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, sbi->ll_cache);
 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, sbi->ll_cache);
 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         sb->s_root = d_make_root(root);
 595         if (sb->s_root == NULL) {
 596                 CERROR("%s: can't make root dentry\n",
 597                         ll_get_fsname(sb, NULL, 0));
 598                 GOTO(out_root, err = -ENOMEM);
 599         }
 600 #ifdef HAVE_DCACHE_LOCK
 601         sb->s_root->d_op = &ll_d_ops;
 602 #endif
 603
 604         sbi->ll_sdev_orig = sb->s_dev;
 605
 606         /* We set sb->s_dev equal on all lustre clients in order to support
 607          * NFS export clustering.  NFSD requires that the FSID be the same
 608          * on all clients. */
 609         /* s_dev is also used in lt_compare() to compare two fs, but that is
 610          * only a node-local comparison. */
 611         uuid = obd_get_uuid(sbi->ll_md_exp);
 612         if (uuid != NULL)
 613                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 614
 615         if (data != NULL)
 616                 OBD_FREE_PTR(data);
 617         if (osfs != NULL)
 618                 OBD_FREE_PTR(osfs);
 619
 620         if (sbi->ll_dt_obd) {
 621                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 622                                         &sbi->ll_dt_obd->obd_kset.kobj,
 623                                         sbi->ll_dt_obd->obd_type->typ_name);
 624                 if (err < 0) {
 625                         CERROR("%s: could not register %s in llite: rc = %d\n",
 626                                dt, ll_get_fsname(sb, NULL, 0), err);
 627                         err = 0;
 628                 }
 629         }
 630
 631         if (sbi->ll_md_obd) {
 632                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 633                                         &sbi->ll_md_obd->obd_kset.kobj,
 634                                         sbi->ll_md_obd->obd_type->typ_name);
 635                 if (err < 0) {
 636                         CERROR("%s: could not register %s in llite: rc = %d\n",
 637                                md, ll_get_fsname(sb, NULL, 0), err);
 638                         err = 0;
 639                 }
 640         }
 641
 642         RETURN(err);
 643 out_root:
 644         if (root)
 645                 iput(root);
 646 out_lock_cn_cb:
 647         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 648 out_dt:
 649         obd_disconnect(sbi->ll_dt_exp);
 650         sbi->ll_dt_exp = NULL;
 651         sbi->ll_dt_obd = NULL;
 652 out_md_fid:
 653         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 654 out_md:
 655         obd_disconnect(sbi->ll_md_exp);
 656         sbi->ll_md_exp = NULL;
 657         sbi->ll_md_obd = NULL;
 658 out:
 659         if (data != NULL)
 660                 OBD_FREE_PTR(data);
 661         if (osfs != NULL)
 662                 OBD_FREE_PTR(osfs);
 663         return err;
 664 }
 665
 666 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 667 {
 668         int size, rc;
 669
 670         size = sizeof(*lmmsize);
 671         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 672                           KEY_MAX_EASIZE, &size, lmmsize);
 673         if (rc != 0) {
 674                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 675                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 676                 RETURN(rc);
 677         }
 678
 679         size = sizeof(int);
 680         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 681                           KEY_MAX_EASIZE, &size, lmmsize);
 682         if (rc)
 683                 CERROR("Get max mdsize error rc %d\n", rc);
 684
 685         RETURN(rc);
 686 }
 687
 688 /**
 689  * Get the value of the default_easize parameter.
 690  *
 691  * \see client_obd::cl_default_mds_easize
 692  *
 693  * \param[in] sbi       superblock info for this filesystem
 694  * \param[out] lmmsize  pointer to storage location for value
 695  *
 696  * \retval 0            on success
 697  * \retval negative     negated errno on failure
 698  */
 699 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 700 {
 701         int size, rc;
 702
 703         size = sizeof(int);
 704         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 705                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 706         if (rc)
 707                 CERROR("Get default mdsize error rc %d\n", rc);
 708
 709         RETURN(rc);
 710 }
 711
 712 /**
 713  * Set the default_easize parameter to the given value.
 714  *
 715  * \see client_obd::cl_default_mds_easize
 716  *
 717  * \param[in] sbi       superblock info for this filesystem
 718  * \param[in] lmmsize   the size to set
 719  *
 720  * \retval 0            on success
 721  * \retval negative     negated errno on failure
 722  */
 723 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 724 {
 725         int rc;
 726
 727         if (lmmsize < sizeof(struct lov_mds_md) ||
 728             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 729                 return -EINVAL;
 730
 731         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 732                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 733                                 sizeof(int), &lmmsize, NULL);
 734
 735         RETURN(rc);
 736 }
 737
 738 static void client_common_put_super(struct super_block *sb)
 739 {
 740         struct ll_sb_info *sbi = ll_s2sbi(sb);
 741         ENTRY;
 742
 743         cl_sb_fini(sb);
 744
 745         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 746         obd_disconnect(sbi->ll_dt_exp);
 747         sbi->ll_dt_exp = NULL;
 748
 749         ll_debugfs_unregister_super(sb);
 750
 751         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 752         obd_disconnect(sbi->ll_md_exp);
 753         sbi->ll_md_exp = NULL;
 754
 755         EXIT;
 756 }
 757
 758 void ll_kill_super(struct super_block *sb)
 759 {
 760         struct ll_sb_info *sbi;
 761         ENTRY;
 762
 763         /* not init sb ?*/
 764         if (!(sb->s_flags & MS_ACTIVE))
 765                 return;
 766
 767         sbi = ll_s2sbi(sb);
 768         /* we need restore s_dev from changed for clustred NFS before put_super
 769          * because new kernels have cached s_dev and change sb->s_dev in
 770          * put_super not affected real removing devices */
 771         if (sbi) {
 772                 sb->s_dev = sbi->ll_sdev_orig;
 773                 sbi->ll_umounting = 1;
 774
 775                 /* wait running statahead threads to quit */
 776                 while (atomic_read(&sbi->ll_sa_running) > 0) {
 777                         set_current_state(TASK_UNINTERRUPTIBLE);
 778                         schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC >> 3));
 779                 }
 780         }
 781
 782         EXIT;
 783 }
 784
 785 static inline int ll_set_opt(const char *opt, char *data, int fl)
 786 {
 787         if (strncmp(opt, data, strlen(opt)) != 0)
 788                 return 0;
 789         else
 790                 return fl;
 791 }
 792
 793 /* non-client-specific mount options are parsed in lmd_parse */
 794 static int ll_options(char *options, struct ll_sb_info *sbi)
 795 {
 796         int tmp;
 797         char *s1 = options, *s2;
 798         int *flags = &sbi->ll_flags;
 799         ENTRY;
 800
 801         if (!options)
 802                 RETURN(0);
 803
 804         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 805
 806         while (*s1) {
 807                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 808                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 809                 if (tmp) {
 810                         *flags |= tmp;
 811                         goto next;
 812                 }
 813                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 814                 if (tmp) {
 815                         *flags |= tmp;
 816                         goto next;
 817                 }
 818                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 819                 if (tmp) {
 820                         *flags |= tmp;
 821                         goto next;
 822                 }
 823                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 824                 if (tmp) {
 825                         *flags &= ~tmp;
 826                         goto next;
 827                 }
 828                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 829                 if (tmp) {
 830                         *flags |= tmp;
 831                         goto next;
 832                 }
 833                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 834                 if (tmp) {
 835                         *flags &= ~tmp;
 836                         goto next;
 837                 }
 838                 tmp = ll_set_opt("context", s1, 1);
 839                 if (tmp)
 840                         goto next;
 841                 tmp = ll_set_opt("fscontext", s1, 1);
 842                 if (tmp)
 843                         goto next;
 844                 tmp = ll_set_opt("defcontext", s1, 1);
 845                 if (tmp)
 846                         goto next;
 847                 tmp = ll_set_opt("rootcontext", s1, 1);
 848                 if (tmp)
 849                         goto next;
 850                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 851                 if (tmp) {
 852                         *flags |= tmp;
 853                         goto next;
 854                 }
 855                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 856                 if (tmp) {
 857                         *flags &= ~tmp;
 858                         goto next;
 859                 }
 860
 861                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 862                 if (tmp) {
 863                         *flags |= tmp;
 864                         sbi->ll_checksum_set = 1;
 865                         goto next;
 866                 }
 867                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 868                 if (tmp) {
 869                         *flags &= ~tmp;
 870                         sbi->ll_checksum_set = 1;
 871                         goto next;
 872                 }
 873                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 874                 if (tmp) {
 875                         *flags |= tmp;
 876                         goto next;
 877                 }
 878                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 879                 if (tmp) {
 880                         *flags &= ~tmp;
 881                         goto next;
 882                 }
 883                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 884                 if (tmp) {
 885                         *flags |= tmp;
 886                         goto next;
 887                 }
 888                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 889                 if (tmp) {
 890                         *flags &= ~tmp;
 891                         goto next;
 892                 }
 893                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
 894                 if (tmp) {
 895                         *flags |= tmp;
 896                         goto next;
 897                 }
 898                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
 899                 if (tmp) {
 900                         *flags |= tmp;
 901                         goto next;
 902                 }
 903                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
 904                 if (tmp) {
 905                         *flags &= ~tmp;
 906                         goto next;
 907                 }
 908                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
 909                 if (tmp) {
 910                         *flags |= tmp;
 911                         goto next;
 912                 }
 913                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
 914                                    s1);
 915                 RETURN(-EINVAL);
 916
 917 next:
 918                 /* Find next opt */
 919                 s2 = strchr(s1, ',');
 920                 if (s2 == NULL)
 921                         break;
 922                 s1 = s2 + 1;
 923         }
 924         RETURN(0);
 925 }
 926
 927 void ll_lli_init(struct ll_inode_info *lli)
 928 {
 929         lli->lli_inode_magic = LLI_INODE_MAGIC;
 930         lli->lli_flags = 0;
 931         spin_lock_init(&lli->lli_lock);
 932         lli->lli_posix_acl = NULL;
 933         /* Do not set lli_fid, it has been initialized already. */
 934         fid_zero(&lli->lli_pfid);
 935         lli->lli_mds_read_och = NULL;
 936         lli->lli_mds_write_och = NULL;
 937         lli->lli_mds_exec_och = NULL;
 938         lli->lli_open_fd_read_count = 0;
 939         lli->lli_open_fd_write_count = 0;
 940         lli->lli_open_fd_exec_count = 0;
 941         mutex_init(&lli->lli_och_mutex);
 942         spin_lock_init(&lli->lli_agl_lock);
 943         spin_lock_init(&lli->lli_layout_lock);
 944         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
 945         lli->lli_clob = NULL;
 946
 947         init_rwsem(&lli->lli_xattrs_list_rwsem);
 948         mutex_init(&lli->lli_xattrs_enq_lock);
 949
 950         LASSERT(lli->lli_vfs_inode.i_mode != 0);
 951         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
 952                 mutex_init(&lli->lli_readdir_mutex);
 953                 lli->lli_opendir_key = NULL;
 954                 lli->lli_sai = NULL;
 955                 spin_lock_init(&lli->lli_sa_lock);
 956                 lli->lli_opendir_pid = 0;
 957                 lli->lli_sa_enabled = 0;
 958                 lli->lli_def_stripe_offset = -1;
 959                 init_rwsem(&lli->lli_lsm_sem);
 960         } else {
 961                 mutex_init(&lli->lli_size_mutex);
 962                 lli->lli_symlink_name = NULL;
 963                 init_rwsem(&lli->lli_trunc_sem);
 964                 range_lock_tree_init(&lli->lli_write_tree);
 965                 init_rwsem(&lli->lli_glimpse_sem);
 966                 lli->lli_glimpse_time = ktime_set(0, 0);
 967                 INIT_LIST_HEAD(&lli->lli_agl_list);
 968                 lli->lli_agl_index = 0;
 969                 lli->lli_async_rc = 0;
 970                 spin_lock_init(&lli->lli_heat_lock);
 971                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
 972                 lli->lli_heat_flags = 0;
 973         }
 974         mutex_init(&lli->lli_layout_mutex);
 975         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
 976 }
 977
 978 #define MAX_STRING_SIZE 128
 979
 980 #ifndef HAVE_SUPER_SETUP_BDI_NAME
 981
 982 #define LSI_BDI_INITIALIZED     0x00400000
 983
 984 #ifndef HAVE_BDI_CAP_MAP_COPY
 985 # define BDI_CAP_MAP_COPY       0
 986 #endif
 987
 988 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
 989 {
 990         struct  lustre_sb_info *lsi = s2lsi(sb);
 991         char buf[MAX_STRING_SIZE];
 992         va_list args;
 993         int err;
 994
 995         err = bdi_init(&lsi->lsi_bdi);
 996         if (err)
 997                 return err;
 998
 999         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1000         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1001         lsi->lsi_bdi.name = "lustre";
1002         va_start(args, fmt);
1003         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1004         va_end(args);
1005         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1006         va_end(args);
1007         if (!err)
1008                 sb->s_bdi = &lsi->lsi_bdi;
1009
1010         return err;
1011 }
1012 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1013
1014 int ll_fill_super(struct super_block *sb, struct vfsmount *mnt)
1015 {
1016         struct  lustre_profile *lprof = NULL;
1017         struct  lustre_sb_info *lsi = s2lsi(sb);
1018         struct  ll_sb_info *sbi = NULL;
1019         char    *dt = NULL, *md = NULL;
1020         char    *profilenm = get_profile_name(sb);
1021         struct config_llog_instance *cfg;
1022         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1023         const int instlen = 16 + 2;
1024         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1025         char name[MAX_STRING_SIZE];
1026         int md_len = 0;
1027         int dt_len = 0;
1028         char *ptr;
1029         int len;
1030         int err;
1031
1032         ENTRY;
1033         /* for ASLR, to map between cfg_instance and hashed ptr */
1034         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1035                profilenm, cfg_instance, sb);
1036
1037         try_module_get(THIS_MODULE);
1038
1039         OBD_ALLOC_PTR(cfg);
1040         if (cfg == NULL)
1041                 GOTO(out_free_cfg, err = -ENOMEM);
1042
1043         /* client additional sb info */
1044         lsi->lsi_llsbi = sbi = ll_init_sbi();
1045         if (!sbi)
1046                 GOTO(out_free_cfg, err = -ENOMEM);
1047
1048         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1049         if (err)
1050                 GOTO(out_free_cfg, err);
1051
1052         err = super_setup_bdi_name(sb, "lustre-%016lx", cfg_instance);
1053         if (err)
1054                 GOTO(out_free_cfg, err);
1055
1056 #ifndef HAVE_DCACHE_LOCK
1057         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1058         sb->s_d_op = &ll_d_ops;
1059 #endif
1060         /* Get fsname */
1061         len = strlen(profilenm);
1062         ptr = strrchr(profilenm, '-');
1063         if (ptr && (strcmp(ptr, "-client") == 0))
1064                 len -= 7;
1065
1066         /* Mount info */
1067         snprintf(name, MAX_STRING_SIZE, "%.*s-%016lx", len,
1068                  profilenm, cfg_instance);
1069
1070         /* Call ll_debugfs_register_super() before lustre_process_log()
1071          * so that "llite.*.*" params can be processed correctly.
1072          */
1073         err = ll_debugfs_register_super(sb, name);
1074         if (err < 0) {
1075                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1076                        ll_get_fsname(sb, NULL, 0), err);
1077                 err = 0;
1078         }
1079
1080         /* The cfg_instance is a value unique to this super, in case some
1081          * joker tries to mount the same fs at two mount points.
1082          */
1083         cfg->cfg_instance = cfg_instance;
1084         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1085         cfg->cfg_callback = class_config_llog_handler;
1086         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1087         /* set up client obds */
1088         err = lustre_process_log(sb, profilenm, cfg);
1089         if (err < 0)
1090                 GOTO(out_debugfs, err);
1091
1092         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1093         lprof = class_get_profile(profilenm);
1094         if (lprof == NULL) {
1095                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1096                                    " read from the MGS.  Does that filesystem "
1097                                    "exist?\n", profilenm);
1098                 GOTO(out_debugfs, err = -EINVAL);
1099         }
1100         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1101                lprof->lp_md, lprof->lp_dt);
1102
1103         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1104         OBD_ALLOC(dt, dt_len);
1105         if (!dt)
1106                 GOTO(out_profile, err = -ENOMEM);
1107         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1108
1109         md_len = strlen(lprof->lp_md) + instlen + 2;
1110         OBD_ALLOC(md, md_len);
1111         if (!md)
1112                 GOTO(out_free_dt, err = -ENOMEM);
1113         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1114
1115         /* connections, registrations, sb setup */
1116         err = client_common_fill_super(sb, md, dt, mnt);
1117         if (err < 0)
1118                 GOTO(out_free_md, err);
1119
1120         sbi->ll_client_common_fill_super_succeeded = 1;
1121
1122 out_free_md:
1123         if (md)
1124                 OBD_FREE(md, md_len);
1125 out_free_dt:
1126         if (dt)
1127                 OBD_FREE(dt, dt_len);
1128 out_profile:
1129         if (lprof)
1130                 class_put_profile(lprof);
1131 out_debugfs:
1132         if (err < 0)
1133                 ll_debugfs_unregister_super(sb);
1134 out_free_cfg:
1135         if (cfg)
1136                 OBD_FREE_PTR(cfg);
1137
1138         if (err)
1139                 ll_put_super(sb);
1140         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1141                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1142         RETURN(err);
1143 } /* ll_fill_super */
1144
1145 void ll_put_super(struct super_block *sb)
1146 {
1147         struct config_llog_instance cfg, params_cfg;
1148         struct obd_device *obd;
1149         struct lustre_sb_info *lsi = s2lsi(sb);
1150         struct ll_sb_info *sbi = ll_s2sbi(sb);
1151         char *profilenm = get_profile_name(sb);
1152         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1153         long ccc_count;
1154         int next, force = 1, rc = 0;
1155         ENTRY;
1156
1157         if (!sbi)
1158                 GOTO(out_no_sbi, 0);
1159
1160         /* Should replace instance_id with something better for ASLR */
1161         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1162                profilenm, cfg_instance, sb);
1163
1164         cfg.cfg_instance = cfg_instance;
1165         lustre_end_log(sb, profilenm, &cfg);
1166
1167         params_cfg.cfg_instance = cfg_instance;
1168         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1169
1170         if (sbi->ll_md_exp) {
1171                 obd = class_exp2obd(sbi->ll_md_exp);
1172                 if (obd)
1173                         force = obd->obd_force;
1174         }
1175
1176         /* Wait for unstable pages to be committed to stable storage */
1177         if (force == 0) {
1178                 struct l_wait_info lwi = LWI_INTR(LWI_ON_SIGNAL_NOOP, NULL);
1179                 rc = l_wait_event(sbi->ll_cache->ccc_unstable_waitq,
1180                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0,
1181                         &lwi);
1182         }
1183
1184         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1185         if (force == 0 && rc != -EINTR)
1186                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1187
1188         /* We need to set force before the lov_disconnect in
1189            lustre_common_put_super, since l_d cleans up osc's as well. */
1190         if (force) {
1191                 next = 0;
1192                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1193                                                      &next)) != NULL) {
1194                         obd->obd_force = force;
1195                 }
1196         }
1197
1198         if (sbi->ll_client_common_fill_super_succeeded) {
1199                 /* Only if client_common_fill_super succeeded */
1200                 client_common_put_super(sb);
1201         }
1202
1203         next = 0;
1204         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)) !=NULL) {
1205                 class_manual_cleanup(obd);
1206         }
1207
1208         if (sbi->ll_flags & LL_SBI_VERBOSE)
1209                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1210
1211         if (profilenm)
1212                 class_del_profile(profilenm);
1213
1214 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1215         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1216                 bdi_destroy(&lsi->lsi_bdi);
1217                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1218         }
1219 #endif
1220
1221         ll_free_sbi(sb);
1222         lsi->lsi_llsbi = NULL;
1223 out_no_sbi:
1224         lustre_common_put_super(sb);
1225
1226         cl_env_cache_purge(~0);
1227
1228         module_put(THIS_MODULE);
1229
1230         EXIT;
1231 } /* client_put_super */
1232
1233 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1234 {
1235         struct inode *inode = NULL;
1236
1237         /* NOTE: we depend on atomic igrab() -bzzz */
1238         lock_res_and_lock(lock);
1239         if (lock->l_resource->lr_lvb_inode) {
1240                 struct ll_inode_info * lli;
1241                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1242                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1243                         inode = igrab(lock->l_resource->lr_lvb_inode);
1244                 } else {
1245                         inode = lock->l_resource->lr_lvb_inode;
1246                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1247                                          D_WARNING, lock, "lr_lvb_inode %p is "
1248                                          "bogus: magic %08x",
1249                                          lock->l_resource->lr_lvb_inode,
1250                                          lli->lli_inode_magic);
1251                         inode = NULL;
1252                 }
1253         }
1254         unlock_res_and_lock(lock);
1255         return inode;
1256 }
1257
1258 void ll_dir_clear_lsm_md(struct inode *inode)
1259 {
1260         struct ll_inode_info *lli = ll_i2info(inode);
1261
1262         LASSERT(S_ISDIR(inode->i_mode));
1263
1264         if (lli->lli_lsm_md != NULL) {
1265                 lmv_free_memmd(lli->lli_lsm_md);
1266                 lli->lli_lsm_md = NULL;
1267         }
1268 }
1269
1270 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1271                                       const struct lu_fid *fid,
1272                                       struct lustre_md *md)
1273 {
1274         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1275         struct mdt_body         *body = md->body;
1276         struct inode            *inode;
1277         ino_t                   ino;
1278         ENTRY;
1279
1280         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1281         inode = iget_locked(sb, ino);
1282         if (inode == NULL) {
1283                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1284                        ll_get_fsname(sb, NULL, 0), PFID(fid));
1285                 RETURN(ERR_PTR(-ENOENT));
1286         }
1287
1288         if (inode->i_state & I_NEW) {
1289                 struct ll_inode_info *lli = ll_i2info(inode);
1290                 struct lmv_stripe_md *lsm = md->lmv;
1291
1292                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1293                                 (body->mbo_mode & S_IFMT);
1294                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1295                          PFID(fid));
1296
1297                 inode->i_mtime.tv_sec = 0;
1298                 inode->i_atime.tv_sec = 0;
1299                 inode->i_ctime.tv_sec = 0;
1300                 inode->i_rdev = 0;
1301
1302 #ifdef HAVE_BACKING_DEV_INFO
1303                 /* initializing backing dev info. */
1304                 inode->i_mapping->backing_dev_info =
1305                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1306 #endif
1307                 inode->i_op = &ll_dir_inode_operations;
1308                 inode->i_fop = &ll_dir_operations;
1309                 lli->lli_fid = *fid;
1310                 ll_lli_init(lli);
1311
1312                 LASSERT(lsm != NULL);
1313                 /* master object FID */
1314                 lli->lli_pfid = body->mbo_fid1;
1315                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1316                        lli, PFID(fid), PFID(&lli->lli_pfid));
1317                 unlock_new_inode(inode);
1318         }
1319
1320         RETURN(inode);
1321 }
1322
1323 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1324 {
1325         struct lu_fid *fid;
1326         struct lmv_stripe_md *lsm = md->lmv;
1327         struct ll_inode_info *lli = ll_i2info(inode);
1328         int i;
1329
1330         LASSERT(lsm != NULL);
1331
1332         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1333                 ll_get_fsname(inode->i_sb, NULL, 0),
1334                 PFID(&lli->lli_fid));
1335         lsm_md_dump(D_INODE, lsm);
1336
1337         /* XXX sigh, this lsm_root initialization should be in
1338          * LMV layer, but it needs ll_iget right now, so we
1339          * put this here right now. */
1340         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1341                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1342                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1343                 /* Unfortunately ll_iget will call ll_update_inode,
1344                  * where the initialization of slave inode is slightly
1345                  * different, so it reset lsm_md to NULL to avoid
1346                  * initializing lsm for slave inode. */
1347                 lsm->lsm_md_oinfo[i].lmo_root =
1348                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1349                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1350                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1351
1352                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1353                         while (i-- > 0) {
1354                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1355                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1356                         }
1357                         return rc;
1358                 }
1359         }
1360
1361         lli->lli_lsm_md = lsm;
1362
1363         return 0;
1364 }
1365
1366 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1367 {
1368         struct ll_inode_info *lli = ll_i2info(inode);
1369         struct lmv_stripe_md *lsm = md->lmv;
1370         int rc = 0;
1371
1372         ENTRY;
1373
1374         LASSERT(S_ISDIR(inode->i_mode));
1375         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1376                PFID(ll_inode2fid(inode)));
1377
1378         /*
1379          * no striped information from request, lustre_md from req does not
1380          * include stripeEA, see ll_md_setattr()
1381          */
1382         if (!lsm)
1383                 RETURN(0);
1384
1385         /*
1386          * normally dir layout doesn't change, only take read lock to check
1387          * that to avoid blocking other MD operations.
1388          */
1389         if (lli->lli_lsm_md)
1390                 down_read(&lli->lli_lsm_sem);
1391         else
1392                 down_write(&lli->lli_lsm_sem);
1393
1394         /*
1395          * if dir layout mismatch, check whether version is increased, which
1396          * means layout is changed, this happens in dir migration and lfsck.
1397          */
1398         if (lli->lli_lsm_md && !lsm_md_eq(lli->lli_lsm_md, lsm)) {
1399                 if (lsm->lsm_md_layout_version <=
1400                     lli->lli_lsm_md->lsm_md_layout_version) {
1401                         CERROR("%s: "DFID" dir layout mismatch:\n",
1402                                 ll_get_fsname(inode->i_sb, NULL, 0),
1403                                 PFID(&lli->lli_fid));
1404                         lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1405                         lsm_md_dump(D_ERROR, lsm);
1406                         GOTO(unlock, rc = -EINVAL);
1407                 }
1408
1409                 /* layout changed, switch to write lock */
1410                 up_read(&lli->lli_lsm_sem);
1411                 down_write(&lli->lli_lsm_sem);
1412                 ll_dir_clear_lsm_md(inode);
1413         }
1414
1415         /* set directory layout */
1416         if (!lli->lli_lsm_md) {
1417                 struct cl_attr  *attr;
1418
1419                 rc = ll_init_lsm_md(inode, md);
1420                 up_write(&lli->lli_lsm_sem);
1421                 if (rc != 0)
1422                         RETURN(rc);
1423
1424                 /* set md->lmv to NULL, so the following free lustre_md
1425                  * will not free this lsm */
1426                 md->lmv = NULL;
1427
1428                 /*
1429                  * md_merge_attr() may take long, since lsm is already set,
1430                  * switch to read lock.
1431                  */
1432                 down_read(&lli->lli_lsm_sem);
1433
1434                 OBD_ALLOC_PTR(attr);
1435                 if (attr == NULL)
1436                         GOTO(unlock, rc = -ENOMEM);
1437
1438                 /* validate the lsm */
1439                 rc = md_merge_attr(ll_i2mdexp(inode), lsm, attr,
1440                                    ll_md_blocking_ast);
1441                 if (rc != 0) {
1442                         OBD_FREE_PTR(attr);
1443                         GOTO(unlock, rc);
1444                 }
1445
1446                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1447                         md->body->mbo_nlink = attr->cat_nlink;
1448                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1449                         md->body->mbo_size = attr->cat_size;
1450                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1451                         md->body->mbo_atime = attr->cat_atime;
1452                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1453                         md->body->mbo_ctime = attr->cat_ctime;
1454                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1455                         md->body->mbo_mtime = attr->cat_mtime;
1456
1457                 OBD_FREE_PTR(attr);
1458         }
1459 unlock:
1460         up_read(&lli->lli_lsm_sem);
1461
1462         RETURN(rc);
1463 }
1464
1465 void ll_clear_inode(struct inode *inode)
1466 {
1467         struct ll_inode_info *lli = ll_i2info(inode);
1468         struct ll_sb_info *sbi = ll_i2sbi(inode);
1469         ENTRY;
1470
1471         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1472                PFID(ll_inode2fid(inode)), inode);
1473
1474         if (S_ISDIR(inode->i_mode)) {
1475                 /* these should have been cleared in ll_file_release */
1476                 LASSERT(lli->lli_opendir_key == NULL);
1477                 LASSERT(lli->lli_sai == NULL);
1478                 LASSERT(lli->lli_opendir_pid == 0);
1479         }
1480
1481         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1482
1483         LASSERT(!lli->lli_open_fd_write_count);
1484         LASSERT(!lli->lli_open_fd_read_count);
1485         LASSERT(!lli->lli_open_fd_exec_count);
1486
1487         if (lli->lli_mds_write_och)
1488                 ll_md_real_close(inode, FMODE_WRITE);
1489         if (lli->lli_mds_exec_och)
1490                 ll_md_real_close(inode, FMODE_EXEC);
1491         if (lli->lli_mds_read_och)
1492                 ll_md_real_close(inode, FMODE_READ);
1493
1494         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1495                 OBD_FREE(lli->lli_symlink_name,
1496                          strlen(lli->lli_symlink_name) + 1);
1497                 lli->lli_symlink_name = NULL;
1498         }
1499
1500         ll_xattr_cache_destroy(inode);
1501
1502 #ifdef CONFIG_FS_POSIX_ACL
1503         forget_all_cached_acls(inode);
1504         if (lli->lli_posix_acl) {
1505                 posix_acl_release(lli->lli_posix_acl);
1506                 lli->lli_posix_acl = NULL;
1507         }
1508 #endif
1509         lli->lli_inode_magic = LLI_INODE_DEAD;
1510
1511         if (S_ISDIR(inode->i_mode))
1512                 ll_dir_clear_lsm_md(inode);
1513         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1514                 LASSERT(list_empty(&lli->lli_agl_list));
1515
1516         /*
1517          * XXX This has to be done before lsm is freed below, because
1518          * cl_object still uses inode lsm.
1519          */
1520         cl_inode_fini(inode);
1521
1522         EXIT;
1523 }
1524
1525 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1526 {
1527         struct lustre_md md;
1528         struct inode *inode = dentry->d_inode;
1529         struct ll_sb_info *sbi = ll_i2sbi(inode);
1530         struct ptlrpc_request *request = NULL;
1531         int rc, ia_valid;
1532         ENTRY;
1533
1534         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1535                                      LUSTRE_OPC_ANY, NULL);
1536         if (IS_ERR(op_data))
1537                 RETURN(PTR_ERR(op_data));
1538
1539         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1540         if (rc) {
1541                 ptlrpc_req_finished(request);
1542                 if (rc == -ENOENT) {
1543                         clear_nlink(inode);
1544                         /* Unlinked special device node? Or just a race?
1545                          * Pretend we done everything. */
1546                         if (!S_ISREG(inode->i_mode) &&
1547                             !S_ISDIR(inode->i_mode)) {
1548                                 ia_valid = op_data->op_attr.ia_valid;
1549                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1550                                 rc = simple_setattr(dentry, &op_data->op_attr);
1551                                 op_data->op_attr.ia_valid = ia_valid;
1552                         }
1553                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1554                         CERROR("md_setattr fails: rc = %d\n", rc);
1555                 }
1556                 RETURN(rc);
1557         }
1558
1559         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1560                               sbi->ll_md_exp, &md);
1561         if (rc) {
1562                 ptlrpc_req_finished(request);
1563                 RETURN(rc);
1564         }
1565
1566         ia_valid = op_data->op_attr.ia_valid;
1567         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1568          * cache is not cleared yet. */
1569         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1570         if (S_ISREG(inode->i_mode))
1571                 inode_lock(inode);
1572         rc = simple_setattr(dentry, &op_data->op_attr);
1573         if (S_ISREG(inode->i_mode))
1574                 inode_unlock(inode);
1575         op_data->op_attr.ia_valid = ia_valid;
1576
1577         rc = ll_update_inode(inode, &md);
1578         ptlrpc_req_finished(request);
1579
1580         RETURN(rc);
1581 }
1582
1583 /* If this inode has objects allocated to it (lsm != NULL), then the OST
1584  * object(s) determine the file size and mtime.  Otherwise, the MDS will
1585  * keep these values until such a time that objects are allocated for it.
1586  * We do the MDS operations first, as it is checking permissions for us.
1587  * We don't to the MDS RPC if there is nothing that we want to store there,
1588  * otherwise there is no harm in updating mtime/atime on the MDS if we are
1589  * going to do an RPC anyways.
1590  *
1591  * If we are doing a truncate, we will send the mtime and ctime updates
1592  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
1593  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
1594  * at the same time.
1595  *
1596  * In case of HSMimport, we only set attr on MDS.
1597  */
1598 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
1599                    enum op_xvalid xvalid, bool hsm_import)
1600 {
1601         struct inode *inode = dentry->d_inode;
1602         struct ll_inode_info *lli = ll_i2info(inode);
1603         struct md_op_data *op_data = NULL;
1604         int rc = 0;
1605         ENTRY;
1606
1607         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
1608                "valid %x, hsm_import %d\n",
1609                ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid),
1610                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
1611                hsm_import);
1612
1613         if (attr->ia_valid & ATTR_SIZE) {
1614                 /* Check new size against VFS/VM file size limit and rlimit */
1615                 rc = inode_newsize_ok(inode, attr->ia_size);
1616                 if (rc)
1617                         RETURN(rc);
1618
1619                 /* The maximum Lustre file size is variable, based on the
1620                  * OST maximum object size and number of stripes.  This
1621                  * needs another check in addition to the VFS check above. */
1622                 if (attr->ia_size > ll_file_maxbytes(inode)) {
1623                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
1624                                PFID(&lli->lli_fid), attr->ia_size,
1625                                ll_file_maxbytes(inode));
1626                         RETURN(-EFBIG);
1627                 }
1628
1629                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
1630         }
1631
1632         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
1633         if (attr->ia_valid & TIMES_SET_FLAGS) {
1634                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
1635                     !cfs_capable(CFS_CAP_FOWNER))
1636                         RETURN(-EPERM);
1637         }
1638
1639         /* We mark all of the fields "set" so MDS/OST does not re-set them */
1640         if (!(xvalid & OP_XVALID_CTIME_SET) &&
1641              (attr->ia_valid & ATTR_CTIME)) {
1642                 attr->ia_ctime = current_time(inode);
1643                 xvalid |= OP_XVALID_CTIME_SET;
1644         }
1645         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
1646             (attr->ia_valid & ATTR_ATIME)) {
1647                 attr->ia_atime = current_time(inode);
1648                 attr->ia_valid |= ATTR_ATIME_SET;
1649         }
1650         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
1651             (attr->ia_valid & ATTR_MTIME)) {
1652                 attr->ia_mtime = current_time(inode);
1653                 attr->ia_valid |= ATTR_MTIME_SET;
1654         }
1655
1656         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
1657                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
1658                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
1659                        ktime_get_real_seconds());
1660
1661         if (S_ISREG(inode->i_mode)) {
1662                 if (attr->ia_valid & ATTR_SIZE)
1663                         inode_dio_write_done(inode);
1664                 inode_unlock(inode);
1665         }
1666
1667         /* We always do an MDS RPC, even if we're only changing the size;
1668          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
1669
1670         OBD_ALLOC_PTR(op_data);
1671         if (op_data == NULL)
1672                 GOTO(out, rc = -ENOMEM);
1673
1674         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
1675                 /* If we are changing file size, file content is
1676                  * modified, flag it.
1677                  */
1678                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1679                 op_data->op_bias |= MDS_DATA_MODIFIED;
1680                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
1681         }
1682
1683         if (attr->ia_valid & ATTR_FILE) {
1684                 struct ll_file_data *fd = LUSTRE_FPRIVATE(attr->ia_file);
1685
1686                 if (fd->fd_lease_och)
1687                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
1688         }
1689
1690         op_data->op_attr = *attr;
1691         op_data->op_xvalid = xvalid;
1692
1693         rc = ll_md_setattr(dentry, op_data);
1694         if (rc)
1695                 GOTO(out, rc);
1696
1697         if (!S_ISREG(inode->i_mode) || hsm_import)
1698                 GOTO(out, rc = 0);
1699
1700         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
1701                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
1702             xvalid & OP_XVALID_CTIME_SET) {
1703                 /* For truncate and utimes sending attributes to OSTs, setting
1704                  * mtime/atime to the past will be performed under PW [0:EOF]
1705                  * extent lock (new_size:EOF for truncate).  It may seem
1706                  * excessive to send mtime/atime updates to OSTs when not
1707                  * setting times to past, but it is necessary due to possible
1708                  * time de-synchronization between MDT inode and OST objects
1709                  */
1710                 rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, 0);
1711         }
1712
1713         /* If the file was restored, it needs to set dirty flag.
1714          *
1715          * We've already sent MDS_DATA_MODIFIED flag in
1716          * ll_md_setattr() for truncate. However, the MDT refuses to
1717          * set the HS_DIRTY flag on released files, so we have to set
1718          * it again if the file has been restored. Please check how
1719          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
1720          *
1721          * Please notice that if the file is not released, the previous
1722          * MDS_DATA_MODIFIED has taken effect and usually
1723          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
1724          * This way we can save an RPC for common open + trunc
1725          * operation. */
1726         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
1727                 struct hsm_state_set hss = {
1728                         .hss_valid = HSS_SETMASK,
1729                         .hss_setmask = HS_DIRTY,
1730                 };
1731                 int rc2;
1732
1733                 rc2 = ll_hsm_state_set(inode, &hss);
1734                 /* truncate and write can happen at the same time, so that
1735                  * the file can be set modified even though the file is not
1736                  * restored from released state, and ll_hsm_state_set() is
1737                  * not applicable for the file, and rc2 < 0 is normal in this
1738                  * case. */
1739                 if (rc2 < 0)
1740                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
1741                                PFID(ll_inode2fid(inode)), rc2);
1742         }
1743
1744         EXIT;
1745 out:
1746         if (op_data != NULL)
1747                 ll_finish_md_op_data(op_data);
1748
1749         if (S_ISREG(inode->i_mode)) {
1750                 inode_lock(inode);
1751                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
1752                         inode_dio_wait(inode);
1753                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
1754                  * flag.  ll_update_inode (called from ll_md_setattr), clears
1755                  * inode flags, so there is a gap where S_NOSEC is not set.
1756                  * This can cause a writer to take the i_mutex unnecessarily,
1757                  * but this is safe to do and should be rare. */
1758                 inode_has_no_xattr(inode);
1759         }
1760
1761         ll_stats_ops_tally(ll_i2sbi(inode), (attr->ia_valid & ATTR_SIZE) ?
1762                         LPROC_LL_TRUNC : LPROC_LL_SETATTR, 1);
1763
1764         return rc;
1765 }
1766
1767 int ll_setattr(struct dentry *de, struct iattr *attr)
1768 {
1769         int mode = de->d_inode->i_mode;
1770         enum op_xvalid xvalid = 0;
1771
1772         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
1773                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
1774                 xvalid |= OP_XVALID_OWNEROVERRIDE;
1775
1776         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
1777                                (ATTR_SIZE|ATTR_MODE)) &&
1778             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
1779              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1780               !(attr->ia_mode & S_ISGID))))
1781                 attr->ia_valid |= ATTR_FORCE;
1782
1783         if ((attr->ia_valid & ATTR_MODE) &&
1784             (mode & S_ISUID) &&
1785             !(attr->ia_mode & S_ISUID) &&
1786             !(attr->ia_valid & ATTR_KILL_SUID))
1787                 attr->ia_valid |= ATTR_KILL_SUID;
1788
1789         if ((attr->ia_valid & ATTR_MODE) &&
1790             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
1791             !(attr->ia_mode & S_ISGID) &&
1792             !(attr->ia_valid & ATTR_KILL_SGID))
1793                 attr->ia_valid |= ATTR_KILL_SGID;
1794
1795         return ll_setattr_raw(de, attr, xvalid, false);
1796 }
1797
1798 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
1799                        u32 flags)
1800 {
1801         struct obd_statfs obd_osfs = { 0 };
1802         time64_t max_age;
1803         int rc;
1804
1805         ENTRY;
1806         max_age = ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS;
1807
1808         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
1809         if (rc)
1810                 RETURN(rc);
1811
1812         osfs->os_type = LL_SUPER_MAGIC;
1813
1814         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
1815               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
1816
1817         if (osfs->os_state & OS_STATE_SUM)
1818                 GOTO(out, rc);
1819
1820         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
1821                 flags |= OBD_STATFS_NODELAY;
1822
1823         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
1824         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
1825                 GOTO(out, rc = 0);
1826
1827         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
1828                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
1829                obd_osfs.os_files);
1830
1831         osfs->os_bsize = obd_osfs.os_bsize;
1832         osfs->os_blocks = obd_osfs.os_blocks;
1833         osfs->os_bfree = obd_osfs.os_bfree;
1834         osfs->os_bavail = obd_osfs.os_bavail;
1835
1836         /* If we have _some_ OSTs, but don't have as many free objects on the
1837          * OSTs as inodes on the MDTs, reduce the reported number of inodes
1838          * to compensate, so that the "inodes in use" number is correct.
1839          * This should be kept in sync with lod_statfs() behaviour.
1840          */
1841         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
1842                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
1843                                  obd_osfs.os_ffree;
1844                 osfs->os_ffree = obd_osfs.os_ffree;
1845         }
1846
1847 out:
1848         RETURN(rc);
1849 }
1850 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
1851 {
1852         struct super_block *sb = de->d_sb;
1853         struct obd_statfs osfs;
1854         __u64 fsid = huge_encode_dev(sb->s_dev);
1855         int rc;
1856
1857         CDEBUG(D_VFSTRACE, "VFS Op: at %llu jiffies\n", get_jiffies_64());
1858         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STAFS, 1);
1859
1860         /* Some amount of caching on the client is allowed */
1861         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
1862         if (rc)
1863                 return rc;
1864
1865         statfs_unpack(sfs, &osfs);
1866
1867         /* We need to downshift for all 32-bit kernels, because we can't
1868          * tell if the kernel is being called via sys_statfs64() or not.
1869          * Stop before overflowing f_bsize - in which case it is better
1870          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
1871         if (sizeof(long) < 8) {
1872                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
1873                         sfs->f_bsize <<= 1;
1874
1875                         osfs.os_blocks >>= 1;
1876                         osfs.os_bfree >>= 1;
1877                         osfs.os_bavail >>= 1;
1878                 }
1879         }
1880
1881         sfs->f_blocks = osfs.os_blocks;
1882         sfs->f_bfree = osfs.os_bfree;
1883         sfs->f_bavail = osfs.os_bavail;
1884         sfs->f_fsid.val[0] = (__u32)fsid;
1885         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
1886         return 0;
1887 }
1888
1889 void ll_inode_size_lock(struct inode *inode)
1890 {
1891         struct ll_inode_info *lli;
1892
1893         LASSERT(!S_ISDIR(inode->i_mode));
1894
1895         lli = ll_i2info(inode);
1896         mutex_lock(&lli->lli_size_mutex);
1897 }
1898
1899 void ll_inode_size_unlock(struct inode *inode)
1900 {
1901         struct ll_inode_info *lli;
1902
1903         lli = ll_i2info(inode);
1904         mutex_unlock(&lli->lli_size_mutex);
1905 }
1906
1907 void ll_update_inode_flags(struct inode *inode, int ext_flags)
1908 {
1909         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
1910         if (ext_flags & LUSTRE_PROJINHERIT_FL)
1911                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1912         else
1913                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
1914 }
1915
1916 int ll_update_inode(struct inode *inode, struct lustre_md *md)
1917 {
1918         struct ll_inode_info *lli = ll_i2info(inode);
1919         struct mdt_body *body = md->body;
1920         struct ll_sb_info *sbi = ll_i2sbi(inode);
1921         int rc = 0;
1922
1923         if (body->mbo_valid & OBD_MD_FLEASIZE) {
1924                 rc = cl_file_inode_init(inode, md);
1925                 if (rc)
1926                         return rc;
1927         }
1928
1929         if (S_ISDIR(inode->i_mode)) {
1930                 rc = ll_update_lsm_md(inode, md);
1931                 if (rc != 0)
1932                         return rc;
1933         }
1934
1935 #ifdef CONFIG_FS_POSIX_ACL
1936         if (body->mbo_valid & OBD_MD_FLACL) {
1937                 spin_lock(&lli->lli_lock);
1938                 if (lli->lli_posix_acl)
1939                         posix_acl_release(lli->lli_posix_acl);
1940                 lli->lli_posix_acl = md->posix_acl;
1941                 spin_unlock(&lli->lli_lock);
1942         }
1943 #endif
1944         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
1945                                         sbi->ll_flags & LL_SBI_32BIT_API);
1946         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
1947
1948         if (body->mbo_valid & OBD_MD_FLATIME) {
1949                 if (body->mbo_atime > inode->i_atime.tv_sec)
1950                         inode->i_atime.tv_sec = body->mbo_atime;
1951                 lli->lli_atime = body->mbo_atime;
1952         }
1953
1954         if (body->mbo_valid & OBD_MD_FLMTIME) {
1955                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
1956                         CDEBUG(D_INODE,
1957                                "setting ino %lu mtime from %lld to %llu\n",
1958                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
1959                                body->mbo_mtime);
1960                         inode->i_mtime.tv_sec = body->mbo_mtime;
1961                 }
1962                 lli->lli_mtime = body->mbo_mtime;
1963         }
1964
1965         if (body->mbo_valid & OBD_MD_FLCTIME) {
1966                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
1967                         inode->i_ctime.tv_sec = body->mbo_ctime;
1968                 lli->lli_ctime = body->mbo_ctime;
1969         }
1970
1971         /* Clear i_flags to remove S_NOSEC before permissions are updated */
1972         if (body->mbo_valid & OBD_MD_FLFLAGS)
1973                 ll_update_inode_flags(inode, body->mbo_flags);
1974         if (body->mbo_valid & OBD_MD_FLMODE)
1975                 inode->i_mode = (inode->i_mode & S_IFMT) |
1976                                 (body->mbo_mode & ~S_IFMT);
1977
1978         if (body->mbo_valid & OBD_MD_FLTYPE)
1979                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1980                                 (body->mbo_mode & S_IFMT);
1981
1982         LASSERT(inode->i_mode != 0);
1983         if (S_ISREG(inode->i_mode))
1984                 inode->i_blkbits = min(PTLRPC_MAX_BRW_BITS + 1,
1985                                        LL_MAX_BLKSIZE_BITS);
1986         else
1987                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
1988
1989         if (body->mbo_valid & OBD_MD_FLUID)
1990                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
1991         if (body->mbo_valid & OBD_MD_FLGID)
1992                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
1993         if (body->mbo_valid & OBD_MD_FLPROJID)
1994                 lli->lli_projid = body->mbo_projid;
1995         if (body->mbo_valid & OBD_MD_FLNLINK)
1996                 set_nlink(inode, body->mbo_nlink);
1997         if (body->mbo_valid & OBD_MD_FLRDEV)
1998                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
1999
2000         if (body->mbo_valid & OBD_MD_FLID) {
2001                 /* FID shouldn't be changed! */
2002                 if (fid_is_sane(&lli->lli_fid)) {
2003                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2004                                  "Trying to change FID "DFID
2005                                  " to the "DFID", inode "DFID"(%p)\n",
2006                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2007                                  PFID(ll_inode2fid(inode)), inode);
2008                 } else {
2009                         lli->lli_fid = body->mbo_fid1;
2010                 }
2011         }
2012
2013         LASSERT(fid_seq(&lli->lli_fid) != 0);
2014
2015         if (body->mbo_valid & OBD_MD_FLSIZE) {
2016                 i_size_write(inode, body->mbo_size);
2017
2018                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2019                        PFID(ll_inode2fid(inode)),
2020                        (unsigned long long)body->mbo_size);
2021
2022                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2023                         inode->i_blocks = body->mbo_blocks;
2024         }
2025
2026         if (body->mbo_valid & OBD_MD_TSTATE) {
2027                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2028                  * clear it when done to ensure to start again
2029                  * glimpsing updated attrs
2030                  */
2031                 if (body->mbo_t_state & MS_RESTORE)
2032                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2033                 else
2034                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2035         }
2036
2037         return 0;
2038 }
2039
2040 int ll_read_inode2(struct inode *inode, void *opaque)
2041 {
2042         struct lustre_md *md = opaque;
2043         struct ll_inode_info *lli = ll_i2info(inode);
2044         int     rc;
2045         ENTRY;
2046
2047         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2048                PFID(&lli->lli_fid), inode);
2049
2050         /* Core attributes from the MDS first.  This is a new inode, and
2051          * the VFS doesn't zero times in the core inode so we have to do
2052          * it ourselves.  They will be overwritten by either MDS or OST
2053          * attributes - we just need to make sure they aren't newer.
2054          */
2055         inode->i_mtime.tv_sec = 0;
2056         inode->i_atime.tv_sec = 0;
2057         inode->i_ctime.tv_sec = 0;
2058         inode->i_rdev = 0;
2059         rc = ll_update_inode(inode, md);
2060         if (rc != 0)
2061                 RETURN(rc);
2062
2063         /* OIDEBUG(inode); */
2064
2065 #ifdef HAVE_BACKING_DEV_INFO
2066         /* initializing backing dev info. */
2067         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2068 #endif
2069         if (S_ISREG(inode->i_mode)) {
2070                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2071                 inode->i_op = &ll_file_inode_operations;
2072                 inode->i_fop = sbi->ll_fop;
2073                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2074                 EXIT;
2075         } else if (S_ISDIR(inode->i_mode)) {
2076                 inode->i_op = &ll_dir_inode_operations;
2077                 inode->i_fop = &ll_dir_operations;
2078                 EXIT;
2079         } else if (S_ISLNK(inode->i_mode)) {
2080                 inode->i_op = &ll_fast_symlink_inode_operations;
2081                 EXIT;
2082         } else {
2083                 inode->i_op = &ll_special_inode_operations;
2084
2085                 init_special_inode(inode, inode->i_mode,
2086                                    inode->i_rdev);
2087
2088                 EXIT;
2089         }
2090
2091         return 0;
2092 }
2093
2094 void ll_delete_inode(struct inode *inode)
2095 {
2096         struct ll_inode_info *lli = ll_i2info(inode);
2097         struct address_space *mapping = &inode->i_data;
2098         unsigned long nrpages;
2099         ENTRY;
2100
2101         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL)
2102                 /* It is last chance to write out dirty pages,
2103                  * otherwise we may lose data while umount */
2104                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, CL_FSYNC_LOCAL, 1);
2105
2106         truncate_inode_pages_final(mapping);
2107
2108         /* Workaround for LU-118: Note nrpages may not be totally updated when
2109          * truncate_inode_pages() returns, as there can be a page in the process
2110          * of deletion (inside __delete_from_page_cache()) in the specified
2111          * range. Thus mapping->nrpages can be non-zero when this function
2112          * returns even after truncation of the whole mapping.  Only do this if
2113          * npages isn't already zero.
2114          */
2115         nrpages = mapping->nrpages;
2116         if (nrpages) {
2117                 spin_lock_irq(&mapping->tree_lock);
2118                 nrpages = mapping->nrpages;
2119                 spin_unlock_irq(&mapping->tree_lock);
2120         } /* Workaround end */
2121
2122         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2123                  "see https://jira.whamcloud.com/browse/LU-118\n",
2124                  ll_get_fsname(inode->i_sb, NULL, 0),
2125                  PFID(ll_inode2fid(inode)), inode, nrpages);
2126
2127 #ifdef HAVE_SBOPS_EVICT_INODE
2128         ll_clear_inode(inode);
2129 #endif
2130         clear_inode(inode);
2131
2132         EXIT;
2133 }
2134
2135 int ll_iocontrol(struct inode *inode, struct file *file,
2136                  unsigned int cmd, unsigned long arg)
2137 {
2138         struct ll_sb_info *sbi = ll_i2sbi(inode);
2139         struct ptlrpc_request *req = NULL;
2140         int rc, flags = 0;
2141         ENTRY;
2142
2143         switch (cmd) {
2144         case FS_IOC_GETFLAGS: {
2145                 struct mdt_body *body;
2146                 struct md_op_data *op_data;
2147
2148                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2149                                              0, 0, LUSTRE_OPC_ANY,
2150                                              NULL);
2151                 if (IS_ERR(op_data))
2152                         RETURN(PTR_ERR(op_data));
2153
2154                 op_data->op_valid = OBD_MD_FLFLAGS;
2155                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2156                 ll_finish_md_op_data(op_data);
2157                 if (rc) {
2158                         CERROR("%s: failure inode "DFID": rc = %d\n",
2159                                sbi->ll_md_exp->exp_obd->obd_name,
2160                                PFID(ll_inode2fid(inode)), rc);
2161                         RETURN(-abs(rc));
2162                 }
2163
2164                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2165
2166                 flags = body->mbo_flags;
2167
2168                 ptlrpc_req_finished(req);
2169
2170                 RETURN(put_user(flags, (int __user *)arg));
2171         }
2172         case FS_IOC_SETFLAGS: {
2173                 struct iattr *attr;
2174                 struct md_op_data *op_data;
2175                 struct cl_object *obj;
2176                 struct fsxattr fa = { 0 };
2177
2178                 if (get_user(flags, (int __user *)arg))
2179                         RETURN(-EFAULT);
2180
2181                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2182                 if (flags & LUSTRE_PROJINHERIT_FL)
2183                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2184
2185                 rc = ll_ioctl_check_project(inode, &fa);
2186                 if (rc)
2187                         RETURN(rc);
2188
2189                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2190                                              LUSTRE_OPC_ANY, NULL);
2191                 if (IS_ERR(op_data))
2192                         RETURN(PTR_ERR(op_data));
2193
2194                 op_data->op_attr_flags = flags;
2195                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2196                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2197                 ll_finish_md_op_data(op_data);
2198                 ptlrpc_req_finished(req);
2199                 if (rc)
2200                         RETURN(rc);
2201
2202                 ll_update_inode_flags(inode, flags);
2203
2204                 obj = ll_i2info(inode)->lli_clob;
2205                 if (obj == NULL)
2206                         RETURN(0);
2207
2208                 OBD_ALLOC_PTR(attr);
2209                 if (attr == NULL)
2210                         RETURN(-ENOMEM);
2211
2212                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2213
2214                 OBD_FREE_PTR(attr);
2215                 RETURN(rc);
2216         }
2217         default:
2218                 RETURN(-ENOSYS);
2219         }
2220
2221         RETURN(0);
2222 }
2223
2224 int ll_flush_ctx(struct inode *inode)
2225 {
2226         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2227
2228         CDEBUG(D_SEC, "flush context for user %d\n",
2229                from_kuid(&init_user_ns, current_uid()));
2230
2231         obd_set_info_async(NULL, sbi->ll_md_exp,
2232                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2233                            0, NULL, NULL);
2234         obd_set_info_async(NULL, sbi->ll_dt_exp,
2235                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2236                            0, NULL, NULL);
2237         return 0;
2238 }
2239
2240 /* umount -f client means force down, don't save state */
2241 void ll_umount_begin(struct super_block *sb)
2242 {
2243         struct ll_sb_info *sbi = ll_s2sbi(sb);
2244         struct obd_device *obd;
2245         struct obd_ioctl_data *ioc_data;
2246         struct l_wait_info lwi;
2247         wait_queue_head_t waitq;
2248         ENTRY;
2249
2250         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2251                sb->s_count, atomic_read(&sb->s_active));
2252
2253         obd = class_exp2obd(sbi->ll_md_exp);
2254         if (obd == NULL) {
2255                 CERROR("Invalid MDC connection handle %#llx\n",
2256                        sbi->ll_md_exp->exp_handle.h_cookie);
2257                 EXIT;
2258                 return;
2259         }
2260         obd->obd_force = 1;
2261
2262         obd = class_exp2obd(sbi->ll_dt_exp);
2263         if (obd == NULL) {
2264                 CERROR("Invalid LOV connection handle %#llx\n",
2265                        sbi->ll_dt_exp->exp_handle.h_cookie);
2266                 EXIT;
2267                 return;
2268         }
2269         obd->obd_force = 1;
2270
2271         OBD_ALLOC_PTR(ioc_data);
2272         if (ioc_data) {
2273                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2274                               sizeof *ioc_data, ioc_data, NULL);
2275
2276                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2277                               sizeof *ioc_data, ioc_data, NULL);
2278
2279                 OBD_FREE_PTR(ioc_data);
2280         }
2281
2282         /* Really, we'd like to wait until there are no requests outstanding,
2283          * and then continue.  For now, we just periodically checking for vfs
2284          * to decrement mnt_cnt and hope to finish it within 10sec.
2285          */
2286         init_waitqueue_head(&waitq);
2287         lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(10),
2288                                    cfs_time_seconds(1), NULL, NULL);
2289         l_wait_event(waitq, may_umount(sbi->ll_mnt.mnt), &lwi);
2290
2291         EXIT;
2292 }
2293
2294 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2295 {
2296         struct ll_sb_info *sbi = ll_s2sbi(sb);
2297         char *profilenm = get_profile_name(sb);
2298         int err;
2299         __u32 read_only;
2300
2301         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
2302                 read_only = *flags & MS_RDONLY;
2303                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2304                                          sizeof(KEY_READ_ONLY),
2305                                          KEY_READ_ONLY, sizeof(read_only),
2306                                          &read_only, NULL);
2307                 if (err) {
2308                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2309                                       profilenm, read_only ?
2310                                       "read-only" : "read-write", err);
2311                         return err;
2312                 }
2313
2314                 if (read_only)
2315                         sb->s_flags |= MS_RDONLY;
2316                 else
2317                         sb->s_flags &= ~MS_RDONLY;
2318
2319                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2320                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2321                                       read_only ?  "read-only" : "read-write");
2322         }
2323         return 0;
2324 }
2325
2326 /**
2327  * Cleanup the open handle that is cached on MDT-side.
2328  *
2329  * For open case, the client side open handling thread may hit error
2330  * after the MDT grant the open. Under such case, the client should
2331  * send close RPC to the MDT as cleanup; otherwise, the open handle
2332  * on the MDT will be leaked there until the client umount or evicted.
2333  *
2334  * In further, if someone unlinked the file, because the open handle
2335  * holds the reference on such file/object, then it will block the
2336  * subsequent threads that want to locate such object via FID.
2337  *
2338  * \param[in] sb        super block for this file-system
2339  * \param[in] open_req  pointer to the original open request
2340  */
2341 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2342 {
2343         struct mdt_body                 *body;
2344         struct md_op_data               *op_data;
2345         struct ptlrpc_request           *close_req = NULL;
2346         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2347         ENTRY;
2348
2349         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2350         OBD_ALLOC_PTR(op_data);
2351         if (op_data == NULL) {
2352                 CWARN("%s: cannot allocate op_data to release open handle for "
2353                       DFID"\n",
2354                       ll_get_fsname(sb, NULL, 0), PFID(&body->mbo_fid1));
2355
2356                 RETURN_EXIT;
2357         }
2358
2359         op_data->op_fid1 = body->mbo_fid1;
2360         op_data->op_open_handle = body->mbo_open_handle;
2361         op_data->op_mod_time = ktime_get_real_seconds();
2362         md_close(exp, op_data, NULL, &close_req);
2363         ptlrpc_req_finished(close_req);
2364         ll_finish_md_op_data(op_data);
2365
2366         EXIT;
2367 }
2368
2369 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2370                   struct super_block *sb, struct lookup_intent *it)
2371 {
2372         struct ll_sb_info *sbi = NULL;
2373         struct lustre_md md = { NULL };
2374         int rc;
2375         ENTRY;
2376
2377         LASSERT(*inode || sb);
2378         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2379         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2380                               sbi->ll_md_exp, &md);
2381         if (rc != 0)
2382                 GOTO(cleanup, rc);
2383
2384         if (*inode) {
2385                 rc = ll_update_inode(*inode, &md);
2386                 if (rc != 0)
2387                         GOTO(out, rc);
2388         } else {
2389                 LASSERT(sb != NULL);
2390
2391                 /*
2392                  * At this point server returns to client's same fid as client
2393                  * generated for creating. So using ->fid1 is okay here.
2394                  */
2395                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2396                         CERROR("%s: Fid is insane "DFID"\n",
2397                                 ll_get_fsname(sb, NULL, 0),
2398                                 PFID(&md.body->mbo_fid1));
2399                         GOTO(out, rc = -EINVAL);
2400                 }
2401
2402                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2403                                              sbi->ll_flags & LL_SBI_32BIT_API),
2404                                  &md);
2405                 if (IS_ERR(*inode)) {
2406 #ifdef CONFIG_FS_POSIX_ACL
2407                         if (md.posix_acl) {
2408                                 posix_acl_release(md.posix_acl);
2409                                 md.posix_acl = NULL;
2410                         }
2411 #endif
2412                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2413                         *inode = NULL;
2414                         CERROR("new_inode -fatal: rc %d\n", rc);
2415                         GOTO(out, rc);
2416                 }
2417         }
2418
2419         /* Handling piggyback layout lock.
2420          * Layout lock can be piggybacked by getattr and open request.
2421          * The lsm can be applied to inode only if it comes with a layout lock
2422          * otherwise correct layout may be overwritten, for example:
2423          * 1. proc1: mdt returns a lsm but not granting layout
2424          * 2. layout was changed by another client
2425          * 3. proc2: refresh layout and layout lock granted
2426          * 4. proc1: to apply a stale layout */
2427         if (it != NULL && it->it_lock_mode != 0) {
2428                 struct lustre_handle lockh;
2429                 struct ldlm_lock *lock;
2430
2431                 lockh.cookie = it->it_lock_handle;
2432                 lock = ldlm_handle2lock(&lockh);
2433                 LASSERT(lock != NULL);
2434                 if (ldlm_has_layout(lock)) {
2435                         struct cl_object_conf conf;
2436
2437                         memset(&conf, 0, sizeof(conf));
2438                         conf.coc_opc = OBJECT_CONF_SET;
2439                         conf.coc_inode = *inode;
2440                         conf.coc_lock = lock;
2441                         conf.u.coc_layout = md.layout;
2442                         (void)ll_layout_conf(*inode, &conf);
2443                 }
2444                 LDLM_LOCK_PUT(lock);
2445         }
2446
2447         GOTO(out, rc = 0);
2448
2449 out:
2450         md_free_lustre_md(sbi->ll_md_exp, &md);
2451
2452 cleanup:
2453         if (rc != 0 && it != NULL && it->it_op & IT_OPEN)
2454                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2455
2456         return rc;
2457 }
2458
2459 int ll_obd_statfs(struct inode *inode, void __user *arg)
2460 {
2461         struct ll_sb_info *sbi = NULL;
2462         struct obd_export *exp;
2463         char *buf = NULL;
2464         struct obd_ioctl_data *data = NULL;
2465         __u32 type;
2466         int len = 0, rc;
2467
2468         if (!inode || !(sbi = ll_i2sbi(inode)))
2469                 GOTO(out_statfs, rc = -EINVAL);
2470
2471         rc = obd_ioctl_getdata(&buf, &len, arg);
2472         if (rc)
2473                 GOTO(out_statfs, rc);
2474
2475         data = (void*)buf;
2476         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
2477             !data->ioc_pbuf1 || !data->ioc_pbuf2)
2478                 GOTO(out_statfs, rc = -EINVAL);
2479
2480         if (data->ioc_inllen1 != sizeof(__u32) ||
2481             data->ioc_inllen2 != sizeof(__u32) ||
2482             data->ioc_plen1 != sizeof(struct obd_statfs) ||
2483             data->ioc_plen2 != sizeof(struct obd_uuid))
2484                 GOTO(out_statfs, rc = -EINVAL);
2485
2486         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
2487         if (type & LL_STATFS_LMV)
2488                 exp = sbi->ll_md_exp;
2489         else if (type & LL_STATFS_LOV)
2490                 exp = sbi->ll_dt_exp;
2491         else
2492                 GOTO(out_statfs, rc = -ENODEV);
2493
2494         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
2495         if (rc)
2496                 GOTO(out_statfs, rc);
2497 out_statfs:
2498         OBD_FREE_LARGE(buf, len);
2499         return rc;
2500 }
2501
2502 /*
2503  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
2504  * be called early to avoid deadlock.
2505  */
2506 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
2507 {
2508         if (op_data->op_mea2_sem) {
2509                 up_read(op_data->op_mea2_sem);
2510                 op_data->op_mea2_sem = NULL;
2511         }
2512
2513         if (op_data->op_mea1_sem) {
2514                 up_read(op_data->op_mea1_sem);
2515                 op_data->op_mea1_sem = NULL;
2516         }
2517 }
2518
2519 /* this function prepares md_op_data hint for passing it down to MD stack. */
2520 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
2521                                       struct inode *i1, struct inode *i2,
2522                                       const char *name, size_t namelen,
2523                                       __u32 mode, __u32 opc, void *data)
2524 {
2525         LASSERT(i1 != NULL);
2526
2527         if (name == NULL) {
2528                 /* Do not reuse namelen for something else. */
2529                 if (namelen != 0)
2530                         return ERR_PTR(-EINVAL);
2531         } else {
2532                 if (namelen > ll_i2sbi(i1)->ll_namelen)
2533                         return ERR_PTR(-ENAMETOOLONG);
2534
2535                 if (!lu_name_is_valid_2(name, namelen))
2536                         return ERR_PTR(-EINVAL);
2537         }
2538
2539         if (op_data == NULL)
2540                 OBD_ALLOC_PTR(op_data);
2541
2542         if (op_data == NULL)
2543                 return ERR_PTR(-ENOMEM);
2544
2545         ll_i2gids(op_data->op_suppgids, i1, i2);
2546         op_data->op_fid1 = *ll_inode2fid(i1);
2547         op_data->op_default_stripe_offset = -1;
2548
2549         if (S_ISDIR(i1->i_mode)) {
2550                 down_read(&ll_i2info(i1)->lli_lsm_sem);
2551                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
2552                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
2553                 if (opc == LUSTRE_OPC_MKDIR)
2554                         op_data->op_default_stripe_offset =
2555                                    ll_i2info(i1)->lli_def_stripe_offset;
2556         }
2557
2558         if (i2) {
2559                 op_data->op_fid2 = *ll_inode2fid(i2);
2560                 if (S_ISDIR(i2->i_mode)) {
2561                         if (i2 != i1) {
2562                                 down_read(&ll_i2info(i2)->lli_lsm_sem);
2563                                 op_data->op_mea2_sem =
2564                                                 &ll_i2info(i2)->lli_lsm_sem;
2565                         }
2566                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
2567                 }
2568         } else {
2569                 fid_zero(&op_data->op_fid2);
2570         }
2571
2572         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
2573                 op_data->op_cli_flags |= CLI_HASH64;
2574
2575         if (ll_need_32bit_api(ll_i2sbi(i1)))
2576                 op_data->op_cli_flags |= CLI_API32;
2577
2578         op_data->op_name = name;
2579         op_data->op_namelen = namelen;
2580         op_data->op_mode = mode;
2581         op_data->op_mod_time = ktime_get_real_seconds();
2582         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
2583         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
2584         op_data->op_cap = cfs_curproc_cap_pack();
2585         op_data->op_mds = 0;
2586         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
2587              filename_is_volatile(name, namelen, &op_data->op_mds)) {
2588                 op_data->op_bias |= MDS_CREATE_VOLATILE;
2589         }
2590         op_data->op_data = data;
2591
2592         return op_data;
2593 }
2594
2595 void ll_finish_md_op_data(struct md_op_data *op_data)
2596 {
2597         ll_unlock_md_op_lsm(op_data);
2598         security_release_secctx(op_data->op_file_secctx,
2599                                 op_data->op_file_secctx_size);
2600         OBD_FREE_PTR(op_data);
2601 }
2602
2603 #ifdef HAVE_SUPEROPS_USE_DENTRY
2604 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
2605 #else
2606 int ll_show_options(struct seq_file *seq, struct vfsmount *vfs)
2607 #endif
2608 {
2609         struct ll_sb_info *sbi;
2610
2611 #ifdef HAVE_SUPEROPS_USE_DENTRY
2612         LASSERT((seq != NULL) && (dentry != NULL));
2613         sbi = ll_s2sbi(dentry->d_sb);
2614 #else
2615         LASSERT((seq != NULL) && (vfs != NULL));
2616         sbi = ll_s2sbi(vfs->mnt_sb);
2617 #endif
2618
2619         if (sbi->ll_flags & LL_SBI_NOLCK)
2620                 seq_puts(seq, ",nolock");
2621
2622         if (sbi->ll_flags & LL_SBI_FLOCK)
2623                 seq_puts(seq, ",flock");
2624
2625         if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
2626                 seq_puts(seq, ",localflock");
2627
2628         if (sbi->ll_flags & LL_SBI_USER_XATTR)
2629                 seq_puts(seq, ",user_xattr");
2630
2631         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2632                 seq_puts(seq, ",lazystatfs");
2633
2634         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
2635                 seq_puts(seq, ",user_fid2path");
2636
2637         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
2638                 seq_puts(seq, ",always_ping");
2639
2640         RETURN(0);
2641 }
2642
2643 /**
2644  * Get obd name by cmd, and copy out to user space
2645  */
2646 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
2647 {
2648         struct ll_sb_info *sbi = ll_i2sbi(inode);
2649         struct obd_device *obd;
2650         ENTRY;
2651
2652         if (cmd == OBD_IOC_GETDTNAME)
2653                 obd = class_exp2obd(sbi->ll_dt_exp);
2654         else if (cmd == OBD_IOC_GETMDNAME)
2655                 obd = class_exp2obd(sbi->ll_md_exp);
2656         else
2657                 RETURN(-EINVAL);
2658
2659         if (!obd)
2660                 RETURN(-ENOENT);
2661
2662         if (copy_to_user((void __user *)arg, obd->obd_name,
2663                          strlen(obd->obd_name) + 1))
2664                 RETURN(-EFAULT);
2665
2666         RETURN(0);
2667 }
2668
2669 /**
2670  * Get lustre file system name by \a sbi. If \a buf is provided(non-NULL), the
2671  * fsname will be returned in this buffer; otherwise, a static buffer will be
2672  * used to store the fsname and returned to caller.
2673  */
2674 char *ll_get_fsname(struct super_block *sb, char *buf, int buflen)
2675 {
2676         static char fsname_static[MTI_NAME_MAXLEN];
2677         struct lustre_sb_info *lsi = s2lsi(sb);
2678         char *ptr;
2679         int len;
2680
2681         if (buf == NULL) {
2682                 /* this means the caller wants to use static buffer
2683                  * and it doesn't care about race. Usually this is
2684                  * in error reporting path */
2685                 buf = fsname_static;
2686                 buflen = sizeof(fsname_static);
2687         }
2688
2689         len = strlen(lsi->lsi_lmd->lmd_profile);
2690         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
2691         if (ptr && (strcmp(ptr, "-client") == 0))
2692                 len -= 7;
2693
2694         if (unlikely(len >= buflen))
2695                 len = buflen - 1;
2696         strncpy(buf, lsi->lsi_lmd->lmd_profile, len);
2697         buf[len] = '\0';
2698
2699         return buf;
2700 }
2701
2702 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
2703 {
2704         char *path = NULL;
2705
2706         struct path p;
2707
2708         p.dentry = dentry;
2709         p.mnt = current->fs->root.mnt;
2710         path_get(&p);
2711         path = d_path(&p, buf, bufsize);
2712         path_put(&p);
2713         return path;
2714 }
2715
2716 void ll_dirty_page_discard_warn(struct page *page, int ioret)
2717 {
2718         char *buf, *path = NULL;
2719         struct dentry *dentry = NULL;
2720         struct inode *inode = page->mapping->host;
2721
2722         /* this can be called inside spin lock so use GFP_ATOMIC. */
2723         buf = (char *)__get_free_page(GFP_ATOMIC);
2724         if (buf != NULL) {
2725                 dentry = d_find_alias(page->mapping->host);
2726                 if (dentry != NULL)
2727                         path = ll_d_path(dentry, buf, PAGE_SIZE);
2728         }
2729
2730         CDEBUG(D_WARNING,
2731                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
2732                "(rc %d)\n", ll_get_fsname(page->mapping->host->i_sb, NULL, 0),
2733                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
2734                PFID(ll_inode2fid(inode)),
2735                (path && !IS_ERR(path)) ? path : "", ioret);
2736
2737         if (dentry != NULL)
2738                 dput(dentry);
2739
2740         if (buf != NULL)
2741                 free_page((unsigned long)buf);
2742 }
2743
2744 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
2745                         struct lov_user_md **kbuf)
2746 {
2747         struct lov_user_md      lum;
2748         ssize_t                 lum_size;
2749         ENTRY;
2750
2751         if (copy_from_user(&lum, md, sizeof(lum)))
2752                 RETURN(-EFAULT);
2753
2754         lum_size = ll_lov_user_md_size(&lum);
2755         if (lum_size < 0)
2756                 RETURN(lum_size);
2757
2758         OBD_ALLOC(*kbuf, lum_size);
2759         if (*kbuf == NULL)
2760                 RETURN(-ENOMEM);
2761
2762         if (copy_from_user(*kbuf, md, lum_size) != 0) {
2763                 OBD_FREE(*kbuf, lum_size);
2764                 RETURN(-EFAULT);
2765         }
2766
2767         RETURN(lum_size);
2768 }
2769
2770 /*
2771  * Compute llite root squash state after a change of root squash
2772  * configuration setting or add/remove of a lnet nid
2773  */
2774 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
2775 {
2776         struct root_squash_info *squash = &sbi->ll_squash;
2777         int i;
2778         bool matched;
2779         struct lnet_process_id id;
2780
2781         /* Update norootsquash flag */
2782         down_write(&squash->rsi_sem);
2783         if (list_empty(&squash->rsi_nosquash_nids))
2784                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2785         else {
2786                 /* Do not apply root squash as soon as one of our NIDs is
2787                  * in the nosquash_nids list */
2788                 matched = false;
2789                 i = 0;
2790                 while (LNetGetId(i++, &id) != -ENOENT) {
2791                         if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
2792                                 continue;
2793                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
2794                                 matched = true;
2795                                 break;
2796                         }
2797                 }
2798                 if (matched)
2799                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
2800                 else
2801                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
2802         }
2803         up_write(&squash->rsi_sem);
2804 }
2805
2806 /**
2807  * Parse linkea content to extract information about a given hardlink
2808  *
2809  * \param[in]   ldata      - Initialized linkea data
2810  * \param[in]   linkno     - Link identifier
2811  * \param[out]  parent_fid - The entry's parent FID
2812  * \param[out]  ln         - Entry name destination buffer
2813  *
2814  * \retval 0 on success
2815  * \retval Appropriate negative error code on failure
2816  */
2817 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
2818                             struct lu_fid *parent_fid, struct lu_name *ln)
2819 {
2820         unsigned int    idx;
2821         int             rc;
2822         ENTRY;
2823
2824         rc = linkea_init_with_rec(ldata);
2825         if (rc < 0)
2826                 RETURN(rc);
2827
2828         if (linkno >= ldata->ld_leh->leh_reccount)
2829                 /* beyond last link */
2830                 RETURN(-ENODATA);
2831
2832         linkea_first_entry(ldata);
2833         for (idx = 0; ldata->ld_lee != NULL; idx++) {
2834                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
2835                                     parent_fid);
2836                 if (idx == linkno)
2837                         break;
2838
2839                 linkea_next_entry(ldata);
2840         }
2841
2842         if (idx < linkno)
2843                 RETURN(-ENODATA);
2844
2845         RETURN(0);
2846 }
2847
2848 /**
2849  * Get parent FID and name of an identified link. Operation is performed for
2850  * a given link number, letting the caller iterate over linkno to list one or
2851  * all links of an entry.
2852  *
2853  * \param[in]     file - File descriptor against which to perform the operation
2854  * \param[in,out] arg  - User-filled structure containing the linkno to operate
2855  *                       on and the available size. It is eventually filled with
2856  *                       the requested information or left untouched on error
2857  *
2858  * \retval - 0 on success
2859  * \retval - Appropriate negative error code on failure
2860  */
2861 int ll_getparent(struct file *file, struct getparent __user *arg)
2862 {
2863         struct inode            *inode = file_inode(file);
2864         struct linkea_data      *ldata;
2865         struct lu_buf            buf = LU_BUF_NULL;
2866         struct lu_name           ln;
2867         struct lu_fid            parent_fid;
2868         __u32                    linkno;
2869         __u32                    name_size;
2870         int                      rc;
2871
2872         ENTRY;
2873
2874         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
2875             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
2876                 RETURN(-EPERM);
2877
2878         if (get_user(name_size, &arg->gp_name_size))
2879                 RETURN(-EFAULT);
2880
2881         if (get_user(linkno, &arg->gp_linkno))
2882                 RETURN(-EFAULT);
2883
2884         if (name_size > PATH_MAX)
2885                 RETURN(-EINVAL);
2886
2887         OBD_ALLOC(ldata, sizeof(*ldata));
2888         if (ldata == NULL)
2889                 RETURN(-ENOMEM);
2890
2891         rc = linkea_data_new(ldata, &buf);
2892         if (rc < 0)
2893                 GOTO(ldata_free, rc);
2894
2895 #ifdef HAVE_XATTR_HANDLER_FLAGS
2896         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
2897                            buf.lb_len, OBD_MD_FLXATTR);
2898 #else
2899         rc = ll_getxattr(file_dentry(file), XATTR_NAME_LINK, buf.lb_buf,
2900                          buf.lb_len);
2901 #endif /* HAVE_XATTR_HANDLER_FLAGS */
2902         if (rc < 0)
2903                 GOTO(lb_free, rc);
2904
2905         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
2906         if (rc < 0)
2907                 GOTO(lb_free, rc);
2908
2909         if (ln.ln_namelen >= name_size)
2910                 GOTO(lb_free, rc = -EOVERFLOW);
2911
2912         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
2913                 GOTO(lb_free, rc = -EFAULT);
2914
2915         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
2916                 GOTO(lb_free, rc = -EFAULT);
2917
2918         if (put_user('\0', arg->gp_name + ln.ln_namelen))
2919                 GOTO(lb_free, rc = -EFAULT);
2920
2921 lb_free:
2922         lu_buf_free(&buf);
2923 ldata_free:
2924         OBD_FREE(ldata, sizeof(*ldata));
2925
2926         RETURN(rc);
2927 }