lustre/llite/llite_lib.c

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