lustre/llite/llite_lib.c

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