lustre/llite/llite_lib.c

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