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