2630         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2631             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2632                 GOTO(out_statfs, rc = -EINVAL);
2633
2634         if (data->ioc_inllen1 != sizeof(__u32) ||
2635             data->ioc_inllen2 != sizeof(__u32) ||
2636             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2637             data->ioc_plen2 != sizeof(struct obd_uuid))
2638                 GOTO(out_statfs, rc = -EINVAL);
2639
2640         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2641         if (type & LL_STATFS_LMV)
2642                 exp = sbi->ll_md_exp;
2643         else if (type & LL_STATFS_LOV)
2644                 exp = sbi->ll_dt_exp;
2645         else
2646                 GOTO(out_statfs, rc = -ENODEV);
2647
2648         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2649         if (rc)
2650                 GOTO(out_statfs, rc);
2651 out_statfs:
2652         OBD_FREE_LARGE(buf, len);
2653         return rc;
2654 }
2655
2656 /*
2657  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2658  * be called early to avoid deadlock.
2659  */
2660 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2661 {
2662         if (op_data->op_mea2_sem) {
2663                 up_read(op_data->op_mea2_sem);
2664                 op_data->op_mea2_sem = NULL;
2665         }
2666
2667         if (op_data->op_mea1_sem) {
2668                 up_read(op_data->op_mea1_sem);
2669                 op_data->op_mea1_sem = NULL;
2670         }
2671 }
2672
2673 /* this function prepares md_op_data hint for passing it down to MD stack. */
2674 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2675                                       struct inode *i1, struct inode *i2,
2676                                       const char *name, size_t namelen,
2677                                       __u32 mode, enum md_op_code opc,
2678                                       void *data)
2679 {
2680         LASSERT(i1 != NULL);
2681
2682         if (name == NULL) {
2683                 /* Do not reuse namelen for something else. */
2684                 if (namelen != 0)
2685                         return ERR_PTR(-EINVAL);
2686         } else {
2687                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2688                         return ERR_PTR(-ENAMETOOLONG);
2689
2690                 if (!lu_name_is_valid_2(name, namelen))
2691                         return ERR_PTR(-EINVAL);
2692         }
2693
2694         if (op_data == NULL)
2695                 OBD_ALLOC_PTR(op_data);
2696
2697         if (op_data == NULL)
2698                 return ERR_PTR(-ENOMEM);
2699
2700         ll_i2gids(op_data->op_suppgids, i1, i2);
2701         op_data->op_fid1 = *ll_inode2fid(i1);
2702         op_data->op_code = opc;
2703
2704         if (S_ISDIR(i1->i_mode)) {
2705                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2706                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2707                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2708                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
2709         }
2710
2711         if (i2) {
2712                 op_data->op_fid2 = *ll_inode2fid(i2);
2713                 if (S_ISDIR(i2->i_mode)) {
2714                         if (i2 != i1) {
2715                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2716                                 op_data->op_mea2_sem =
2717                                                 &ll_i2info(i2)->lli_lsm_sem;
2718                         }
2719                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2720                 }
2721         } else {
2722                 fid_zero(&op_data->op_fid2);
2723         }
2724
2725         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2726                 op_data->op_cli_flags |= CLI_HASH64;
2727
2728         if (ll_need_32bit_api(ll_i2sbi(i1)))
2729                 op_data->op_cli_flags |= CLI_API32;
2730
2731         op_data->op_name = name;
2732         op_data->op_namelen = namelen;
2733         op_data->op_mode = mode;
2734         op_data->op_mod_time = ktime_get_real_seconds();
2735         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2736         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2737         op_data->op_cap = cfs_curproc_cap_pack();
2738         op_data->op_mds = 0;
2739         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2740              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2741                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2742         }
2743         op_data->op_data = data;
2744
2745         return op_data;
2746 }
2747
2748 void ll_finish_md_op_data(struct md_op_data *op_data)
2749 {
2750         ll_unlock_md_op_lsm(op_data);
2751         security_release_secctx(op_data->op_file_secctx,
2752                                 op_data->op_file_secctx_size);
2753         OBD_FREE_PTR(op_data);
2754 }
2755
2756 #ifdef HAVE_SUPEROPS_USE_DENTRY
2757 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2758 #else
2759 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2760 #endif
2761 {
2762         struct ll_sb_info *sbi;
2763
2764 #ifdef HAVE_SUPEROPS_USE_DENTRY
2765         LASSERT((seq != NULL) && (dentry != NULL));
2766         sbi = ll_s2sbi(dentry->d_sb);
2767 #else
2768         LASSERT((seq != NULL) && (vfs != NULL));
2769         sbi = ll_s2sbi(vfs->mnt_sb);
2770 #endif
2771
2772         if (sbi->ll_flags & LL_SBI_NOLCK)
2773                 seq_puts(seq, ",nolock");
2774
2775         if (sbi->ll_flags & LL_SBI_FLOCK)
2776                 seq_puts(seq, ",flock");
2777
2778         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2779                 seq_puts(seq, ",localflock");
2780
2781         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2782                 seq_puts(seq, ",user_xattr");
2783
2784         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2785                 seq_puts(seq, ",lazystatfs");
2786
2787         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2788                 seq_puts(seq, ",user_fid2path");
2789
2790         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2791                 seq_puts(seq, ",always_ping");
2792
2793         RETURN(0);
2794 }
2795
2796 /**
2797  * Get obd name by cmd, and copy out to user space
2798  */
2799 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2800 {
2801         struct ll_sb_info *sbi = ll_i2sbi(inode);
2802         struct obd_device *obd;
2803         ENTRY;
2804
2805         if (cmd == OBD_IOC_GETDTNAME)
2806                 obd = class_exp2obd(sbi->ll_dt_exp);
2807         else if (cmd == OBD_IOC_GETMDNAME)
2808                 obd = class_exp2obd(sbi->ll_md_exp);
2809         else
2810                 RETURN(-EINVAL);
2811
2812         if (!obd)
2813                 RETURN(-ENOENT);
2814
2815         if (copy_to_user((void __user *)arg, obd->obd_name,
2816                          strlen(obd->obd_name) + 1))
2817                 RETURN(-EFAULT);
2818
2819         RETURN(0);
2820 }
2821
2822 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2823 {
2824         char *path = NULL;
2825
2826         struct path p;
2827
2828         p.dentry = dentry;
2829         p.mnt = current->fs->root.mnt;
2830         path_get(&p);
2831         path = d_path(&p, buf, bufsize);
2832         path_put(&p);
2833         return path;
2834 }
2835
2836 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2837 {
2838         char *buf, *path = NULL;
2839         struct dentry *dentry = NULL;
2840         struct inode *inode = page->mapping->host;
2841
2842         /* this can be called inside spin lock so use GFP_ATOMIC. */
2843         buf = (char *)__get_free_page(GFP_ATOMIC);
2844         if (buf != NULL) {
2845                 dentry = d_find_alias(page->mapping->host);
2846                 if (dentry != NULL)
2847                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2848         }
2849
2850         CDEBUG(D_WARNING,
2851                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2852                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
2853                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2854                PFID(ll_inode2fid(inode)),
2855                (path && !IS_ERR(path)) ? path : "", ioret);
2856
2857         if (dentry != NULL)
2858                 dput(dentry);
2859
2860         if (buf != NULL)
2861                 free_page((unsigned long)buf);
2862 }
2863
2864 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2865                         struct lov_user_md **kbuf)
2866 {
2867         struct lov_user_md      lum;
2868         ssize_t                 lum_size;
2869         ENTRY;
2870
2871         if (copy_from_user(&lum, md, sizeof(lum)))
2872                 RETURN(-EFAULT);
2873
2874         lum_size = ll_lov_user_md_size(&lum);
2875         if (lum_size < 0)
2876                 RETURN(lum_size);
2877
2878         OBD_ALLOC(*kbuf, lum_size);
2879         if (*kbuf == NULL)
2880                 RETURN(-ENOMEM);
2881
2882         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2883                 OBD_FREE(*kbuf, lum_size);
2884                 RETURN(-EFAULT);
2885         }
2886
2887         RETURN(lum_size);
2888 }
2889
2890 /*
2891  * Compute llite root squash state after a change of root squash
2892  * configuration setting or add/remove of a lnet nid
2893  */
2894 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2895 {
2896         struct root_squash_info *squash = &sbi->ll_squash;
2897         int i;
2898         bool matched;
2899         struct lnet_process_id id;
2900
2901         /* Update norootsquash flag */
2902         down_write(&squash->rsi_sem);
2903         if (list_empty(&squash->rsi_nosquash_nids))
2904                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2905         else {
2906                 /* Do not apply root squash as soon as one of our NIDs is
2907                  * in the nosquash_nids list */
2908                 matched = false;
2909                 i = 0;
2910                 while (LNetGetId(i++, &id) != -ENOENT) {
2911                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2912                                 continue;
2913                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2914                                 matched = true;
2915                                 break;
2916                         }
2917                 }
2918                 if (matched)
2919                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2920                 else
2921                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2922         }
2923         up_write(&squash->rsi_sem);
2924 }
2925
2926 /**
2927  * Parse linkea content to extract information about a given hardlink
2928  *
2929  * \param[in]   ldata      - Initialized linkea data
2930  * \param[in]   linkno     - Link identifier
2931  * \param[out]  parent_fid - The entry's parent FID
2932  * \param[out]  ln         - Entry name destination buffer
2933  *
2934  * \retval 0 on success
2935  * \retval Appropriate negative error code on failure
2936  */
2937 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2938                             struct lu_fid *parent_fid, struct lu_name *ln)
2939 {
2940         unsigned int    idx;
2941         int             rc;
2942         ENTRY;
2943
2944         rc = linkea_init_with_rec(ldata);
2945         if (rc < 0)
2946                 RETURN(rc);
2947
2948         if (linkno >= ldata->ld_leh->leh_reccount)
2949                 /* beyond last link */
2950                 RETURN(-ENODATA);
2951
2952         linkea_first_entry(ldata);
2953         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2954                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2955                                     parent_fid);
2956                 if (idx == linkno)
2957                         break;
2958
2959                 linkea_next_entry(ldata);
2960         }
2961
2962         if (idx < linkno)
2963                 RETURN(-ENODATA);
2964
2965         RETURN(0);
2966 }
2967
2968 /**
2969  * Get parent FID and name of an identified link. Operation is performed for
2970  * a given link number, letting the caller iterate over linkno to list one or
2971  * all links of an entry.
2972  *
2973  * \param[in]     file - File descriptor against which to perform the operation
2974  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2975  *                       on and the available size. It is eventually filled with
2976  *                       the requested information or left untouched on error
2977  *
2978  * \retval - 0 on success
2979  * \retval - Appropriate negative error code on failure
2980  */
2981 int ll_getparent(struct file *file, struct getparent __user *arg)
2982 {
2983         struct inode            *inode = file_inode(file);
2984         struct linkea_data      *ldata;
2985         struct lu_buf            buf = LU_BUF_NULL;
2986         struct lu_name           ln;
2987         struct lu_fid            parent_fid;
2988         __u32                    linkno;
2989         __u32                    name_size;
2990         int                      rc;
2991
2992         ENTRY;
2993
2994         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2995             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2996                 RETURN(-EPERM);
2997
2998         if (get_user(name_size, &arg->gp_name_size))
2999                 RETURN(-EFAULT);
3000
3001         if (get_user(linkno, &arg->gp_linkno))
3002                 RETURN(-EFAULT);
3003
3004         if (name_size > PATH_MAX)
3005                 RETURN(-EINVAL);
3006
3007         OBD_ALLOC(ldata, sizeof(*ldata));
3008         if (ldata == NULL)
3009                 RETURN(-ENOMEM);
3010
3011         rc = linkea_data_new(ldata, &buf);
3012         if (rc < 0)
3013                 GOTO(ldata_free, rc);
3014
3015 #ifdef HAVE_XATTR_HANDLER_FLAGS
3016         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3017                            buf.lb_len, OBD_MD_FLXATTR);
3018 #else
3019         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
3020                          buf.lb_len);
3021 #endif /* HAVE_XATTR_HANDLER_FLAGS */
3022         if (rc < 0)
3023                 GOTO(lb_free, rc);
3024
3025         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3026         if (rc < 0)
3027                 GOTO(lb_free, rc);
3028
3029         if (ln.ln_namelen >= name_size)
3030                 GOTO(lb_free, rc = -EOVERFLOW);
3031
3032         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3033                 GOTO(lb_free, rc = -EFAULT);
3034
3035         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3036                 GOTO(lb_free, rc = -EFAULT);
3037
3038         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3039                 GOTO(lb_free, rc = -EFAULT);
3040
3041 lb_free:
3042         lu_buf_free(&buf);
3043 ldata_free:
3044         OBD_FREE(ldata, sizeof(*ldata));
3045
3046         RETURN(rc);
3047 }