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/cpu.h>
  40 #include <linux/module.h>
  41 #include <linux/random.h>
  42 #include <linux/statfs.h>
  43 #include <linux/time.h>
  44 #include <linux/types.h>
  45 #include <libcfs/linux/linux-uuid.h>
  46 #include <linux/version.h>
  47 #include <linux/mm.h>
  48 #include <linux/user_namespace.h>
  49 #include <linux/delay.h>
  50 #include <linux/uidgid.h>
  51 #include <linux/security.h>
  52 #include <linux/fs_struct.h>
  53
  54 #ifndef HAVE_CPUS_READ_LOCK
  55 #include <libcfs/linux/linux-cpu.h>
  56 #endif
  57 #include <uapi/linux/lustre/lustre_ioctl.h>
  58 #ifdef HAVE_UAPI_LINUX_MOUNT_H
  59 #include <uapi/linux/mount.h>
  60 #endif
  61
  62 #include <lustre_ha.h>
  63 #include <lustre_dlm.h>
  64 #include <lprocfs_status.h>
  65 #include <lustre_disk.h>
  66 #include <uapi/linux/lustre/lustre_param.h>
  67 #include <lustre_log.h>
  68 #include <cl_object.h>
  69 #include <obd_cksum.h>
  70 #include "llite_internal.h"
  71
  72 struct kmem_cache *ll_file_data_slab;
  73
  74 #ifndef log2
  75 #define log2(n) ffz(~(n))
  76 #endif
  77
  78 /**
  79  * If there is only one number of core visible to Lustre,
  80  * async readahead will be disabled, to avoid massive over
  81  * subscription, we use 1/2 of active cores as default max
  82  * async readahead requests.
  83  */
  84 static inline unsigned int ll_get_ra_async_max_active(void)
  85 {
  86         return cfs_cpt_weight(cfs_cpt_tab, CFS_CPT_ANY) >> 1;
  87 }
  88
  89 static struct ll_sb_info *ll_init_sbi(void)
  90 {
  91         struct ll_sb_info *sbi = NULL;
  92         unsigned long pages;
  93         unsigned long lru_page_max;
  94         struct sysinfo si;
  95         int rc;
  96         int i;
  97
  98         ENTRY;
  99
 100         OBD_ALLOC_PTR(sbi);
 101         if (sbi == NULL)
 102                 RETURN(ERR_PTR(-ENOMEM));
 103
 104         rc = pcc_super_init(&sbi->ll_pcc_super);
 105         if (rc < 0)
 106                 GOTO(out_sbi, rc);
 107
 108         spin_lock_init(&sbi->ll_lock);
 109         mutex_init(&sbi->ll_lco.lco_lock);
 110         spin_lock_init(&sbi->ll_pp_extent_lock);
 111         spin_lock_init(&sbi->ll_process_lock);
 112         sbi->ll_rw_stats_on = 0;
 113         sbi->ll_statfs_max_age = OBD_STATFS_CACHE_SECONDS;
 114
 115         si_meminfo(&si);
 116         pages = si.totalram - si.totalhigh;
 117         lru_page_max = pages / 2;
 118
 119         sbi->ll_ra_info.ra_async_max_active = ll_get_ra_async_max_active();
 120         sbi->ll_ra_info.ll_readahead_wq =
 121                 cfs_cpt_bind_workqueue("ll-readahead-wq", cfs_cpt_tab,
 122                                        0, CFS_CPT_ANY,
 123                                        sbi->ll_ra_info.ra_async_max_active);
 124         if (IS_ERR(sbi->ll_ra_info.ll_readahead_wq))
 125                 GOTO(out_pcc, rc = PTR_ERR(sbi->ll_ra_info.ll_readahead_wq));
 126
 127         /* initialize ll_cache data */
 128         sbi->ll_cache = cl_cache_init(lru_page_max);
 129         if (sbi->ll_cache == NULL)
 130                 GOTO(out_destroy_ra, rc = -ENOMEM);
 131
 132         /* initialize foreign symlink prefix path */
 133         OBD_ALLOC(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
 134         if (sbi->ll_foreign_symlink_prefix == NULL)
 135                 GOTO(out_destroy_ra, rc = -ENOMEM);
 136         memcpy(sbi->ll_foreign_symlink_prefix, "/mnt/", sizeof("/mnt/"));
 137         sbi->ll_foreign_symlink_prefix_size = sizeof("/mnt/");
 138
 139         /* initialize foreign symlink upcall path, none by default */
 140         OBD_ALLOC(sbi->ll_foreign_symlink_upcall, sizeof("none"));
 141         if (sbi->ll_foreign_symlink_upcall == NULL)
 142                 GOTO(out_destroy_ra, rc = -ENOMEM);
 143         memcpy(sbi->ll_foreign_symlink_upcall, "none", sizeof("none"));
 144         sbi->ll_foreign_symlink_upcall_items = NULL;
 145         sbi->ll_foreign_symlink_upcall_nb_items = 0;
 146         init_rwsem(&sbi->ll_foreign_symlink_sem);
 147         /* foreign symlink support (LL_SBI_FOREIGN_SYMLINK in ll_flags)
 148          * not enabled by default
 149          */
 150
 151         sbi->ll_ra_info.ra_max_pages =
 152                 min(pages / 32, SBI_DEFAULT_READ_AHEAD_MAX);
 153         sbi->ll_ra_info.ra_max_pages_per_file =
 154                 min(sbi->ll_ra_info.ra_max_pages / 4,
 155                     SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX);
 156         sbi->ll_ra_info.ra_async_pages_per_file_threshold =
 157                                 sbi->ll_ra_info.ra_max_pages_per_file;
 158         sbi->ll_ra_info.ra_range_pages = SBI_DEFAULT_RA_RANGE_PAGES;
 159         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = -1;
 160         atomic_set(&sbi->ll_ra_info.ra_async_inflight, 0);
 161
 162         sbi->ll_flags |= LL_SBI_VERBOSE;
 163 #ifdef ENABLE_CHECKSUM
 164         sbi->ll_flags |= LL_SBI_CHECKSUM;
 165 #endif
 166 #ifdef ENABLE_FLOCK
 167         sbi->ll_flags |= LL_SBI_FLOCK;
 168 #endif
 169
 170 #ifdef HAVE_LRU_RESIZE_SUPPORT
 171         sbi->ll_flags |= LL_SBI_LRU_RESIZE;
 172 #endif
 173         sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 174
 175         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
 176                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 177                                pp_r_hist.oh_lock);
 178                 spin_lock_init(&sbi->ll_rw_extents_info.pp_extents[i].
 179                                pp_w_hist.oh_lock);
 180         }
 181
 182         /* metadata statahead is enabled by default */
 183         sbi->ll_sa_running_max = LL_SA_RUNNING_DEF;
 184         sbi->ll_sa_max = LL_SA_RPC_DEF;
 185         atomic_set(&sbi->ll_sa_total, 0);
 186         atomic_set(&sbi->ll_sa_wrong, 0);
 187         atomic_set(&sbi->ll_sa_running, 0);
 188         atomic_set(&sbi->ll_agl_total, 0);
 189         sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 190         sbi->ll_flags |= LL_SBI_FAST_READ;
 191         sbi->ll_flags |= LL_SBI_TINY_WRITE;
 192         ll_sbi_set_encrypt(sbi, true);
 193
 194         /* root squash */
 195         sbi->ll_squash.rsi_uid = 0;
 196         sbi->ll_squash.rsi_gid = 0;
 197         INIT_LIST_HEAD(&sbi->ll_squash.rsi_nosquash_nids);
 198         spin_lock_init(&sbi->ll_squash.rsi_lock);
 199
 200         /* Per-filesystem file heat */
 201         sbi->ll_heat_decay_weight = SBI_DEFAULT_HEAT_DECAY_WEIGHT;
 202         sbi->ll_heat_period_second = SBI_DEFAULT_HEAT_PERIOD_SECOND;
 203         RETURN(sbi);
 204 out_destroy_ra:
 205         if (sbi->ll_foreign_symlink_prefix)
 206                 OBD_FREE(sbi->ll_foreign_symlink_prefix, sizeof("/mnt/"));
 207         if (sbi->ll_cache) {
 208                 cl_cache_decref(sbi->ll_cache);
 209                 sbi->ll_cache = NULL;
 210         }
 211         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 212 out_pcc:
 213         pcc_super_fini(&sbi->ll_pcc_super);
 214 out_sbi:
 215         OBD_FREE_PTR(sbi);
 216         RETURN(ERR_PTR(rc));
 217 }
 218
 219 static void ll_free_sbi(struct super_block *sb)
 220 {
 221         struct ll_sb_info *sbi = ll_s2sbi(sb);
 222         ENTRY;
 223
 224         if (sbi != NULL) {
 225                 if (!list_empty(&sbi->ll_squash.rsi_nosquash_nids))
 226                         cfs_free_nidlist(&sbi->ll_squash.rsi_nosquash_nids);
 227                 if (sbi->ll_ra_info.ll_readahead_wq)
 228                         destroy_workqueue(sbi->ll_ra_info.ll_readahead_wq);
 229                 if (sbi->ll_cache != NULL) {
 230                         cl_cache_decref(sbi->ll_cache);
 231                         sbi->ll_cache = NULL;
 232                 }
 233                 if (sbi->ll_foreign_symlink_prefix) {
 234                         OBD_FREE(sbi->ll_foreign_symlink_prefix,
 235                                  sbi->ll_foreign_symlink_prefix_size);
 236                         sbi->ll_foreign_symlink_prefix = NULL;
 237                 }
 238                 if (sbi->ll_foreign_symlink_upcall) {
 239                         OBD_FREE(sbi->ll_foreign_symlink_upcall,
 240                                  strlen(sbi->ll_foreign_symlink_upcall) +
 241                                        1);
 242                         sbi->ll_foreign_symlink_upcall = NULL;
 243                 }
 244                 if (sbi->ll_foreign_symlink_upcall_items) {
 245                         int i;
 246                         int nb_items = sbi->ll_foreign_symlink_upcall_nb_items;
 247                         struct ll_foreign_symlink_upcall_item *items =
 248                                 sbi->ll_foreign_symlink_upcall_items;
 249
 250                         for (i = 0 ; i < nb_items; i++)
 251                                 if (items[i].type == STRING_TYPE)
 252                                         OBD_FREE(items[i].string,
 253                                                        items[i].size);
 254
 255                         OBD_FREE_LARGE(items, nb_items *
 256                                 sizeof(struct ll_foreign_symlink_upcall_item));
 257                         sbi->ll_foreign_symlink_upcall_items = NULL;
 258                 }
 259                 pcc_super_fini(&sbi->ll_pcc_super);
 260                 OBD_FREE(sbi, sizeof(*sbi));
 261         }
 262         EXIT;
 263 }
 264
 265 static int client_common_fill_super(struct super_block *sb, char *md, char *dt)
 266 {
 267         struct inode *root = NULL;
 268         struct ll_sb_info *sbi = ll_s2sbi(sb);
 269         struct obd_statfs *osfs = NULL;
 270         struct ptlrpc_request *request = NULL;
 271         struct obd_connect_data *data = NULL;
 272         struct obd_uuid *uuid;
 273         struct md_op_data *op_data;
 274         struct lustre_md lmd;
 275         u64 valid;
 276         int size, err, checksum;
 277
 278         ENTRY;
 279         sbi->ll_md_obd = class_name2obd(md);
 280         if (!sbi->ll_md_obd) {
 281                 CERROR("MD %s: not setup or attached\n", md);
 282                 RETURN(-EINVAL);
 283         }
 284
 285         OBD_ALLOC_PTR(data);
 286         if (data == NULL)
 287                 RETURN(-ENOMEM);
 288
 289         OBD_ALLOC_PTR(osfs);
 290         if (osfs == NULL) {
 291                 OBD_FREE_PTR(data);
 292                 RETURN(-ENOMEM);
 293         }
 294
 295         /* pass client page size via ocd_grant_blkbits, the server should report
 296          * back its backend blocksize for grant calculation purpose */
 297         data->ocd_grant_blkbits = PAGE_SHIFT;
 298
 299         /* indicate MDT features supported by this client */
 300         data->ocd_connect_flags = OBD_CONNECT_IBITS    | OBD_CONNECT_NODEVOH  |
 301                                   OBD_CONNECT_ATTRFID  | OBD_CONNECT_GRANT |
 302                                   OBD_CONNECT_VERSION  | OBD_CONNECT_BRW_SIZE |
 303                                   OBD_CONNECT_SRVLOCK  | OBD_CONNECT_TRUNCLOCK|
 304                                   OBD_CONNECT_MDS_CAPA | OBD_CONNECT_OSS_CAPA |
 305                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID     |
 306                                   OBD_CONNECT_AT       | OBD_CONNECT_LOV_V3   |
 307                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 308                                   OBD_CONNECT_64BITHASH |
 309                                   OBD_CONNECT_EINPROGRESS |
 310                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 311                                   OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS|
 312                                   OBD_CONNECT_MAX_EASIZE |
 313                                   OBD_CONNECT_FLOCK_DEAD |
 314                                   OBD_CONNECT_DISP_STRIPE | OBD_CONNECT_LFSCK |
 315                                   OBD_CONNECT_OPEN_BY_FID |
 316                                   OBD_CONNECT_DIR_STRIPE |
 317                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_CKSUM |
 318                                   OBD_CONNECT_SUBTREE |
 319                                   OBD_CONNECT_MULTIMODRPCS |
 320                                   OBD_CONNECT_GRANT_PARAM |
 321                                   OBD_CONNECT_SHORTIO | OBD_CONNECT_FLAGS2;
 322
 323         data->ocd_connect_flags2 = OBD_CONNECT2_DIR_MIGRATE |
 324                                    OBD_CONNECT2_SUM_STATFS |
 325                                    OBD_CONNECT2_OVERSTRIPING |
 326                                    OBD_CONNECT2_FLR |
 327                                    OBD_CONNECT2_LOCK_CONVERT |
 328                                    OBD_CONNECT2_ARCHIVE_ID_ARRAY |
 329                                    OBD_CONNECT2_INC_XID |
 330                                    OBD_CONNECT2_LSOM |
 331                                    OBD_CONNECT2_ASYNC_DISCARD |
 332                                    OBD_CONNECT2_PCC |
 333                                    OBD_CONNECT2_CRUSH | OBD_CONNECT2_LSEEK |
 334                                    OBD_CONNECT2_GETATTR_PFID |
 335                                    OBD_CONNECT2_DOM_LVB;
 336
 337 #ifdef HAVE_LRU_RESIZE_SUPPORT
 338         if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
 339                 data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 340 #endif
 341         data->ocd_connect_flags |= OBD_CONNECT_ACL_FLAGS;
 342
 343         data->ocd_cksum_types = obd_cksum_types_supported_client();
 344
 345         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_LIGHTWEIGHT))
 346                 /* flag mdc connection as lightweight, only used for test
 347                  * purpose, use with care */
 348                 data->ocd_connect_flags |= OBD_CONNECT_LIGHTWEIGHT;
 349
 350         data->ocd_ibits_known = MDS_INODELOCK_FULL;
 351         data->ocd_version = LUSTRE_VERSION_CODE;
 352
 353         if (sb->s_flags & SB_RDONLY)
 354                 data->ocd_connect_flags |= OBD_CONNECT_RDONLY;
 355         if (sbi->ll_flags & LL_SBI_USER_XATTR)
 356                 data->ocd_connect_flags |= OBD_CONNECT_XATTR;
 357
 358 #ifdef SB_NOSEC
 359         /* Setting this indicates we correctly support S_NOSEC (See kernel
 360          * commit 9e1f1de02c2275d7172e18dc4e7c2065777611bf)
 361          */
 362         sb->s_flags |= SB_NOSEC;
 363 #endif
 364
 365         if (sbi->ll_flags & LL_SBI_FLOCK)
 366                 sbi->ll_fop = &ll_file_operations_flock;
 367         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
 368                 sbi->ll_fop = &ll_file_operations;
 369         else
 370                 sbi->ll_fop = &ll_file_operations_noflock;
 371
 372         /* always ping even if server suppress_pings */
 373         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 374                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 375
 376         obd_connect_set_secctx(data);
 377         if (ll_sbi_has_encrypt(sbi))
 378                 obd_connect_set_enc(data);
 379
 380 #if defined(CONFIG_SECURITY)
 381         data->ocd_connect_flags2 |= OBD_CONNECT2_SELINUX_POLICY;
 382 #endif
 383
 384         data->ocd_brw_size = MD_MAX_BRW_SIZE;
 385
 386         err = obd_connect(NULL, &sbi->ll_md_exp, sbi->ll_md_obd,
 387                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 388         if (err == -EBUSY) {
 389                 LCONSOLE_ERROR_MSG(0x14f, "An MDT (md %s) is performing "
 390                                    "recovery, of which this client is not a "
 391                                    "part. Please wait for recovery to complete,"
 392                                    " abort, or time out.\n", md);
 393                 GOTO(out, err);
 394         } else if (err) {
 395                 CERROR("cannot connect to %s: rc = %d\n", md, err);
 396                 GOTO(out, err);
 397         }
 398
 399         sbi->ll_md_exp->exp_connect_data = *data;
 400
 401         err = obd_fid_init(sbi->ll_md_exp->exp_obd, sbi->ll_md_exp,
 402                            LUSTRE_SEQ_METADATA);
 403         if (err) {
 404                 CERROR("%s: Can't init metadata layer FID infrastructure, "
 405                        "rc = %d\n", sbi->ll_md_exp->exp_obd->obd_name, err);
 406                 GOTO(out_md, err);
 407         }
 408
 409         /* For mount, we only need fs info from MDT0, and also in DNE, it
 410          * can make sure the client can be mounted as long as MDT0 is
 411          * avaible */
 412         err = obd_statfs(NULL, sbi->ll_md_exp, osfs,
 413                         ktime_get_seconds() - sbi->ll_statfs_max_age,
 414                         OBD_STATFS_FOR_MDT0);
 415         if (err)
 416                 GOTO(out_md_fid, err);
 417
 418         /* This needs to be after statfs to ensure connect has finished.
 419          * Note that "data" does NOT contain the valid connect reply.
 420          * If connecting to a 1.8 server there will be no LMV device, so
 421          * we can access the MDC export directly and exp_connect_flags will
 422          * be non-zero, but if accessing an upgraded 2.1 server it will
 423          * have the correct flags filled in.
 424          * XXX: fill in the LMV exp_connect_flags from MDC(s). */
 425         valid = exp_connect_flags(sbi->ll_md_exp) & CLIENT_CONNECT_MDT_REQD;
 426         if (exp_connect_flags(sbi->ll_md_exp) != 0 &&
 427             valid != CLIENT_CONNECT_MDT_REQD) {
 428                 char *buf;
 429
 430                 OBD_ALLOC_WAIT(buf, PAGE_SIZE);
 431                 obd_connect_flags2str(buf, PAGE_SIZE,
 432                                       valid ^ CLIENT_CONNECT_MDT_REQD, 0, ",");
 433                 LCONSOLE_ERROR_MSG(0x170, "Server %s does not support "
 434                                    "feature(s) needed for correct operation "
 435                                    "of this client (%s). Please upgrade "
 436                                    "server or downgrade client.\n",
 437                                    sbi->ll_md_exp->exp_obd->obd_name, buf);
 438                 OBD_FREE(buf, PAGE_SIZE);
 439                 GOTO(out_md_fid, err = -EPROTO);
 440         }
 441
 442         size = sizeof(*data);
 443         err = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_CONN_DATA),
 444                            KEY_CONN_DATA,  &size, data);
 445         if (err) {
 446                 CERROR("%s: Get connect data failed: rc = %d\n",
 447                        sbi->ll_md_exp->exp_obd->obd_name, err);
 448                 GOTO(out_md_fid, err);
 449         }
 450
 451         LASSERT(osfs->os_bsize);
 452         sb->s_blocksize = osfs->os_bsize;
 453         sb->s_blocksize_bits = log2(osfs->os_bsize);
 454         sb->s_magic = LL_SUPER_MAGIC;
 455         sb->s_maxbytes = MAX_LFS_FILESIZE;
 456         sbi->ll_namelen = osfs->os_namelen;
 457         sbi->ll_mnt.mnt = current->fs->root.mnt;
 458
 459         if ((sbi->ll_flags & LL_SBI_USER_XATTR) &&
 460             !(data->ocd_connect_flags & OBD_CONNECT_XATTR)) {
 461                 LCONSOLE_INFO("Disabling user_xattr feature because "
 462                               "it is not supported on the server\n");
 463                 sbi->ll_flags &= ~LL_SBI_USER_XATTR;
 464         }
 465
 466         if (data->ocd_connect_flags & OBD_CONNECT_ACL) {
 467 #ifdef SB_POSIXACL
 468                 sb->s_flags |= SB_POSIXACL;
 469 #endif
 470                 sbi->ll_flags |= LL_SBI_ACL;
 471         } else {
 472                 LCONSOLE_INFO("client wants to enable acl, but mdt not!\n");
 473 #ifdef SB_POSIXACL
 474                 sb->s_flags &= ~SB_POSIXACL;
 475 #endif
 476                 sbi->ll_flags &= ~LL_SBI_ACL;
 477         }
 478
 479         if (data->ocd_connect_flags & OBD_CONNECT_64BITHASH)
 480                 sbi->ll_flags |= LL_SBI_64BIT_HASH;
 481
 482         if (data->ocd_connect_flags & OBD_CONNECT_LAYOUTLOCK)
 483                 sbi->ll_flags |= LL_SBI_LAYOUT_LOCK;
 484
 485         if (obd_connect_has_secctx(data))
 486                 sbi->ll_flags |= LL_SBI_FILE_SECCTX;
 487
 488         if (ll_sbi_has_encrypt(sbi) && !obd_connect_has_enc(data)) {
 489                 if (ll_sbi_has_test_dummy_encryption(sbi))
 490                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 491                                       sbi->ll_fsname,
 492                                       sbi->ll_md_exp->exp_obd->obd_name);
 493                 ll_sbi_set_encrypt(sbi, false);
 494         }
 495
 496         if (data->ocd_ibits_known & MDS_INODELOCK_XATTR) {
 497                 if (!(data->ocd_connect_flags & OBD_CONNECT_MAX_EASIZE)) {
 498                         LCONSOLE_INFO("%s: disabling xattr cache due to "
 499                                       "unknown maximum xattr size.\n", dt);
 500                 } else if (!sbi->ll_xattr_cache_set) {
 501                         /* If xattr_cache is already set (no matter 0 or 1)
 502                          * during processing llog, it won't be enabled here. */
 503                         sbi->ll_flags |= LL_SBI_XATTR_CACHE;
 504                         sbi->ll_xattr_cache_enabled = 1;
 505                 }
 506         }
 507
 508         sbi->ll_dt_obd = class_name2obd(dt);
 509         if (!sbi->ll_dt_obd) {
 510                 CERROR("DT %s: not setup or attached\n", dt);
 511                 GOTO(out_md_fid, err = -ENODEV);
 512         }
 513
 514         /* pass client page size via ocd_grant_blkbits, the server should report
 515          * back its backend blocksize for grant calculation purpose */
 516         data->ocd_grant_blkbits = PAGE_SHIFT;
 517
 518         /* indicate OST features supported by this client */
 519         data->ocd_connect_flags = OBD_CONNECT_GRANT | OBD_CONNECT_VERSION |
 520                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_BRW_SIZE |
 521                                   OBD_CONNECT_CANCELSET | OBD_CONNECT_FID |
 522                                   OBD_CONNECT_SRVLOCK | OBD_CONNECT_TRUNCLOCK|
 523                                   OBD_CONNECT_AT | OBD_CONNECT_OSS_CAPA |
 524                                   OBD_CONNECT_VBR | OBD_CONNECT_FULL20 |
 525                                   OBD_CONNECT_64BITHASH | OBD_CONNECT_MAXBYTES |
 526                                   OBD_CONNECT_EINPROGRESS |
 527                                   OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
 528                                   OBD_CONNECT_LAYOUTLOCK |
 529                                   OBD_CONNECT_PINGLESS | OBD_CONNECT_LFSCK |
 530                                   OBD_CONNECT_BULK_MBITS | OBD_CONNECT_SHORTIO |
 531                                   OBD_CONNECT_FLAGS2 | OBD_CONNECT_GRANT_SHRINK;
 532         data->ocd_connect_flags2 = OBD_CONNECT2_LOCKAHEAD |
 533                                    OBD_CONNECT2_INC_XID | OBD_CONNECT2_LSEEK;
 534
 535         if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_GRANT_PARAM))
 536                 data->ocd_connect_flags |= OBD_CONNECT_GRANT_PARAM;
 537
 538         /* OBD_CONNECT_CKSUM should always be set, even if checksums are
 539          * disabled by default, because it can still be enabled on the
 540          * fly via /sys. As a consequence, we still need to come to an
 541          * agreement on the supported algorithms at connect time
 542          */
 543         data->ocd_connect_flags |= OBD_CONNECT_CKSUM;
 544
 545         if (OBD_FAIL_CHECK(OBD_FAIL_OSC_CKSUM_ADLER_ONLY))
 546                 data->ocd_cksum_types = OBD_CKSUM_ADLER;
 547         else
 548                 data->ocd_cksum_types = obd_cksum_types_supported_client();
 549
 550 #ifdef HAVE_LRU_RESIZE_SUPPORT
 551         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 552 #endif
 553         /* always ping even if server suppress_pings */
 554         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
 555                 data->ocd_connect_flags &= ~OBD_CONNECT_PINGLESS;
 556
 557         if (ll_sbi_has_encrypt(sbi))
 558                 obd_connect_set_enc(data);
 559
 560         CDEBUG(D_RPCTRACE, "ocd_connect_flags: %#llx ocd_version: %d "
 561                "ocd_grant: %d\n", data->ocd_connect_flags,
 562                data->ocd_version, data->ocd_grant);
 563
 564         sbi->ll_dt_obd->obd_upcall.onu_owner = &sbi->ll_lco;
 565         sbi->ll_dt_obd->obd_upcall.onu_upcall = cl_ocd_update;
 566
 567         data->ocd_brw_size = DT_MAX_BRW_SIZE;
 568
 569         err = obd_connect(NULL, &sbi->ll_dt_exp, sbi->ll_dt_obd,
 570                           &sbi->ll_sb_uuid, data, sbi->ll_cache);
 571         if (err == -EBUSY) {
 572                 LCONSOLE_ERROR_MSG(0x150, "An OST (dt %s) is performing "
 573                                    "recovery, of which this client is not a "
 574                                    "part.  Please wait for recovery to "
 575                                    "complete, abort, or time out.\n", dt);
 576                 GOTO(out_md, err);
 577         } else if (err) {
 578                 CERROR("%s: Cannot connect to %s: rc = %d\n",
 579                        sbi->ll_dt_exp->exp_obd->obd_name, dt, err);
 580                 GOTO(out_md, err);
 581         }
 582
 583         if (ll_sbi_has_encrypt(sbi) &&
 584             !obd_connect_has_enc(&sbi->ll_dt_obd->u.lov.lov_ocd)) {
 585                 if (ll_sbi_has_test_dummy_encryption(sbi))
 586                         LCONSOLE_WARN("%s: server %s does not support encryption feature, encryption deactivated.\n",
 587                                       sbi->ll_fsname, dt);
 588                 ll_sbi_set_encrypt(sbi, false);
 589         } else if (ll_sbi_has_test_dummy_encryption(sbi)) {
 590                 LCONSOLE_WARN("Test dummy encryption mode enabled\n");
 591         }
 592
 593         sbi->ll_dt_exp->exp_connect_data = *data;
 594
 595         /* Don't change value if it was specified in the config log */
 596         if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
 597                 sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 598                         max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
 599                               (data->ocd_brw_size >> PAGE_SHIFT));
 600                 if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
 601                     sbi->ll_ra_info.ra_max_pages_per_file)
 602                         sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
 603                                 sbi->ll_ra_info.ra_max_pages_per_file;
 604         }
 605
 606         err = obd_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
 607                            LUSTRE_SEQ_METADATA);
 608         if (err) {
 609                 CERROR("%s: Can't init data layer FID infrastructure, "
 610                        "rc = %d\n", sbi->ll_dt_exp->exp_obd->obd_name, err);
 611                 GOTO(out_dt, err);
 612         }
 613
 614         mutex_lock(&sbi->ll_lco.lco_lock);
 615         sbi->ll_lco.lco_flags = data->ocd_connect_flags;
 616         sbi->ll_lco.lco_md_exp = sbi->ll_md_exp;
 617         sbi->ll_lco.lco_dt_exp = sbi->ll_dt_exp;
 618         mutex_unlock(&sbi->ll_lco.lco_lock);
 619
 620         fid_zero(&sbi->ll_root_fid);
 621         err = md_get_root(sbi->ll_md_exp, get_mount_fileset(sb),
 622                            &sbi->ll_root_fid);
 623         if (err) {
 624                 CERROR("cannot mds_connect: rc = %d\n", err);
 625                 GOTO(out_lock_cn_cb, err);
 626         }
 627         if (!fid_is_sane(&sbi->ll_root_fid)) {
 628                 CERROR("%s: Invalid root fid "DFID" during mount\n",
 629                        sbi->ll_md_exp->exp_obd->obd_name,
 630                        PFID(&sbi->ll_root_fid));
 631                 GOTO(out_lock_cn_cb, err = -EINVAL);
 632         }
 633         CDEBUG(D_SUPER, "rootfid "DFID"\n", PFID(&sbi->ll_root_fid));
 634
 635         sb->s_op = &lustre_super_operations;
 636         sb->s_xattr = ll_xattr_handlers;
 637 #if THREAD_SIZE >= 8192 /*b=17630*/
 638         sb->s_export_op = &lustre_export_operations;
 639 #endif
 640 #ifdef HAVE_LUSTRE_CRYPTO
 641         llcrypt_set_ops(sb, &lustre_cryptops);
 642 #endif
 643
 644         /* make root inode
 645          * XXX: move this to after cbd setup? */
 646         valid = OBD_MD_FLGETATTR | OBD_MD_FLBLOCKS | OBD_MD_FLMODEASIZE;
 647         if (sbi->ll_flags & LL_SBI_ACL)
 648                 valid |= OBD_MD_FLACL;
 649
 650         OBD_ALLOC_PTR(op_data);
 651         if (op_data == NULL)
 652                 GOTO(out_lock_cn_cb, err = -ENOMEM);
 653
 654         op_data->op_fid1 = sbi->ll_root_fid;
 655         op_data->op_mode = 0;
 656         op_data->op_valid = valid;
 657
 658         err = md_getattr(sbi->ll_md_exp, op_data, &request);
 659
 660         OBD_FREE_PTR(op_data);
 661         if (err) {
 662                 CERROR("%s: md_getattr failed for root: rc = %d\n",
 663                        sbi->ll_md_exp->exp_obd->obd_name, err);
 664                 GOTO(out_lock_cn_cb, err);
 665         }
 666
 667         err = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
 668                                sbi->ll_md_exp, &lmd);
 669         if (err) {
 670                 CERROR("failed to understand root inode md: rc = %d\n", err);
 671                 ptlrpc_req_finished(request);
 672                 GOTO(out_lock_cn_cb, err);
 673         }
 674
 675         LASSERT(fid_is_sane(&sbi->ll_root_fid));
 676         root = ll_iget(sb, cl_fid_build_ino(&sbi->ll_root_fid,
 677                                             sbi->ll_flags & LL_SBI_32BIT_API),
 678                        &lmd);
 679         md_free_lustre_md(sbi->ll_md_exp, &lmd);
 680         ptlrpc_req_finished(request);
 681
 682         if (IS_ERR(root)) {
 683                 lmd_clear_acl(&lmd);
 684                 err = IS_ERR(root) ? PTR_ERR(root) : -EBADF;
 685                 root = NULL;
 686                 CERROR("%s: bad ll_iget() for root: rc = %d\n",
 687                        sbi->ll_fsname, err);
 688                 GOTO(out_root, err);
 689         }
 690
 691         checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
 692         if (sbi->ll_checksum_set) {
 693                 err = obd_set_info_async(NULL, sbi->ll_dt_exp,
 694                                          sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 695                                          sizeof(checksum), &checksum, NULL);
 696                 if (err) {
 697                         CERROR("%s: Set checksum failed: rc = %d\n",
 698                                sbi->ll_dt_exp->exp_obd->obd_name, err);
 699                         GOTO(out_root, err);
 700                 }
 701         }
 702         cl_sb_init(sb);
 703
 704         sb->s_root = d_make_root(root);
 705         if (sb->s_root == NULL) {
 706                 err = -ENOMEM;
 707                 CERROR("%s: can't make root dentry: rc = %d\n",
 708                        sbi->ll_fsname, err);
 709                 GOTO(out_root, err);
 710         }
 711
 712         sbi->ll_sdev_orig = sb->s_dev;
 713
 714         /* We set sb->s_dev equal on all lustre clients in order to support
 715          * NFS export clustering.  NFSD requires that the FSID be the same
 716          * on all clients. */
 717         /* s_dev is also used in lt_compare() to compare two fs, but that is
 718          * only a node-local comparison. */
 719         uuid = obd_get_uuid(sbi->ll_md_exp);
 720         if (uuid != NULL)
 721                 sb->s_dev = get_uuid2int(uuid->uuid, strlen(uuid->uuid));
 722
 723         if (data != NULL)
 724                 OBD_FREE_PTR(data);
 725         if (osfs != NULL)
 726                 OBD_FREE_PTR(osfs);
 727
 728         if (sbi->ll_dt_obd) {
 729                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 730                                         &sbi->ll_dt_obd->obd_kset.kobj,
 731                                         sbi->ll_dt_obd->obd_type->typ_name);
 732                 if (err < 0) {
 733                         CERROR("%s: could not register %s in llite: rc = %d\n",
 734                                dt, sbi->ll_fsname, err);
 735                         err = 0;
 736                 }
 737         }
 738
 739         if (sbi->ll_md_obd) {
 740                 err = sysfs_create_link(&sbi->ll_kset.kobj,
 741                                         &sbi->ll_md_obd->obd_kset.kobj,
 742                                         sbi->ll_md_obd->obd_type->typ_name);
 743                 if (err < 0) {
 744                         CERROR("%s: could not register %s in llite: rc = %d\n",
 745                                md, sbi->ll_fsname, err);
 746                         err = 0;
 747                 }
 748         }
 749
 750         RETURN(err);
 751 out_root:
 752         if (root)
 753                 iput(root);
 754 out_lock_cn_cb:
 755         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 756 out_dt:
 757         obd_disconnect(sbi->ll_dt_exp);
 758         sbi->ll_dt_exp = NULL;
 759         sbi->ll_dt_obd = NULL;
 760 out_md_fid:
 761         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 762 out_md:
 763         obd_disconnect(sbi->ll_md_exp);
 764         sbi->ll_md_exp = NULL;
 765         sbi->ll_md_obd = NULL;
 766 out:
 767         if (data != NULL)
 768                 OBD_FREE_PTR(data);
 769         if (osfs != NULL)
 770                 OBD_FREE_PTR(osfs);
 771         return err;
 772 }
 773
 774 int ll_get_max_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 775 {
 776         int size, rc;
 777
 778         size = sizeof(*lmmsize);
 779         rc = obd_get_info(NULL, sbi->ll_dt_exp, sizeof(KEY_MAX_EASIZE),
 780                           KEY_MAX_EASIZE, &size, lmmsize);
 781         if (rc != 0) {
 782                 CERROR("%s: cannot get max LOV EA size: rc = %d\n",
 783                        sbi->ll_dt_exp->exp_obd->obd_name, rc);
 784                 RETURN(rc);
 785         }
 786
 787         CDEBUG(D_INFO, "max LOV ea size: %d\n", *lmmsize);
 788
 789         size = sizeof(int);
 790         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_MAX_EASIZE),
 791                           KEY_MAX_EASIZE, &size, lmmsize);
 792         if (rc)
 793                 CERROR("Get max mdsize error rc %d\n", rc);
 794
 795         CDEBUG(D_INFO, "max LMV ea size: %d\n", *lmmsize);
 796
 797         RETURN(rc);
 798 }
 799
 800 /**
 801  * Get the value of the default_easize parameter.
 802  *
 803  * \see client_obd::cl_default_mds_easize
 804  *
 805  * \param[in] sbi       superblock info for this filesystem
 806  * \param[out] lmmsize  pointer to storage location for value
 807  *
 808  * \retval 0            on success
 809  * \retval negative     negated errno on failure
 810  */
 811 int ll_get_default_mdsize(struct ll_sb_info *sbi, int *lmmsize)
 812 {
 813         int size, rc;
 814
 815         size = sizeof(int);
 816         rc = obd_get_info(NULL, sbi->ll_md_exp, sizeof(KEY_DEFAULT_EASIZE),
 817                          KEY_DEFAULT_EASIZE, &size, lmmsize);
 818         if (rc)
 819                 CERROR("Get default mdsize error rc %d\n", rc);
 820
 821         RETURN(rc);
 822 }
 823
 824 /**
 825  * Set the default_easize parameter to the given value.
 826  *
 827  * \see client_obd::cl_default_mds_easize
 828  *
 829  * \param[in] sbi       superblock info for this filesystem
 830  * \param[in] lmmsize   the size to set
 831  *
 832  * \retval 0            on success
 833  * \retval negative     negated errno on failure
 834  */
 835 int ll_set_default_mdsize(struct ll_sb_info *sbi, int lmmsize)
 836 {
 837         int rc;
 838
 839         if (lmmsize < sizeof(struct lov_mds_md) ||
 840             lmmsize > OBD_MAX_DEFAULT_EA_SIZE)
 841                 return -EINVAL;
 842
 843         rc = obd_set_info_async(NULL, sbi->ll_md_exp,
 844                                 sizeof(KEY_DEFAULT_EASIZE), KEY_DEFAULT_EASIZE,
 845                                 sizeof(int), &lmmsize, NULL);
 846
 847         RETURN(rc);
 848 }
 849
 850 static void client_common_put_super(struct super_block *sb)
 851 {
 852         struct ll_sb_info *sbi = ll_s2sbi(sb);
 853         ENTRY;
 854
 855         cl_sb_fini(sb);
 856
 857         obd_fid_fini(sbi->ll_dt_exp->exp_obd);
 858         obd_disconnect(sbi->ll_dt_exp);
 859         sbi->ll_dt_exp = NULL;
 860
 861         ll_debugfs_unregister_super(sb);
 862
 863         obd_fid_fini(sbi->ll_md_exp->exp_obd);
 864         obd_disconnect(sbi->ll_md_exp);
 865         sbi->ll_md_exp = NULL;
 866
 867         EXIT;
 868 }
 869
 870 void ll_kill_super(struct super_block *sb)
 871 {
 872         struct ll_sb_info *sbi;
 873         ENTRY;
 874
 875         /* not init sb ?*/
 876         if (!(sb->s_flags & SB_ACTIVE))
 877                 return;
 878
 879         sbi = ll_s2sbi(sb);
 880         /* we need restore s_dev from changed for clustred NFS before put_super
 881          * because new kernels have cached s_dev and change sb->s_dev in
 882          * put_super not affected real removing devices */
 883         if (sbi) {
 884                 sb->s_dev = sbi->ll_sdev_orig;
 885
 886                 /* wait running statahead threads to quit */
 887                 while (atomic_read(&sbi->ll_sa_running) > 0)
 888                         schedule_timeout_uninterruptible(
 889                                 cfs_time_seconds(1) >> 3);
 890         }
 891
 892         EXIT;
 893 }
 894
 895 static inline int ll_set_opt(const char *opt, char *data, int fl)
 896 {
 897         if (strncmp(opt, data, strlen(opt)) != 0)
 898                 return 0;
 899         else
 900                 return fl;
 901 }
 902
 903 /* non-client-specific mount options are parsed in lmd_parse */
 904 static int ll_options(char *options, struct ll_sb_info *sbi)
 905 {
 906         int tmp;
 907         char *s1 = options, *s2;
 908         int *flags = &sbi->ll_flags;
 909         ENTRY;
 910
 911         if (!options)
 912                 RETURN(0);
 913
 914         CDEBUG(D_CONFIG, "Parsing opts %s\n", options);
 915
 916         while (*s1) {
 917                 CDEBUG(D_SUPER, "next opt=%s\n", s1);
 918                 tmp = ll_set_opt("nolock", s1, LL_SBI_NOLCK);
 919                 if (tmp) {
 920                         *flags |= tmp;
 921                         goto next;
 922                 }
 923                 tmp = ll_set_opt("flock", s1, LL_SBI_FLOCK);
 924                 if (tmp) {
 925                         *flags = (*flags & ~LL_SBI_LOCALFLOCK) | tmp;
 926                         goto next;
 927                 }
 928                 tmp = ll_set_opt("localflock", s1, LL_SBI_LOCALFLOCK);
 929                 if (tmp) {
 930                         *flags = (*flags & ~LL_SBI_FLOCK) | tmp;
 931                         goto next;
 932                 }
 933                 tmp = ll_set_opt("noflock", s1, LL_SBI_FLOCK|LL_SBI_LOCALFLOCK);
 934                 if (tmp) {
 935                         *flags &= ~tmp;
 936                         goto next;
 937                 }
 938                 tmp = ll_set_opt("user_xattr", s1, LL_SBI_USER_XATTR);
 939                 if (tmp) {
 940                         *flags |= tmp;
 941                         goto next;
 942                 }
 943                 tmp = ll_set_opt("nouser_xattr", s1, LL_SBI_USER_XATTR);
 944                 if (tmp) {
 945                         *flags &= ~tmp;
 946                         goto next;
 947                 }
 948                 tmp = ll_set_opt("context", s1, 1);
 949                 if (tmp)
 950                         goto next;
 951                 tmp = ll_set_opt("fscontext", s1, 1);
 952                 if (tmp)
 953                         goto next;
 954                 tmp = ll_set_opt("defcontext", s1, 1);
 955                 if (tmp)
 956                         goto next;
 957                 tmp = ll_set_opt("rootcontext", s1, 1);
 958                 if (tmp)
 959                         goto next;
 960                 tmp = ll_set_opt("user_fid2path", s1, LL_SBI_USER_FID2PATH);
 961                 if (tmp) {
 962                         *flags |= tmp;
 963                         goto next;
 964                 }
 965                 tmp = ll_set_opt("nouser_fid2path", s1, LL_SBI_USER_FID2PATH);
 966                 if (tmp) {
 967                         *flags &= ~tmp;
 968                         goto next;
 969                 }
 970
 971                 tmp = ll_set_opt("checksum", s1, LL_SBI_CHECKSUM);
 972                 if (tmp) {
 973                         *flags |= tmp;
 974                         sbi->ll_checksum_set = 1;
 975                         goto next;
 976                 }
 977                 tmp = ll_set_opt("nochecksum", s1, LL_SBI_CHECKSUM);
 978                 if (tmp) {
 979                         *flags &= ~tmp;
 980                         sbi->ll_checksum_set = 1;
 981                         goto next;
 982                 }
 983                 tmp = ll_set_opt("lruresize", s1, LL_SBI_LRU_RESIZE);
 984                 if (tmp) {
 985                         *flags |= tmp;
 986                         goto next;
 987                 }
 988                 tmp = ll_set_opt("nolruresize", s1, LL_SBI_LRU_RESIZE);
 989                 if (tmp) {
 990                         *flags &= ~tmp;
 991                         goto next;
 992                 }
 993                 tmp = ll_set_opt("lazystatfs", s1, LL_SBI_LAZYSTATFS);
 994                 if (tmp) {
 995                         *flags |= tmp;
 996                         goto next;
 997                 }
 998                 tmp = ll_set_opt("nolazystatfs", s1, LL_SBI_LAZYSTATFS);
 999                 if (tmp) {
1000                         *flags &= ~tmp;
1001                         goto next;
1002                 }
1003                 tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
1004                 if (tmp) {
1005                         *flags |= tmp;
1006                         goto next;
1007                 }
1008                 tmp = ll_set_opt("verbose", s1, LL_SBI_VERBOSE);
1009                 if (tmp) {
1010                         *flags |= tmp;
1011                         goto next;
1012                 }
1013                 tmp = ll_set_opt("noverbose", s1, LL_SBI_VERBOSE);
1014                 if (tmp) {
1015                         *flags &= ~tmp;
1016                         goto next;
1017                 }
1018                 tmp = ll_set_opt("always_ping", s1, LL_SBI_ALWAYS_PING);
1019                 if (tmp) {
1020                         *flags |= tmp;
1021                         goto next;
1022                 }
1023                 tmp = ll_set_opt("test_dummy_encryption", s1,
1024                                  LL_SBI_TEST_DUMMY_ENCRYPTION);
1025                 if (tmp) {
1026 #ifdef HAVE_LUSTRE_CRYPTO
1027                         *flags |= tmp;
1028 #else
1029                         LCONSOLE_WARN("Test dummy encryption mount option ignored: encryption not supported\n");
1030 #endif
1031                         goto next;
1032                 }
1033                 tmp = ll_set_opt("noencrypt", s1, LL_SBI_ENCRYPT);
1034                 if (tmp) {
1035 #ifdef HAVE_LUSTRE_CRYPTO
1036                         *flags &= ~tmp;
1037 #else
1038                         LCONSOLE_WARN("noencrypt mount option ignored: encryption not supported\n");
1039 #endif
1040                         goto next;
1041                 }
1042                 tmp = ll_set_opt("foreign_symlink", s1, LL_SBI_FOREIGN_SYMLINK);
1043                 if (tmp) {
1044                         int prefix_pos = sizeof("foreign_symlink=") - 1;
1045                         int equal_pos = sizeof("foreign_symlink=") - 2;
1046
1047                         /* non-default prefix provided ? */
1048                         if (strlen(s1) >= sizeof("foreign_symlink=") &&
1049                             *(s1 + equal_pos) == '=') {
1050                                 char *old = sbi->ll_foreign_symlink_prefix;
1051                                 size_t old_len =
1052                                         sbi->ll_foreign_symlink_prefix_size;
1053
1054                                 /* path must be absolute */
1055                                 if (*(s1 + sizeof("foreign_symlink=")
1056                                       - 1) != '/') {
1057                                         LCONSOLE_ERROR_MSG(0x152,
1058                                                 "foreign prefix '%s' must be an absolute path\n",
1059                                                 s1 + prefix_pos);
1060                                         RETURN(-EINVAL);
1061                                 }
1062                                 /* last option ? */
1063                                 s2 = strchrnul(s1 + prefix_pos, ',');
1064
1065                                 if (sbi->ll_foreign_symlink_prefix) {
1066                                         sbi->ll_foreign_symlink_prefix = NULL;
1067                                         sbi->ll_foreign_symlink_prefix_size = 0;
1068                                 }
1069                                 /* alloc for path length and '\0' */
1070                                 OBD_ALLOC(sbi->ll_foreign_symlink_prefix,
1071                                                 s2 - (s1 + prefix_pos) + 1);
1072                                 if (!sbi->ll_foreign_symlink_prefix) {
1073                                         /* restore previous */
1074                                         sbi->ll_foreign_symlink_prefix = old;
1075                                         sbi->ll_foreign_symlink_prefix_size =
1076                                                 old_len;
1077                                         RETURN(-ENOMEM);
1078                                 }
1079                                 if (old)
1080                                         OBD_FREE(old, old_len);
1081                                 strncpy(sbi->ll_foreign_symlink_prefix,
1082                                         s1 + prefix_pos,
1083                                         s2 - (s1 + prefix_pos));
1084                                 sbi->ll_foreign_symlink_prefix_size =
1085                                         s2 - (s1 + prefix_pos) + 1;
1086                         } else {
1087                                 LCONSOLE_ERROR_MSG(0x152,
1088                                                    "invalid %s option\n", s1);
1089                         }
1090                         /* enable foreign symlink support */
1091                         *flags |= tmp;
1092                         goto next;
1093                 }
1094                 LCONSOLE_ERROR_MSG(0x152, "Unknown option '%s', won't mount.\n",
1095                                    s1);
1096                 RETURN(-EINVAL);
1097
1098 next:
1099                 /* Find next opt */
1100                 s2 = strchr(s1, ',');
1101                 if (s2 == NULL)
1102                         break;
1103                 s1 = s2 + 1;
1104         }
1105         RETURN(0);
1106 }
1107
1108 void ll_lli_init(struct ll_inode_info *lli)
1109 {
1110         lli->lli_inode_magic = LLI_INODE_MAGIC;
1111         lli->lli_flags = 0;
1112         spin_lock_init(&lli->lli_lock);
1113         lli->lli_posix_acl = NULL;
1114         /* Do not set lli_fid, it has been initialized already. */
1115         fid_zero(&lli->lli_pfid);
1116         lli->lli_mds_read_och = NULL;
1117         lli->lli_mds_write_och = NULL;
1118         lli->lli_mds_exec_och = NULL;
1119         lli->lli_open_fd_read_count = 0;
1120         lli->lli_open_fd_write_count = 0;
1121         lli->lli_open_fd_exec_count = 0;
1122         mutex_init(&lli->lli_och_mutex);
1123         spin_lock_init(&lli->lli_agl_lock);
1124         spin_lock_init(&lli->lli_layout_lock);
1125         ll_layout_version_set(lli, CL_LAYOUT_GEN_NONE);
1126         lli->lli_clob = NULL;
1127
1128         init_rwsem(&lli->lli_xattrs_list_rwsem);
1129         mutex_init(&lli->lli_xattrs_enq_lock);
1130
1131         LASSERT(lli->lli_vfs_inode.i_mode != 0);
1132         if (S_ISDIR(lli->lli_vfs_inode.i_mode)) {
1133                 lli->lli_opendir_key = NULL;
1134                 lli->lli_sai = NULL;
1135                 spin_lock_init(&lli->lli_sa_lock);
1136                 lli->lli_opendir_pid = 0;
1137                 lli->lli_sa_enabled = 0;
1138                 init_rwsem(&lli->lli_lsm_sem);
1139         } else {
1140                 mutex_init(&lli->lli_size_mutex);
1141                 mutex_init(&lli->lli_setattr_mutex);
1142                 lli->lli_symlink_name = NULL;
1143                 ll_trunc_sem_init(&lli->lli_trunc_sem);
1144                 range_lock_tree_init(&lli->lli_write_tree);
1145                 init_rwsem(&lli->lli_glimpse_sem);
1146                 lli->lli_glimpse_time = ktime_set(0, 0);
1147                 INIT_LIST_HEAD(&lli->lli_agl_list);
1148                 lli->lli_agl_index = 0;
1149                 lli->lli_async_rc = 0;
1150                 spin_lock_init(&lli->lli_heat_lock);
1151                 obd_heat_clear(lli->lli_heat_instances, OBD_HEAT_COUNT);
1152                 lli->lli_heat_flags = 0;
1153                 mutex_init(&lli->lli_pcc_lock);
1154                 lli->lli_pcc_state = PCC_STATE_FL_NONE;
1155                 lli->lli_pcc_inode = NULL;
1156                 lli->lli_pcc_dsflags = PCC_DATASET_INVALID;
1157                 lli->lli_pcc_generation = 0;
1158                 mutex_init(&lli->lli_group_mutex);
1159                 lli->lli_group_users = 0;
1160                 lli->lli_group_gid = 0;
1161         }
1162         mutex_init(&lli->lli_layout_mutex);
1163         memset(lli->lli_jobid, 0, sizeof(lli->lli_jobid));
1164 }
1165
1166 #define MAX_STRING_SIZE 128
1167
1168 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1169
1170 #define LSI_BDI_INITIALIZED     0x00400000
1171
1172 #ifndef HAVE_BDI_CAP_MAP_COPY
1173 # define BDI_CAP_MAP_COPY       0
1174 #endif
1175
1176 static int super_setup_bdi_name(struct super_block *sb, char *fmt, ...)
1177 {
1178         struct  lustre_sb_info *lsi = s2lsi(sb);
1179         char buf[MAX_STRING_SIZE];
1180         va_list args;
1181         int err;
1182
1183         err = bdi_init(&lsi->lsi_bdi);
1184         if (err)
1185                 return err;
1186
1187         lsi->lsi_flags |= LSI_BDI_INITIALIZED;
1188         lsi->lsi_bdi.capabilities = BDI_CAP_MAP_COPY;
1189         lsi->lsi_bdi.name = "lustre";
1190         va_start(args, fmt);
1191         vsnprintf(buf, MAX_STRING_SIZE, fmt, args);
1192         va_end(args);
1193         err = bdi_register(&lsi->lsi_bdi, NULL, "%s", buf);
1194         va_end(args);
1195         if (!err)
1196                 sb->s_bdi = &lsi->lsi_bdi;
1197
1198         return err;
1199 }
1200 #endif /* !HAVE_SUPER_SETUP_BDI_NAME */
1201
1202 int ll_fill_super(struct super_block *sb)
1203 {
1204         struct  lustre_profile *lprof = NULL;
1205         struct  lustre_sb_info *lsi = s2lsi(sb);
1206         struct  ll_sb_info *sbi = NULL;
1207         char    *dt = NULL, *md = NULL;
1208         char    *profilenm = get_profile_name(sb);
1209         struct config_llog_instance *cfg;
1210         /* %p for void* in printf needs 16+2 characters: 0xffffffffffffffff */
1211         const int instlen = LUSTRE_MAXINSTANCE + 2;
1212         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1213         char name[MAX_STRING_SIZE];
1214         int md_len = 0;
1215         int dt_len = 0;
1216         uuid_t uuid;
1217         char *ptr;
1218         int len;
1219         int err;
1220
1221         ENTRY;
1222         /* for ASLR, to map between cfg_instance and hashed ptr */
1223         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1224                profilenm, cfg_instance, sb);
1225
1226         OBD_ALLOC_PTR(cfg);
1227         if (cfg == NULL)
1228                 GOTO(out_free_cfg, err = -ENOMEM);
1229
1230         /* client additional sb info */
1231         lsi->lsi_llsbi = sbi = ll_init_sbi();
1232         if (IS_ERR(sbi))
1233                 GOTO(out_free_cfg, err = PTR_ERR(sbi));
1234
1235         err = ll_options(lsi->lsi_lmd->lmd_opts, sbi);
1236         if (err)
1237                 GOTO(out_free_cfg, err);
1238
1239         /* kernel >= 2.6.38 store dentry operations in sb->s_d_op. */
1240         sb->s_d_op = &ll_d_ops;
1241
1242         /* UUID handling */
1243         generate_random_uuid(uuid.b);
1244         snprintf(sbi->ll_sb_uuid.uuid, sizeof(sbi->ll_sb_uuid), "%pU", uuid.b);
1245
1246         CDEBUG(D_CONFIG, "llite sb uuid: %s\n", sbi->ll_sb_uuid.uuid);
1247
1248         /* Get fsname */
1249         len = strlen(profilenm);
1250         ptr = strrchr(profilenm, '-');
1251         if (ptr && (strcmp(ptr, "-client") == 0))
1252                 len -= 7;
1253
1254         if (len > LUSTRE_MAXFSNAME) {
1255                 if (unlikely(len >= MAX_STRING_SIZE))
1256                         len = MAX_STRING_SIZE - 1;
1257                 strncpy(name, profilenm, len);
1258                 name[len] = '\0';
1259                 err = -ENAMETOOLONG;
1260                 CERROR("%s: fsname longer than %u characters: rc = %d\n",
1261                        name, LUSTRE_MAXFSNAME, err);
1262                 GOTO(out_free_cfg, err);
1263         }
1264         strncpy(sbi->ll_fsname, profilenm, len);
1265         sbi->ll_fsname[len] = '\0';
1266
1267         /* Mount info */
1268         snprintf(name, sizeof(name), "%.*s-%016lx", len,
1269                  profilenm, cfg_instance);
1270
1271         err = super_setup_bdi_name(sb, "%s", name);
1272         if (err)
1273                 GOTO(out_free_cfg, err);
1274
1275         /* Call ll_debugfs_register_super() before lustre_process_log()
1276          * so that "llite.*.*" params can be processed correctly.
1277          */
1278         err = ll_debugfs_register_super(sb, name);
1279         if (err < 0) {
1280                 CERROR("%s: could not register mountpoint in llite: rc = %d\n",
1281                        sbi->ll_fsname, err);
1282                 err = 0;
1283         }
1284
1285         /* The cfg_instance is a value unique to this super, in case some
1286          * joker tries to mount the same fs at two mount points.
1287          */
1288         cfg->cfg_instance = cfg_instance;
1289         cfg->cfg_uuid = lsi->lsi_llsbi->ll_sb_uuid;
1290         cfg->cfg_callback = class_config_llog_handler;
1291         cfg->cfg_sub_clds = CONFIG_SUB_CLIENT;
1292         /* set up client obds */
1293         err = lustre_process_log(sb, profilenm, cfg);
1294         if (err < 0)
1295                 GOTO(out_debugfs, err);
1296
1297         /* Profile set with LCFG_MOUNTOPT so we can find our mdc and osc obds */
1298         lprof = class_get_profile(profilenm);
1299         if (lprof == NULL) {
1300                 LCONSOLE_ERROR_MSG(0x156, "The client profile '%s' could not be"
1301                                    " read from the MGS.  Does that filesystem "
1302                                    "exist?\n", profilenm);
1303                 GOTO(out_debugfs, err = -EINVAL);
1304         }
1305         CDEBUG(D_CONFIG, "Found profile %s: mdc=%s osc=%s\n", profilenm,
1306                lprof->lp_md, lprof->lp_dt);
1307
1308         dt_len = strlen(lprof->lp_dt) + instlen + 2;
1309         OBD_ALLOC(dt, dt_len);
1310         if (!dt)
1311                 GOTO(out_profile, err = -ENOMEM);
1312         snprintf(dt, dt_len - 1, "%s-%016lx", lprof->lp_dt, cfg_instance);
1313
1314         md_len = strlen(lprof->lp_md) + instlen + 2;
1315         OBD_ALLOC(md, md_len);
1316         if (!md)
1317                 GOTO(out_free_dt, err = -ENOMEM);
1318         snprintf(md, md_len - 1, "%s-%016lx", lprof->lp_md, cfg_instance);
1319
1320         /* connections, registrations, sb setup */
1321         err = client_common_fill_super(sb, md, dt);
1322         if (err < 0)
1323                 GOTO(out_free_md, err);
1324
1325         sbi->ll_client_common_fill_super_succeeded = 1;
1326
1327 out_free_md:
1328         if (md)
1329                 OBD_FREE(md, md_len);
1330 out_free_dt:
1331         if (dt)
1332                 OBD_FREE(dt, dt_len);
1333 out_profile:
1334         if (lprof)
1335                 class_put_profile(lprof);
1336 out_debugfs:
1337         if (err < 0)
1338                 ll_debugfs_unregister_super(sb);
1339 out_free_cfg:
1340         if (cfg)
1341                 OBD_FREE_PTR(cfg);
1342
1343         if (err)
1344                 ll_put_super(sb);
1345         else if (sbi->ll_flags & LL_SBI_VERBOSE)
1346                 LCONSOLE_WARN("Mounted %s\n", profilenm);
1347         RETURN(err);
1348 } /* ll_fill_super */
1349
1350 void ll_put_super(struct super_block *sb)
1351 {
1352         struct config_llog_instance cfg, params_cfg;
1353         struct obd_device *obd;
1354         struct lustre_sb_info *lsi = s2lsi(sb);
1355         struct ll_sb_info *sbi = ll_s2sbi(sb);
1356         char *profilenm = get_profile_name(sb);
1357         unsigned long cfg_instance = ll_get_cfg_instance(sb);
1358         long ccc_count;
1359         int next, force = 1, rc = 0;
1360         ENTRY;
1361
1362         if (IS_ERR(sbi))
1363                 GOTO(out_no_sbi, 0);
1364
1365         /* Should replace instance_id with something better for ASLR */
1366         CDEBUG(D_VFSTRACE, "VFS Op: cfg_instance %s-%016lx (sb %p)\n",
1367                profilenm, cfg_instance, sb);
1368
1369         cfg.cfg_instance = cfg_instance;
1370         lustre_end_log(sb, profilenm, &cfg);
1371
1372         params_cfg.cfg_instance = cfg_instance;
1373         lustre_end_log(sb, PARAMS_FILENAME, &params_cfg);
1374
1375         if (sbi->ll_md_exp) {
1376                 obd = class_exp2obd(sbi->ll_md_exp);
1377                 if (obd)
1378                         force = obd->obd_force;
1379         }
1380
1381         /* Wait for unstable pages to be committed to stable storage */
1382         if (force == 0) {
1383                 rc = l_wait_event_abortable(
1384                         sbi->ll_cache->ccc_unstable_waitq,
1385                         atomic_long_read(&sbi->ll_cache->ccc_unstable_nr) == 0);
1386         }
1387
1388         ccc_count = atomic_long_read(&sbi->ll_cache->ccc_unstable_nr);
1389         if (force == 0 && rc != -ERESTARTSYS)
1390                 LASSERTF(ccc_count == 0, "count: %li\n", ccc_count);
1391
1392         /* We need to set force before the lov_disconnect in
1393          * lustre_common_put_super, since l_d cleans up osc's as well.
1394          */
1395         if (force) {
1396                 next = 0;
1397                 while ((obd = class_devices_in_group(&sbi->ll_sb_uuid,
1398                                                      &next)) != NULL) {
1399                         obd->obd_force = force;
1400                 }
1401         }
1402
1403         if (sbi->ll_client_common_fill_super_succeeded) {
1404                 /* Only if client_common_fill_super succeeded */
1405                 client_common_put_super(sb);
1406         }
1407
1408         next = 0;
1409         while ((obd = class_devices_in_group(&sbi->ll_sb_uuid, &next)))
1410                 class_manual_cleanup(obd);
1411
1412         if (sbi->ll_flags & LL_SBI_VERBOSE)
1413                 LCONSOLE_WARN("Unmounted %s\n", profilenm ? profilenm : "");
1414
1415         if (profilenm)
1416                 class_del_profile(profilenm);
1417
1418 #ifndef HAVE_SUPER_SETUP_BDI_NAME
1419         if (lsi->lsi_flags & LSI_BDI_INITIALIZED) {
1420                 bdi_destroy(&lsi->lsi_bdi);
1421                 lsi->lsi_flags &= ~LSI_BDI_INITIALIZED;
1422         }
1423 #endif
1424
1425         ll_free_sbi(sb);
1426         lsi->lsi_llsbi = NULL;
1427 out_no_sbi:
1428         lustre_common_put_super(sb);
1429
1430         cl_env_cache_purge(~0);
1431
1432         module_put(THIS_MODULE);
1433
1434         EXIT;
1435 } /* client_put_super */
1436
1437 struct inode *ll_inode_from_resource_lock(struct ldlm_lock *lock)
1438 {
1439         struct inode *inode = NULL;
1440
1441         /* NOTE: we depend on atomic igrab() -bzzz */
1442         lock_res_and_lock(lock);
1443         if (lock->l_resource->lr_lvb_inode) {
1444                 struct ll_inode_info * lli;
1445                 lli = ll_i2info(lock->l_resource->lr_lvb_inode);
1446                 if (lli->lli_inode_magic == LLI_INODE_MAGIC) {
1447                         inode = igrab(lock->l_resource->lr_lvb_inode);
1448                 } else {
1449                         inode = lock->l_resource->lr_lvb_inode;
1450                         LDLM_DEBUG_LIMIT(inode->i_state & I_FREEING ?  D_INFO :
1451                                          D_WARNING, lock, "lr_lvb_inode %p is "
1452                                          "bogus: magic %08x",
1453                                          lock->l_resource->lr_lvb_inode,
1454                                          lli->lli_inode_magic);
1455                         inode = NULL;
1456                 }
1457         }
1458         unlock_res_and_lock(lock);
1459         return inode;
1460 }
1461
1462 void ll_dir_clear_lsm_md(struct inode *inode)
1463 {
1464         struct ll_inode_info *lli = ll_i2info(inode);
1465
1466         LASSERT(S_ISDIR(inode->i_mode));
1467
1468         if (lli->lli_lsm_md) {
1469                 lmv_free_memmd(lli->lli_lsm_md);
1470                 lli->lli_lsm_md = NULL;
1471         }
1472
1473         if (lli->lli_default_lsm_md) {
1474                 lmv_free_memmd(lli->lli_default_lsm_md);
1475                 lli->lli_default_lsm_md = NULL;
1476         }
1477 }
1478
1479 static struct inode *ll_iget_anon_dir(struct super_block *sb,
1480                                       const struct lu_fid *fid,
1481                                       struct lustre_md *md)
1482 {
1483         struct ll_sb_info       *sbi = ll_s2sbi(sb);
1484         struct mdt_body         *body = md->body;
1485         struct inode            *inode;
1486         ino_t                   ino;
1487         ENTRY;
1488
1489         ino = cl_fid_build_ino(fid, sbi->ll_flags & LL_SBI_32BIT_API);
1490         inode = iget_locked(sb, ino);
1491         if (inode == NULL) {
1492                 CERROR("%s: failed get simple inode "DFID": rc = -ENOENT\n",
1493                        sbi->ll_fsname, PFID(fid));
1494                 RETURN(ERR_PTR(-ENOENT));
1495         }
1496
1497         if (inode->i_state & I_NEW) {
1498                 struct ll_inode_info *lli = ll_i2info(inode);
1499                 struct lmv_stripe_md *lsm = md->lmv;
1500
1501                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
1502                                 (body->mbo_mode & S_IFMT);
1503                 LASSERTF(S_ISDIR(inode->i_mode), "Not slave inode "DFID"\n",
1504                          PFID(fid));
1505
1506                 inode->i_mtime.tv_sec = 0;
1507                 inode->i_atime.tv_sec = 0;
1508                 inode->i_ctime.tv_sec = 0;
1509                 inode->i_rdev = 0;
1510
1511 #ifdef HAVE_BACKING_DEV_INFO
1512                 /* initializing backing dev info. */
1513                 inode->i_mapping->backing_dev_info =
1514                                                 &s2lsi(inode->i_sb)->lsi_bdi;
1515 #endif
1516                 inode->i_op = &ll_dir_inode_operations;
1517                 inode->i_fop = &ll_dir_operations;
1518                 lli->lli_fid = *fid;
1519                 ll_lli_init(lli);
1520
1521                 LASSERT(lsm != NULL);
1522                 /* master object FID */
1523                 lli->lli_pfid = body->mbo_fid1;
1524                 CDEBUG(D_INODE, "lli %p slave "DFID" master "DFID"\n",
1525                        lli, PFID(fid), PFID(&lli->lli_pfid));
1526                 unlock_new_inode(inode);
1527         }
1528
1529         RETURN(inode);
1530 }
1531
1532 static int ll_init_lsm_md(struct inode *inode, struct lustre_md *md)
1533 {
1534         struct lu_fid *fid;
1535         struct lmv_stripe_md *lsm = md->lmv;
1536         struct ll_inode_info *lli = ll_i2info(inode);
1537         int i;
1538
1539         LASSERT(lsm != NULL);
1540
1541         CDEBUG(D_INODE, "%s: "DFID" set dir layout:\n",
1542                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1543         lsm_md_dump(D_INODE, lsm);
1544
1545         if (!lmv_dir_striped(lsm))
1546                 goto out;
1547
1548         /* XXX sigh, this lsm_root initialization should be in
1549          * LMV layer, but it needs ll_iget right now, so we
1550          * put this here right now. */
1551         for (i = 0; i < lsm->lsm_md_stripe_count; i++) {
1552                 fid = &lsm->lsm_md_oinfo[i].lmo_fid;
1553                 LASSERT(lsm->lsm_md_oinfo[i].lmo_root == NULL);
1554
1555                 if (!fid_is_sane(fid))
1556                         continue;
1557
1558                 /* Unfortunately ll_iget will call ll_update_inode,
1559                  * where the initialization of slave inode is slightly
1560                  * different, so it reset lsm_md to NULL to avoid
1561                  * initializing lsm for slave inode. */
1562                 lsm->lsm_md_oinfo[i].lmo_root =
1563                                 ll_iget_anon_dir(inode->i_sb, fid, md);
1564                 if (IS_ERR(lsm->lsm_md_oinfo[i].lmo_root)) {
1565                         int rc = PTR_ERR(lsm->lsm_md_oinfo[i].lmo_root);
1566
1567                         lsm->lsm_md_oinfo[i].lmo_root = NULL;
1568                         while (i-- > 0) {
1569                                 iput(lsm->lsm_md_oinfo[i].lmo_root);
1570                                 lsm->lsm_md_oinfo[i].lmo_root = NULL;
1571                         }
1572                         return rc;
1573                 }
1574         }
1575 out:
1576         lli->lli_lsm_md = lsm;
1577
1578         return 0;
1579 }
1580
1581 static void ll_update_default_lsm_md(struct inode *inode, struct lustre_md *md)
1582 {
1583         struct ll_inode_info *lli = ll_i2info(inode);
1584
1585         if (!md->default_lmv) {
1586                 /* clear default lsm */
1587                 if (lli->lli_default_lsm_md) {
1588                         down_write(&lli->lli_lsm_sem);
1589                         if (lli->lli_default_lsm_md) {
1590                                 lmv_free_memmd(lli->lli_default_lsm_md);
1591                                 lli->lli_default_lsm_md = NULL;
1592                         }
1593                         up_write(&lli->lli_lsm_sem);
1594                 }
1595         } else if (lli->lli_default_lsm_md) {
1596                 /* update default lsm if it changes */
1597                 down_read(&lli->lli_lsm_sem);
1598                 if (lli->lli_default_lsm_md &&
1599                     !lsm_md_eq(lli->lli_default_lsm_md, md->default_lmv)) {
1600                         up_read(&lli->lli_lsm_sem);
1601                         down_write(&lli->lli_lsm_sem);
1602                         if (lli->lli_default_lsm_md)
1603                                 lmv_free_memmd(lli->lli_default_lsm_md);
1604                         lli->lli_default_lsm_md = md->default_lmv;
1605                         lsm_md_dump(D_INODE, md->default_lmv);
1606                         md->default_lmv = NULL;
1607                         up_write(&lli->lli_lsm_sem);
1608                 } else {
1609                         up_read(&lli->lli_lsm_sem);
1610                 }
1611         } else {
1612                 /* init default lsm */
1613                 down_write(&lli->lli_lsm_sem);
1614                 lli->lli_default_lsm_md = md->default_lmv;
1615                 lsm_md_dump(D_INODE, md->default_lmv);
1616                 md->default_lmv = NULL;
1617                 up_write(&lli->lli_lsm_sem);
1618         }
1619 }
1620
1621 static int ll_update_lsm_md(struct inode *inode, struct lustre_md *md)
1622 {
1623         struct ll_inode_info *lli = ll_i2info(inode);
1624         struct lmv_stripe_md *lsm = md->lmv;
1625         struct cl_attr  *attr;
1626         int rc = 0;
1627
1628         ENTRY;
1629
1630         LASSERT(S_ISDIR(inode->i_mode));
1631         CDEBUG(D_INODE, "update lsm %p of "DFID"\n", lli->lli_lsm_md,
1632                PFID(ll_inode2fid(inode)));
1633
1634         /* update default LMV */
1635         if (md->default_lmv)
1636                 ll_update_default_lsm_md(inode, md);
1637
1638         /*
1639          * no striped information from request, lustre_md from req does not
1640          * include stripeEA, see ll_md_setattr()
1641          */
1642         if (!lsm)
1643                 RETURN(0);
1644
1645         /*
1646          * normally dir layout doesn't change, only take read lock to check
1647          * that to avoid blocking other MD operations.
1648          */
1649         down_read(&lli->lli_lsm_sem);
1650
1651         /* some current lookup initialized lsm, and unchanged */
1652         if (lli->lli_lsm_md && lsm_md_eq(lli->lli_lsm_md, lsm))
1653                 GOTO(unlock, rc = 0);
1654
1655         /* if dir layout doesn't match, check whether version is increased,
1656          * which means layout is changed, this happens in dir split/merge and
1657          * lfsck.
1658          *
1659          * foreign LMV should not change.
1660          */
1661         if (lli->lli_lsm_md && lmv_dir_striped(lli->lli_lsm_md) &&
1662             lsm->lsm_md_layout_version <=
1663             lli->lli_lsm_md->lsm_md_layout_version) {
1664                 CERROR("%s: "DFID" dir layout mismatch:\n",
1665                        ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid));
1666                 lsm_md_dump(D_ERROR, lli->lli_lsm_md);
1667                 lsm_md_dump(D_ERROR, lsm);
1668                 GOTO(unlock, rc = -EINVAL);
1669         }
1670
1671         up_read(&lli->lli_lsm_sem);
1672         down_write(&lli->lli_lsm_sem);
1673         /* clear existing lsm */
1674         if (lli->lli_lsm_md) {
1675                 lmv_free_memmd(lli->lli_lsm_md);
1676                 lli->lli_lsm_md = NULL;
1677         }
1678
1679         rc = ll_init_lsm_md(inode, md);
1680         up_write(&lli->lli_lsm_sem);
1681
1682         if (rc)
1683                 RETURN(rc);
1684
1685         /* set md->lmv to NULL, so the following free lustre_md will not free
1686          * this lsm.
1687          */
1688         md->lmv = NULL;
1689
1690         /* md_merge_attr() may take long, since lsm is already set, switch to
1691          * read lock.
1692          */
1693         down_read(&lli->lli_lsm_sem);
1694
1695         if (!lmv_dir_striped(lli->lli_lsm_md))
1696                 GOTO(unlock, rc = 0);
1697
1698         OBD_ALLOC_PTR(attr);
1699         if (!attr)
1700                 GOTO(unlock, rc = -ENOMEM);
1701
1702         /* validate the lsm */
1703         rc = md_merge_attr(ll_i2mdexp(inode), lli->lli_lsm_md, attr,
1704                            ll_md_blocking_ast);
1705         if (!rc) {
1706                 if (md->body->mbo_valid & OBD_MD_FLNLINK)
1707                         md->body->mbo_nlink = attr->cat_nlink;
1708                 if (md->body->mbo_valid & OBD_MD_FLSIZE)
1709                         md->body->mbo_size = attr->cat_size;
1710                 if (md->body->mbo_valid & OBD_MD_FLATIME)
1711                         md->body->mbo_atime = attr->cat_atime;
1712                 if (md->body->mbo_valid & OBD_MD_FLCTIME)
1713                         md->body->mbo_ctime = attr->cat_ctime;
1714                 if (md->body->mbo_valid & OBD_MD_FLMTIME)
1715                         md->body->mbo_mtime = attr->cat_mtime;
1716         }
1717
1718         OBD_FREE_PTR(attr);
1719         GOTO(unlock, rc);
1720 unlock:
1721         up_read(&lli->lli_lsm_sem);
1722
1723         return rc;
1724 }
1725
1726 void ll_clear_inode(struct inode *inode)
1727 {
1728         struct ll_inode_info *lli = ll_i2info(inode);
1729         struct ll_sb_info *sbi = ll_i2sbi(inode);
1730
1731         ENTRY;
1732
1733         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
1734                PFID(ll_inode2fid(inode)), inode);
1735
1736         if (S_ISDIR(inode->i_mode)) {
1737                 /* these should have been cleared in ll_file_release */
1738                 LASSERT(lli->lli_opendir_key == NULL);
1739                 LASSERT(lli->lli_sai == NULL);
1740                 LASSERT(lli->lli_opendir_pid == 0);
1741         } else {
1742                 pcc_inode_free(inode);
1743         }
1744
1745         md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
1746
1747         LASSERT(!lli->lli_open_fd_write_count);
1748         LASSERT(!lli->lli_open_fd_read_count);
1749         LASSERT(!lli->lli_open_fd_exec_count);
1750
1751         if (lli->lli_mds_write_och)
1752                 ll_md_real_close(inode, FMODE_WRITE);
1753         if (lli->lli_mds_exec_och)
1754                 ll_md_real_close(inode, FMODE_EXEC);
1755         if (lli->lli_mds_read_och)
1756                 ll_md_real_close(inode, FMODE_READ);
1757
1758         if (S_ISLNK(inode->i_mode) && lli->lli_symlink_name) {
1759                 OBD_FREE(lli->lli_symlink_name,
1760                          strlen(lli->lli_symlink_name) + 1);
1761                 lli->lli_symlink_name = NULL;
1762         }
1763
1764         ll_xattr_cache_destroy(inode);
1765
1766         forget_all_cached_acls(inode);
1767         lli_clear_acl(lli);
1768         lli->lli_inode_magic = LLI_INODE_DEAD;
1769
1770         if (S_ISDIR(inode->i_mode))
1771                 ll_dir_clear_lsm_md(inode);
1772         else if (S_ISREG(inode->i_mode) && !is_bad_inode(inode))
1773                 LASSERT(list_empty(&lli->lli_agl_list));
1774
1775         /*
1776          * XXX This has to be done before lsm is freed below, because
1777          * cl_object still uses inode lsm.
1778          */
1779         cl_inode_fini(inode);
1780
1781         llcrypt_put_encryption_info(inode);
1782
1783         EXIT;
1784 }
1785
1786 static int ll_md_setattr(struct dentry *dentry, struct md_op_data *op_data)
1787 {
1788         struct lustre_md md;
1789         struct inode *inode = dentry->d_inode;
1790         struct ll_sb_info *sbi = ll_i2sbi(inode);
1791         struct ptlrpc_request *request = NULL;
1792         int rc, ia_valid;
1793         ENTRY;
1794
1795         op_data = ll_prep_md_op_data(op_data, inode, NULL, NULL, 0, 0,
1796                                      LUSTRE_OPC_ANY, NULL);
1797         if (IS_ERR(op_data))
1798                 RETURN(PTR_ERR(op_data));
1799
1800         rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &request);
1801         if (rc) {
1802                 ptlrpc_req_finished(request);
1803                 if (rc == -ENOENT) {
1804                         clear_nlink(inode);
1805                         /* Unlinked special device node? Or just a race?
1806                          * Pretend we done everything. */
1807                         if (!S_ISREG(inode->i_mode) &&
1808                             !S_ISDIR(inode->i_mode)) {
1809                                 ia_valid = op_data->op_attr.ia_valid;
1810                                 op_data->op_attr.ia_valid &= ~TIMES_SET_FLAGS;
1811                                 rc = simple_setattr(dentry, &op_data->op_attr);
1812                                 op_data->op_attr.ia_valid = ia_valid;
1813                         }
1814                 } else if (rc != -EPERM && rc != -EACCES && rc != -ETXTBSY) {
1815                         CERROR("md_setattr fails: rc = %d\n", rc);
1816                 }
1817                 RETURN(rc);
1818         }
1819
1820         rc = md_get_lustre_md(sbi->ll_md_exp, request, sbi->ll_dt_exp,
1821                               sbi->ll_md_exp, &md);
1822         if (rc) {
1823                 ptlrpc_req_finished(request);
1824                 RETURN(rc);
1825         }
1826
1827         ia_valid = op_data->op_attr.ia_valid;
1828         /* inode size will be in ll_setattr_ost, can't do it now since dirty
1829          * cache is not cleared yet. */
1830         op_data->op_attr.ia_valid &= ~(TIMES_SET_FLAGS | ATTR_SIZE);
1831         if (S_ISREG(inode->i_mode))
1832                 inode_lock(inode);
1833         rc = simple_setattr(dentry, &op_data->op_attr);
1834         if (S_ISREG(inode->i_mode))
1835                 inode_unlock(inode);
1836         op_data->op_attr.ia_valid = ia_valid;
1837
1838         rc = ll_update_inode(inode, &md);
1839         ptlrpc_req_finished(request);
1840
1841         RETURN(rc);
1842 }
1843
1844 /**
1845  * Zero portion of page that is part of @inode.
1846  * This implies, if necessary:
1847  * - taking cl_lock on range corresponding to concerned page
1848  * - grabbing vm page
1849  * - associating cl_page
1850  * - proceeding to clio read
1851  * - zeroing range in page
1852  * - proceeding to cl_page flush
1853  * - releasing cl_lock
1854  *
1855  * \param[in] inode     inode
1856  * \param[in] index     page index
1857  * \param[in] offset    offset in page to start zero from
1858  * \param[in] len       len to zero
1859  *
1860  * \retval 0            on success
1861  * \retval negative     errno on failure
1862  */
1863 int ll_io_zero_page(struct inode *inode, pgoff_t index, pgoff_t offset,
1864                     unsigned len)
1865 {
1866         struct ll_inode_info *lli = ll_i2info(inode);
1867         struct cl_object *clob = lli->lli_clob;
1868         __u16 refcheck;
1869         struct lu_env *env = NULL;
1870         struct cl_io *io = NULL;
1871         struct cl_page *clpage = NULL;
1872         struct page *vmpage = NULL;
1873         unsigned from = index << PAGE_SHIFT;
1874         struct cl_lock *lock = NULL;
1875         struct cl_lock_descr *descr = NULL;
1876         struct cl_2queue *queue = NULL;
1877         struct cl_sync_io *anchor = NULL;
1878         bool holdinglock = false;
1879         bool lockedbymyself = true;
1880         int rc;
1881
1882         ENTRY;
1883
1884         env = cl_env_get(&refcheck);
1885         if (IS_ERR(env))
1886                 RETURN(PTR_ERR(env));
1887
1888         io = vvp_env_thread_io(env);
1889         io->ci_obj = clob;
1890         rc = cl_io_rw_init(env, io, CIT_WRITE, from, PAGE_SIZE);
1891         if (rc)
1892                 GOTO(putenv, rc);
1893
1894         lock = vvp_env_lock(env);
1895         descr = &lock->cll_descr;
1896         descr->cld_obj   = io->ci_obj;
1897         descr->cld_start = cl_index(io->ci_obj, from);
1898         descr->cld_end   = cl_index(io->ci_obj, from + PAGE_SIZE - 1);
1899         descr->cld_mode  = CLM_WRITE;
1900         descr->cld_enq_flags = CEF_MUST | CEF_NONBLOCK;
1901
1902         /* request lock for page */
1903         rc = cl_lock_request(env, io, lock);
1904         /* -ECANCELED indicates a matching lock with a different extent
1905          * was already present, and -EEXIST indicates a matching lock
1906          * on exactly the same extent was already present.
1907          * In both cases it means we are covered.
1908          */
1909         if (rc == -ECANCELED || rc == -EEXIST)
1910                 rc = 0;
1911         else if (rc < 0)
1912                 GOTO(iofini, rc);
1913         else
1914                 holdinglock = true;
1915
1916         /* grab page */
1917         vmpage = grab_cache_page_nowait(inode->i_mapping, index);
1918         if (vmpage == NULL)
1919                 GOTO(rellock, rc = -EOPNOTSUPP);
1920
1921         if (!PageDirty(vmpage)) {
1922                 /* associate cl_page */
1923                 clpage = cl_page_find(env, clob, vmpage->index,
1924                                       vmpage, CPT_CACHEABLE);
1925                 if (IS_ERR(clpage))
1926                         GOTO(pagefini, rc = PTR_ERR(clpage));
1927
1928                 cl_page_assume(env, io, clpage);
1929         }
1930
1931         if (!PageUptodate(vmpage) && !PageDirty(vmpage) &&
1932             !PageWriteback(vmpage)) {
1933                 /* read page */
1934                 /* set PagePrivate2 to detect special case of empty page
1935                  * in osc_brw_fini_request()
1936                  */
1937                 SetPagePrivate2(vmpage);
1938                 rc = ll_io_read_page(env, io, clpage, NULL);
1939                 if (!PagePrivate2(vmpage))
1940                         /* PagePrivate2 was cleared in osc_brw_fini_request()
1941                          * meaning we read an empty page. In this case, in order
1942                          * to avoid allocating unnecessary block in truncated
1943                          * file, we must not zero and write as below. Subsequent
1944                          * server-side truncate will handle things correctly.
1945                          */
1946                         GOTO(clpfini, rc = 0);
1947                 ClearPagePrivate2(vmpage);
1948                 if (rc)
1949                         GOTO(clpfini, rc);
1950                 lockedbymyself = trylock_page(vmpage);
1951                 cl_page_assume(env, io, clpage);
1952         }
1953
1954         /* zero range in page */
1955         zero_user(vmpage, offset, len);
1956
1957         if (holdinglock && clpage) {
1958                 /* explicitly write newly modified page */
1959                 queue = &io->ci_queue;
1960                 cl_2queue_init(queue);
1961                 anchor = &vvp_env_info(env)->vti_anchor;
1962                 cl_sync_io_init(anchor, 1);
1963                 clpage->cp_sync_io = anchor;
1964                 cl_2queue_add(queue, clpage);
1965                 rc = cl_io_submit_rw(env, io, CRT_WRITE, queue);
1966                 if (rc)
1967                         GOTO(queuefini1, rc);
1968                 rc = cl_sync_io_wait(env, anchor, 0);
1969                 if (rc)
1970                         GOTO(queuefini2, rc);
1971                 cl_page_assume(env, io, clpage);
1972
1973 queuefini2:
1974                 cl_2queue_discard(env, io, queue);
1975 queuefini1:
1976                 cl_2queue_disown(env, io, queue);
1977                 cl_2queue_fini(env, queue);
1978         }
1979
1980 clpfini:
1981         if (clpage)
1982                 cl_page_put(env, clpage);
1983 pagefini:
1984         if (lockedbymyself) {
1985                 unlock_page(vmpage);
1986                 put_page(vmpage);
1987         }
1988 rellock:
1989         if (holdinglock)
1990                 cl_lock_release(env, lock);
1991 iofini:
1992         cl_io_fini(env, io);
1993 putenv:
1994         if (env)
1995                 cl_env_put(env, &refcheck);
1996
1997         RETURN(rc);
1998 }
1999
2000 /* If this inode has objects allocated to it (lsm != NULL), then the OST
2001  * object(s) determine the file size and mtime.  Otherwise, the MDS will
2002  * keep these values until such a time that objects are allocated for it.
2003  * We do the MDS operations first, as it is checking permissions for us.
2004  * We don't to the MDS RPC if there is nothing that we want to store there,
2005  * otherwise there is no harm in updating mtime/atime on the MDS if we are
2006  * going to do an RPC anyways.
2007  *
2008  * If we are doing a truncate, we will send the mtime and ctime updates
2009  * to the OST with the punch RPC, otherwise we do an explicit setattr RPC.
2010  * I don't believe it is possible to get e.g. ATTR_MTIME_SET and ATTR_SIZE
2011  * at the same time.
2012  *
2013  * In case of HSMimport, we only set attr on MDS.
2014  */
2015 int ll_setattr_raw(struct dentry *dentry, struct iattr *attr,
2016                    enum op_xvalid xvalid, bool hsm_import)
2017 {
2018         struct inode *inode = dentry->d_inode;
2019         struct ll_inode_info *lli = ll_i2info(inode);
2020         struct md_op_data *op_data = NULL;
2021         ktime_t kstart = ktime_get();
2022         int rc = 0;
2023
2024         ENTRY;
2025
2026         CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, "
2027                "valid %x, hsm_import %d\n",
2028                ll_i2sbi(inode)->ll_fsname, PFID(&lli->lli_fid),
2029                inode, i_size_read(inode), attr->ia_size, attr->ia_valid,
2030                hsm_import);
2031
2032         if (attr->ia_valid & ATTR_SIZE) {
2033                 /* Check new size against VFS/VM file size limit and rlimit */
2034                 rc = inode_newsize_ok(inode, attr->ia_size);
2035                 if (rc)
2036                         RETURN(rc);
2037
2038                 /* The maximum Lustre file size is variable, based on the
2039                  * OST maximum object size and number of stripes.  This
2040                  * needs another check in addition to the VFS check above. */
2041                 if (attr->ia_size > ll_file_maxbytes(inode)) {
2042                         CDEBUG(D_INODE,"file "DFID" too large %llu > %llu\n",
2043                                PFID(&lli->lli_fid), attr->ia_size,
2044                                ll_file_maxbytes(inode));
2045                         RETURN(-EFBIG);
2046                 }
2047
2048                 attr->ia_valid |= ATTR_MTIME | ATTR_CTIME;
2049         }
2050
2051         /* POSIX: check before ATTR_*TIME_SET set (from inode_change_ok) */
2052         if (attr->ia_valid & TIMES_SET_FLAGS) {
2053                 if ((!uid_eq(current_fsuid(), inode->i_uid)) &&
2054                     !cfs_capable(CAP_FOWNER))
2055                         RETURN(-EPERM);
2056         }
2057
2058         /* We mark all of the fields "set" so MDS/OST does not re-set them */
2059         if (!(xvalid & OP_XVALID_CTIME_SET) &&
2060              (attr->ia_valid & ATTR_CTIME)) {
2061                 attr->ia_ctime = current_time(inode);
2062                 xvalid |= OP_XVALID_CTIME_SET;
2063         }
2064         if (!(attr->ia_valid & ATTR_ATIME_SET) &&
2065             (attr->ia_valid & ATTR_ATIME)) {
2066                 attr->ia_atime = current_time(inode);
2067                 attr->ia_valid |= ATTR_ATIME_SET;
2068         }
2069         if (!(attr->ia_valid & ATTR_MTIME_SET) &&
2070             (attr->ia_valid & ATTR_MTIME)) {
2071                 attr->ia_mtime = current_time(inode);
2072                 attr->ia_valid |= ATTR_MTIME_SET;
2073         }
2074
2075         if (attr->ia_valid & (ATTR_MTIME | ATTR_CTIME))
2076                 CDEBUG(D_INODE, "setting mtime %lld, ctime %lld, now = %lld\n",
2077                        (s64)attr->ia_mtime.tv_sec, (s64)attr->ia_ctime.tv_sec,
2078                        ktime_get_real_seconds());
2079
2080         if (S_ISREG(inode->i_mode))
2081                 inode_unlock(inode);
2082
2083         /* We always do an MDS RPC, even if we're only changing the size;
2084          * only the MDS knows whether truncate() should fail with -ETXTBUSY */
2085
2086         OBD_ALLOC_PTR(op_data);
2087         if (op_data == NULL)
2088                 GOTO(out, rc = -ENOMEM);
2089
2090         if (!hsm_import && attr->ia_valid & ATTR_SIZE) {
2091                 /* If we are changing file size, file content is
2092                  * modified, flag it.
2093                  */
2094                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2095                 op_data->op_bias |= MDS_DATA_MODIFIED;
2096                 ll_file_clear_flag(lli, LLIF_DATA_MODIFIED);
2097         }
2098
2099         if (attr->ia_valid & ATTR_FILE) {
2100                 struct ll_file_data *fd = attr->ia_file->private_data;
2101
2102                 if (fd->fd_lease_och)
2103                         op_data->op_bias |= MDS_TRUNC_KEEP_LEASE;
2104         }
2105
2106         op_data->op_attr = *attr;
2107         op_data->op_xvalid = xvalid;
2108
2109         rc = ll_md_setattr(dentry, op_data);
2110         if (rc)
2111                 GOTO(out, rc);
2112
2113         if (!S_ISREG(inode->i_mode) || hsm_import)
2114                 GOTO(out, rc = 0);
2115
2116         if (attr->ia_valid & (ATTR_SIZE | ATTR_ATIME | ATTR_ATIME_SET |
2117                               ATTR_MTIME | ATTR_MTIME_SET | ATTR_CTIME) ||
2118             xvalid & OP_XVALID_CTIME_SET) {
2119                 bool cached = false;
2120
2121                 rc = pcc_inode_setattr(inode, attr, &cached);
2122                 if (cached) {
2123                         if (rc) {
2124                                 CERROR("%s: PCC inode "DFID" setattr failed: "
2125                                        "rc = %d\n",
2126                                        ll_i2sbi(inode)->ll_fsname,
2127                                        PFID(&lli->lli_fid), rc);
2128                                 GOTO(out, rc);
2129                         }
2130                 } else {
2131                         unsigned int flags = 0;
2132
2133                         /* For truncate and utimes sending attributes to OSTs,
2134                          * setting mtime/atime to the past will be performed
2135                          * under PW [0:EOF] extent lock (new_size:EOF for
2136                          * truncate). It may seem excessive to send mtime/atime
2137                          * updates to OSTs when not setting times to past, but
2138                          * it is necessary due to possible time
2139                          * de-synchronization between MDT inode and OST objects
2140                          */
2141                         if (S_ISREG(inode->i_mode) && IS_ENCRYPTED(inode) &&
2142                             attr->ia_valid & ATTR_SIZE) {
2143                                 xvalid |= OP_XVALID_FLAGS;
2144                                 flags = LUSTRE_ENCRYPT_FL;
2145                                 /* Call to ll_io_zero_page is not necessary if
2146                                  * truncating on PAGE_SIZE boundary, because
2147                                  * whole pages will be wiped.
2148                                  * In case of Direct IO, all we need is to set
2149                                  * new size.
2150                                  */
2151                                 if (attr->ia_size & ~PAGE_MASK &&
2152                                     !(attr->ia_valid & ATTR_FILE &&
2153                                       attr->ia_file->f_flags & O_DIRECT)) {
2154                                         pgoff_t offset =
2155                                                 attr->ia_size & (PAGE_SIZE - 1);
2156
2157                                         rc = ll_io_zero_page(inode,
2158                                                     attr->ia_size >> PAGE_SHIFT,
2159                                                     offset, PAGE_SIZE - offset);
2160                                         if (rc)
2161                                                 GOTO(out, rc);
2162                                 }
2163                         }
2164                         rc = cl_setattr_ost(lli->lli_clob, attr, xvalid, flags);
2165                 }
2166         }
2167
2168         /* If the file was restored, it needs to set dirty flag.
2169          *
2170          * We've already sent MDS_DATA_MODIFIED flag in
2171          * ll_md_setattr() for truncate. However, the MDT refuses to
2172          * set the HS_DIRTY flag on released files, so we have to set
2173          * it again if the file has been restored. Please check how
2174          * LLIF_DATA_MODIFIED is set in vvp_io_setattr_fini().
2175          *
2176          * Please notice that if the file is not released, the previous
2177          * MDS_DATA_MODIFIED has taken effect and usually
2178          * LLIF_DATA_MODIFIED is not set(see vvp_io_setattr_fini()).
2179          * This way we can save an RPC for common open + trunc
2180          * operation. */
2181         if (ll_file_test_and_clear_flag(lli, LLIF_DATA_MODIFIED)) {
2182                 struct hsm_state_set hss = {
2183                         .hss_valid = HSS_SETMASK,
2184                         .hss_setmask = HS_DIRTY,
2185                 };
2186                 int rc2;
2187
2188                 rc2 = ll_hsm_state_set(inode, &hss);
2189                 /* truncate and write can happen at the same time, so that
2190                  * the file can be set modified even though the file is not
2191                  * restored from released state, and ll_hsm_state_set() is
2192                  * not applicable for the file, and rc2 < 0 is normal in this
2193                  * case. */
2194                 if (rc2 < 0)
2195                         CDEBUG(D_INFO, DFID "HSM set dirty failed: rc2 = %d\n",
2196                                PFID(ll_inode2fid(inode)), rc2);
2197         }
2198
2199         EXIT;
2200 out:
2201         if (op_data != NULL)
2202                 ll_finish_md_op_data(op_data);
2203
2204         if (S_ISREG(inode->i_mode)) {
2205                 inode_lock(inode);
2206                 if ((attr->ia_valid & ATTR_SIZE) && !hsm_import)
2207                         inode_dio_wait(inode);
2208                 /* Once we've got the i_mutex, it's safe to set the S_NOSEC
2209                  * flag.  ll_update_inode (called from ll_md_setattr), clears
2210                  * inode flags, so there is a gap where S_NOSEC is not set.
2211                  * This can cause a writer to take the i_mutex unnecessarily,
2212                  * but this is safe to do and should be rare. */
2213                 inode_has_no_xattr(inode);
2214         }
2215
2216         if (!rc)
2217                 ll_stats_ops_tally(ll_i2sbi(inode), attr->ia_valid & ATTR_SIZE ?
2218                                         LPROC_LL_TRUNC : LPROC_LL_SETATTR,
2219                                    ktime_us_delta(ktime_get(), kstart));
2220
2221         return rc;
2222 }
2223
2224 int ll_setattr(struct dentry *de, struct iattr *attr)
2225 {
2226         int mode = de->d_inode->i_mode;
2227         enum op_xvalid xvalid = 0;
2228         int rc;
2229
2230         rc = llcrypt_prepare_setattr(de, attr);
2231         if (rc)
2232                 return rc;
2233
2234         if ((attr->ia_valid & (ATTR_CTIME|ATTR_SIZE|ATTR_MODE)) ==
2235                               (ATTR_CTIME|ATTR_SIZE|ATTR_MODE))
2236                 xvalid |= OP_XVALID_OWNEROVERRIDE;
2237
2238         if (((attr->ia_valid & (ATTR_MODE|ATTR_FORCE|ATTR_SIZE)) ==
2239                                (ATTR_SIZE|ATTR_MODE)) &&
2240             (((mode & S_ISUID) && !(attr->ia_mode & S_ISUID)) ||
2241              (((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2242               !(attr->ia_mode & S_ISGID))))
2243                 attr->ia_valid |= ATTR_FORCE;
2244
2245         if ((attr->ia_valid & ATTR_MODE) &&
2246             (mode & S_ISUID) &&
2247             !(attr->ia_mode & S_ISUID) &&
2248             !(attr->ia_valid & ATTR_KILL_SUID))
2249                 attr->ia_valid |= ATTR_KILL_SUID;
2250
2251         if ((attr->ia_valid & ATTR_MODE) &&
2252             ((mode & (S_ISGID|S_IXGRP)) == (S_ISGID|S_IXGRP)) &&
2253             !(attr->ia_mode & S_ISGID) &&
2254             !(attr->ia_valid & ATTR_KILL_SGID))
2255                 attr->ia_valid |= ATTR_KILL_SGID;
2256
2257         return ll_setattr_raw(de, attr, xvalid, false);
2258 }
2259
2260 int ll_statfs_internal(struct ll_sb_info *sbi, struct obd_statfs *osfs,
2261                        u32 flags)
2262 {
2263         struct obd_statfs obd_osfs = { 0 };
2264         time64_t max_age;
2265         int rc;
2266
2267         ENTRY;
2268         max_age = ktime_get_seconds() - sbi->ll_statfs_max_age;
2269
2270         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
2271                 flags |= OBD_STATFS_NODELAY;
2272
2273         rc = obd_statfs(NULL, sbi->ll_md_exp, osfs, max_age, flags);
2274         if (rc)
2275                 RETURN(rc);
2276
2277         osfs->os_type = LL_SUPER_MAGIC;
2278
2279         CDEBUG(D_SUPER, "MDC blocks %llu/%llu objects %llu/%llu\n",
2280               osfs->os_bavail, osfs->os_blocks, osfs->os_ffree, osfs->os_files);
2281
2282         if (osfs->os_state & OS_STATFS_SUM)
2283                 GOTO(out, rc);
2284
2285         rc = obd_statfs(NULL, sbi->ll_dt_exp, &obd_osfs, max_age, flags);
2286         if (rc) /* Possibly a filesystem with no OSTs.  Report MDT totals. */
2287                 GOTO(out, rc = 0);
2288
2289         CDEBUG(D_SUPER, "OSC blocks %llu/%llu objects %llu/%llu\n",
2290                obd_osfs.os_bavail, obd_osfs.os_blocks, obd_osfs.os_ffree,
2291                obd_osfs.os_files);
2292
2293         osfs->os_bsize = obd_osfs.os_bsize;
2294         osfs->os_blocks = obd_osfs.os_blocks;
2295         osfs->os_bfree = obd_osfs.os_bfree;
2296         osfs->os_bavail = obd_osfs.os_bavail;
2297
2298         /* If we have _some_ OSTs, but don't have as many free objects on the
2299          * OSTs as inodes on the MDTs, reduce the reported number of inodes
2300          * to compensate, so that the "inodes in use" number is correct.
2301          * This should be kept in sync with lod_statfs() behaviour.
2302          */
2303         if (obd_osfs.os_files && obd_osfs.os_ffree < osfs->os_ffree) {
2304                 osfs->os_files = (osfs->os_files - osfs->os_ffree) +
2305                                  obd_osfs.os_ffree;
2306                 osfs->os_ffree = obd_osfs.os_ffree;
2307         }
2308
2309 out:
2310         RETURN(rc);
2311 }
2312
2313 static int ll_statfs_project(struct inode *inode, struct kstatfs *sfs)
2314 {
2315         struct if_quotactl qctl = {
2316                 .qc_cmd = LUSTRE_Q_GETQUOTA,
2317                 .qc_type = PRJQUOTA,
2318                 .qc_valid = QC_GENERAL,
2319         };
2320         u64 limit, curblock;
2321         int ret;
2322
2323         qctl.qc_id = ll_i2info(inode)->lli_projid;
2324         ret = quotactl_ioctl(ll_i2sbi(inode), &qctl);
2325         if (ret) {
2326                 /* ignore errors if project ID does not have
2327                  * a quota limit or feature unsupported.
2328                  */
2329                 if (ret == -ESRCH || ret == -EOPNOTSUPP)
2330                         ret = 0;
2331                 return ret;
2332         }
2333
2334         limit = ((qctl.qc_dqblk.dqb_bsoftlimit ?
2335                  qctl.qc_dqblk.dqb_bsoftlimit :
2336                  qctl.qc_dqblk.dqb_bhardlimit) * 1024) / sfs->f_bsize;
2337         if (limit && sfs->f_blocks > limit) {
2338                 curblock = (qctl.qc_dqblk.dqb_curspace +
2339                                 sfs->f_bsize - 1) / sfs->f_bsize;
2340                 sfs->f_blocks = limit;
2341                 sfs->f_bfree = sfs->f_bavail =
2342                         (sfs->f_blocks > curblock) ?
2343                         (sfs->f_blocks - curblock) : 0;
2344         }
2345
2346         limit = qctl.qc_dqblk.dqb_isoftlimit ?
2347                 qctl.qc_dqblk.dqb_isoftlimit :
2348                 qctl.qc_dqblk.dqb_ihardlimit;
2349         if (limit && sfs->f_files > limit) {
2350                 sfs->f_files = limit;
2351                 sfs->f_ffree = (sfs->f_files >
2352                         qctl.qc_dqblk.dqb_curinodes) ?
2353                         (sfs->f_files - qctl.qc_dqblk.dqb_curinodes) : 0;
2354         }
2355
2356         return 0;
2357 }
2358
2359 int ll_statfs(struct dentry *de, struct kstatfs *sfs)
2360 {
2361         struct super_block *sb = de->d_sb;
2362         struct obd_statfs osfs;
2363         __u64 fsid = huge_encode_dev(sb->s_dev);
2364         ktime_t kstart = ktime_get();
2365         int rc;
2366
2367         CDEBUG(D_VFSTRACE, "VFS Op:sb=%s (%p)\n", sb->s_id, sb);
2368
2369         /* Some amount of caching on the client is allowed */
2370         rc = ll_statfs_internal(ll_s2sbi(sb), &osfs, OBD_STATFS_SUM);
2371         if (rc)
2372                 return rc;
2373
2374         statfs_unpack(sfs, &osfs);
2375
2376         /* We need to downshift for all 32-bit kernels, because we can't
2377          * tell if the kernel is being called via sys_statfs64() or not.
2378          * Stop before overflowing f_bsize - in which case it is better
2379          * to just risk EOVERFLOW if caller is using old sys_statfs(). */
2380         if (sizeof(long) < 8) {
2381                 while (osfs.os_blocks > ~0UL && sfs->f_bsize < 0x40000000) {
2382                         sfs->f_bsize <<= 1;
2383
2384                         osfs.os_blocks >>= 1;
2385                         osfs.os_bfree >>= 1;
2386                         osfs.os_bavail >>= 1;
2387                 }
2388         }
2389
2390         sfs->f_blocks = osfs.os_blocks;
2391         sfs->f_bfree = osfs.os_bfree;
2392         sfs->f_bavail = osfs.os_bavail;
2393         sfs->f_fsid.val[0] = (__u32)fsid;
2394         sfs->f_fsid.val[1] = (__u32)(fsid >> 32);
2395         if (ll_i2info(de->d_inode)->lli_projid)
2396                 return ll_statfs_project(de->d_inode, sfs);
2397
2398         ll_stats_ops_tally(ll_s2sbi(sb), LPROC_LL_STATFS,
2399                            ktime_us_delta(ktime_get(), kstart));
2400
2401         return 0;
2402 }
2403
2404 void ll_inode_size_lock(struct inode *inode)
2405 {
2406         struct ll_inode_info *lli;
2407
2408         LASSERT(!S_ISDIR(inode->i_mode));
2409
2410         lli = ll_i2info(inode);
2411         mutex_lock(&lli->lli_size_mutex);
2412 }
2413
2414 void ll_inode_size_unlock(struct inode *inode)
2415 {
2416         struct ll_inode_info *lli;
2417
2418         lli = ll_i2info(inode);
2419         mutex_unlock(&lli->lli_size_mutex);
2420 }
2421
2422 void ll_update_inode_flags(struct inode *inode, int ext_flags)
2423 {
2424         /* do not clear encryption flag */
2425         ext_flags |= ll_inode_to_ext_flags(inode->i_flags) & LUSTRE_ENCRYPT_FL;
2426         inode->i_flags = ll_ext_to_inode_flags(ext_flags);
2427         if (ext_flags & LUSTRE_PROJINHERIT_FL)
2428                 ll_file_set_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2429         else
2430                 ll_file_clear_flag(ll_i2info(inode), LLIF_PROJECT_INHERIT);
2431 }
2432
2433 int ll_update_inode(struct inode *inode, struct lustre_md *md)
2434 {
2435         struct ll_inode_info *lli = ll_i2info(inode);
2436         struct mdt_body *body = md->body;
2437         struct ll_sb_info *sbi = ll_i2sbi(inode);
2438         int rc = 0;
2439
2440         if (body->mbo_valid & OBD_MD_FLEASIZE) {
2441                 rc = cl_file_inode_init(inode, md);
2442                 if (rc)
2443                         return rc;
2444         }
2445
2446         if (S_ISDIR(inode->i_mode)) {
2447                 rc = ll_update_lsm_md(inode, md);
2448                 if (rc != 0)
2449                         return rc;
2450         }
2451
2452         if (body->mbo_valid & OBD_MD_FLACL)
2453                 lli_replace_acl(lli, md);
2454
2455         inode->i_ino = cl_fid_build_ino(&body->mbo_fid1,
2456                                         sbi->ll_flags & LL_SBI_32BIT_API);
2457         inode->i_generation = cl_fid_build_gen(&body->mbo_fid1);
2458
2459         if (body->mbo_valid & OBD_MD_FLATIME) {
2460                 if (body->mbo_atime > inode->i_atime.tv_sec)
2461                         inode->i_atime.tv_sec = body->mbo_atime;
2462                 lli->lli_atime = body->mbo_atime;
2463         }
2464
2465         if (body->mbo_valid & OBD_MD_FLMTIME) {
2466                 if (body->mbo_mtime > inode->i_mtime.tv_sec) {
2467                         CDEBUG(D_INODE,
2468                                "setting ino %lu mtime from %lld to %llu\n",
2469                                inode->i_ino, (s64)inode->i_mtime.tv_sec,
2470                                body->mbo_mtime);
2471                         inode->i_mtime.tv_sec = body->mbo_mtime;
2472                 }
2473                 lli->lli_mtime = body->mbo_mtime;
2474         }
2475
2476         if (body->mbo_valid & OBD_MD_FLCTIME) {
2477                 if (body->mbo_ctime > inode->i_ctime.tv_sec)
2478                         inode->i_ctime.tv_sec = body->mbo_ctime;
2479                 lli->lli_ctime = body->mbo_ctime;
2480         }
2481
2482         if (body->mbo_valid & OBD_MD_FLBTIME)
2483                 lli->lli_btime = body->mbo_btime;
2484
2485         /* Clear i_flags to remove S_NOSEC before permissions are updated */
2486         if (body->mbo_valid & OBD_MD_FLFLAGS)
2487                 ll_update_inode_flags(inode, body->mbo_flags);
2488         if (body->mbo_valid & OBD_MD_FLMODE)
2489                 inode->i_mode = (inode->i_mode & S_IFMT) |
2490                                 (body->mbo_mode & ~S_IFMT);
2491
2492         if (body->mbo_valid & OBD_MD_FLTYPE)
2493                 inode->i_mode = (inode->i_mode & ~S_IFMT) |
2494                                 (body->mbo_mode & S_IFMT);
2495
2496         LASSERT(inode->i_mode != 0);
2497         if (body->mbo_valid & OBD_MD_FLUID)
2498                 inode->i_uid = make_kuid(&init_user_ns, body->mbo_uid);
2499         if (body->mbo_valid & OBD_MD_FLGID)
2500                 inode->i_gid = make_kgid(&init_user_ns, body->mbo_gid);
2501         if (body->mbo_valid & OBD_MD_FLPROJID)
2502                 lli->lli_projid = body->mbo_projid;
2503         if (body->mbo_valid & OBD_MD_FLNLINK)
2504                 set_nlink(inode, body->mbo_nlink);
2505         if (body->mbo_valid & OBD_MD_FLRDEV)
2506                 inode->i_rdev = old_decode_dev(body->mbo_rdev);
2507
2508         if (body->mbo_valid & OBD_MD_FLID) {
2509                 /* FID shouldn't be changed! */
2510                 if (fid_is_sane(&lli->lli_fid)) {
2511                         LASSERTF(lu_fid_eq(&lli->lli_fid, &body->mbo_fid1),
2512                                  "Trying to change FID "DFID
2513                                  " to the "DFID", inode "DFID"(%p)\n",
2514                                  PFID(&lli->lli_fid), PFID(&body->mbo_fid1),
2515                                  PFID(ll_inode2fid(inode)), inode);
2516                 } else {
2517                         lli->lli_fid = body->mbo_fid1;
2518                 }
2519         }
2520
2521         LASSERT(fid_seq(&lli->lli_fid) != 0);
2522
2523         lli->lli_attr_valid = body->mbo_valid;
2524         if (body->mbo_valid & OBD_MD_FLSIZE) {
2525                 i_size_write(inode, body->mbo_size);
2526
2527                 CDEBUG(D_VFSTRACE, "inode="DFID", updating i_size %llu\n",
2528                        PFID(ll_inode2fid(inode)),
2529                        (unsigned long long)body->mbo_size);
2530
2531                 if (body->mbo_valid & OBD_MD_FLBLOCKS)
2532                         inode->i_blocks = body->mbo_blocks;
2533         } else {
2534                 if (body->mbo_valid & OBD_MD_FLLAZYSIZE)
2535                         lli->lli_lazysize = body->mbo_size;
2536                 if (body->mbo_valid & OBD_MD_FLLAZYBLOCKS)
2537                         lli->lli_lazyblocks = body->mbo_blocks;
2538         }
2539
2540         if (body->mbo_valid & OBD_MD_TSTATE) {
2541                 /* Set LLIF_FILE_RESTORING if restore ongoing and
2542                  * clear it when done to ensure to start again
2543                  * glimpsing updated attrs
2544                  */
2545                 if (body->mbo_t_state & MS_RESTORE)
2546                         ll_file_set_flag(lli, LLIF_FILE_RESTORING);
2547                 else
2548                         ll_file_clear_flag(lli, LLIF_FILE_RESTORING);
2549         }
2550
2551         return 0;
2552 }
2553
2554 int ll_read_inode2(struct inode *inode, void *opaque)
2555 {
2556         struct lustre_md *md = opaque;
2557         struct ll_inode_info *lli = ll_i2info(inode);
2558         int     rc;
2559         ENTRY;
2560
2561         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p)\n",
2562                PFID(&lli->lli_fid), inode);
2563
2564         /* Core attributes from the MDS first.  This is a new inode, and
2565          * the VFS doesn't zero times in the core inode so we have to do
2566          * it ourselves.  They will be overwritten by either MDS or OST
2567          * attributes - we just need to make sure they aren't newer.
2568          */
2569         inode->i_mtime.tv_sec = 0;
2570         inode->i_atime.tv_sec = 0;
2571         inode->i_ctime.tv_sec = 0;
2572         inode->i_rdev = 0;
2573         rc = ll_update_inode(inode, md);
2574         if (rc != 0)
2575                 RETURN(rc);
2576
2577         /* OIDEBUG(inode); */
2578
2579 #ifdef HAVE_BACKING_DEV_INFO
2580         /* initializing backing dev info. */
2581         inode->i_mapping->backing_dev_info = &s2lsi(inode->i_sb)->lsi_bdi;
2582 #endif
2583         if (S_ISREG(inode->i_mode)) {
2584                 struct ll_sb_info *sbi = ll_i2sbi(inode);
2585                 inode->i_op = &ll_file_inode_operations;
2586                 inode->i_fop = sbi->ll_fop;
2587                 inode->i_mapping->a_ops = (struct address_space_operations *)&ll_aops;
2588                 EXIT;
2589         } else if (S_ISDIR(inode->i_mode)) {
2590                 inode->i_op = &ll_dir_inode_operations;
2591                 inode->i_fop = &ll_dir_operations;
2592                 EXIT;
2593         } else if (S_ISLNK(inode->i_mode)) {
2594                 inode->i_op = &ll_fast_symlink_inode_operations;
2595                 EXIT;
2596         } else {
2597                 inode->i_op = &ll_special_inode_operations;
2598
2599                 init_special_inode(inode, inode->i_mode,
2600                                    inode->i_rdev);
2601
2602                 EXIT;
2603         }
2604
2605         return 0;
2606 }
2607
2608 void ll_delete_inode(struct inode *inode)
2609 {
2610         struct ll_inode_info *lli = ll_i2info(inode);
2611         struct address_space *mapping = &inode->i_data;
2612         unsigned long nrpages;
2613         unsigned long flags;
2614
2615         ENTRY;
2616
2617         if (S_ISREG(inode->i_mode) && lli->lli_clob != NULL) {
2618                 /* It is last chance to write out dirty pages,
2619                  * otherwise we may lose data while umount.
2620                  *
2621                  * If i_nlink is 0 then just discard data. This is safe because
2622                  * local inode gets i_nlink 0 from server only for the last
2623                  * unlink, so that file is not opened somewhere else
2624                  */
2625                 cl_sync_file_range(inode, 0, OBD_OBJECT_EOF, inode->i_nlink ?
2626                                    CL_FSYNC_LOCAL : CL_FSYNC_DISCARD, 1);
2627         }
2628         truncate_inode_pages_final(mapping);
2629
2630         /* Workaround for LU-118: Note nrpages may not be totally updated when
2631          * truncate_inode_pages() returns, as there can be a page in the process
2632          * of deletion (inside __delete_from_page_cache()) in the specified
2633          * range. Thus mapping->nrpages can be non-zero when this function
2634          * returns even after truncation of the whole mapping.  Only do this if
2635          * npages isn't already zero.
2636          */
2637         nrpages = mapping->nrpages;
2638         if (nrpages) {
2639                 ll_xa_lock_irqsave(&mapping->i_pages, flags);
2640                 nrpages = mapping->nrpages;
2641                 ll_xa_unlock_irqrestore(&mapping->i_pages, flags);
2642         } /* Workaround end */
2643
2644         LASSERTF(nrpages == 0, "%s: inode="DFID"(%p) nrpages=%lu, "
2645                  "see https://jira.whamcloud.com/browse/LU-118\n",
2646                  ll_i2sbi(inode)->ll_fsname,
2647                  PFID(ll_inode2fid(inode)), inode, nrpages);
2648
2649         ll_clear_inode(inode);
2650         clear_inode(inode);
2651
2652         EXIT;
2653 }
2654
2655 int ll_iocontrol(struct inode *inode, struct file *file,
2656                  unsigned int cmd, unsigned long arg)
2657 {
2658         struct ll_sb_info *sbi = ll_i2sbi(inode);
2659         struct ptlrpc_request *req = NULL;
2660         int rc, flags = 0;
2661         ENTRY;
2662
2663         switch (cmd) {
2664         case FS_IOC_GETFLAGS: {
2665                 struct mdt_body *body;
2666                 struct md_op_data *op_data;
2667
2668                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
2669                                              0, 0, LUSTRE_OPC_ANY,
2670                                              NULL);
2671                 if (IS_ERR(op_data))
2672                         RETURN(PTR_ERR(op_data));
2673
2674                 op_data->op_valid = OBD_MD_FLFLAGS;
2675                 rc = md_getattr(sbi->ll_md_exp, op_data, &req);
2676                 ll_finish_md_op_data(op_data);
2677                 if (rc) {
2678                         CERROR("%s: failure inode "DFID": rc = %d\n",
2679                                sbi->ll_md_exp->exp_obd->obd_name,
2680                                PFID(ll_inode2fid(inode)), rc);
2681                         RETURN(-abs(rc));
2682                 }
2683
2684                 body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
2685
2686                 flags = body->mbo_flags;
2687
2688                 ptlrpc_req_finished(req);
2689
2690                 RETURN(put_user(flags, (int __user *)arg));
2691         }
2692         case FS_IOC_SETFLAGS: {
2693                 struct iattr *attr;
2694                 struct md_op_data *op_data;
2695                 struct cl_object *obj;
2696                 struct fsxattr fa = { 0 };
2697
2698                 if (get_user(flags, (int __user *)arg))
2699                         RETURN(-EFAULT);
2700
2701                 fa.fsx_projid = ll_i2info(inode)->lli_projid;
2702                 if (flags & LUSTRE_PROJINHERIT_FL)
2703                         fa.fsx_xflags = FS_XFLAG_PROJINHERIT;
2704
2705                 rc = ll_ioctl_check_project(inode, &fa);
2706                 if (rc)
2707                         RETURN(rc);
2708
2709                 op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
2710                                              LUSTRE_OPC_ANY, NULL);
2711                 if (IS_ERR(op_data))
2712                         RETURN(PTR_ERR(op_data));
2713
2714                 op_data->op_attr_flags = flags;
2715                 op_data->op_xvalid |= OP_XVALID_FLAGS;
2716                 rc = md_setattr(sbi->ll_md_exp, op_data, NULL, 0, &req);
2717                 ll_finish_md_op_data(op_data);
2718                 ptlrpc_req_finished(req);
2719                 if (rc)
2720                         RETURN(rc);
2721
2722                 ll_update_inode_flags(inode, flags);
2723
2724                 obj = ll_i2info(inode)->lli_clob;
2725                 if (obj == NULL)
2726                         RETURN(0);
2727
2728                 OBD_ALLOC_PTR(attr);
2729                 if (attr == NULL)
2730                         RETURN(-ENOMEM);
2731
2732                 rc = cl_setattr_ost(obj, attr, OP_XVALID_FLAGS, flags);
2733
2734                 OBD_FREE_PTR(attr);
2735                 RETURN(rc);
2736         }
2737         default:
2738                 RETURN(-ENOSYS);
2739         }
2740
2741         RETURN(0);
2742 }
2743
2744 int ll_flush_ctx(struct inode *inode)
2745 {
2746         struct ll_sb_info  *sbi = ll_i2sbi(inode);
2747
2748         CDEBUG(D_SEC, "flush context for user %d\n",
2749                from_kuid(&init_user_ns, current_uid()));
2750
2751         obd_set_info_async(NULL, sbi->ll_md_exp,
2752                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2753                            0, NULL, NULL);
2754         obd_set_info_async(NULL, sbi->ll_dt_exp,
2755                            sizeof(KEY_FLUSH_CTX), KEY_FLUSH_CTX,
2756                            0, NULL, NULL);
2757         return 0;
2758 }
2759
2760 /* umount -f client means force down, don't save state */
2761 void ll_umount_begin(struct super_block *sb)
2762 {
2763         struct ll_sb_info *sbi = ll_s2sbi(sb);
2764         struct obd_device *obd;
2765         struct obd_ioctl_data *ioc_data;
2766         int cnt;
2767         ENTRY;
2768
2769         CDEBUG(D_VFSTRACE, "VFS Op: superblock %p count %d active %d\n", sb,
2770                sb->s_count, atomic_read(&sb->s_active));
2771
2772         obd = class_exp2obd(sbi->ll_md_exp);
2773         if (obd == NULL) {
2774                 CERROR("Invalid MDC connection handle %#llx\n",
2775                        sbi->ll_md_exp->exp_handle.h_cookie);
2776                 EXIT;
2777                 return;
2778         }
2779         obd->obd_force = 1;
2780
2781         obd = class_exp2obd(sbi->ll_dt_exp);
2782         if (obd == NULL) {
2783                 CERROR("Invalid LOV connection handle %#llx\n",
2784                        sbi->ll_dt_exp->exp_handle.h_cookie);
2785                 EXIT;
2786                 return;
2787         }
2788         obd->obd_force = 1;
2789
2790         OBD_ALLOC_PTR(ioc_data);
2791         if (ioc_data) {
2792                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_md_exp,
2793                               sizeof *ioc_data, ioc_data, NULL);
2794
2795                 obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_dt_exp,
2796                               sizeof *ioc_data, ioc_data, NULL);
2797
2798                 OBD_FREE_PTR(ioc_data);
2799         }
2800
2801         /* Really, we'd like to wait until there are no requests outstanding,
2802          * and then continue.  For now, we just periodically checking for vfs
2803          * to decrement mnt_cnt and hope to finish it within 10sec.
2804          */
2805         cnt = 10;
2806         while (cnt > 0 &&
2807                !may_umount(sbi->ll_mnt.mnt)) {
2808                 ssleep(1);
2809                 cnt -= 1;
2810         }
2811
2812         EXIT;
2813 }
2814
2815 int ll_remount_fs(struct super_block *sb, int *flags, char *data)
2816 {
2817         struct ll_sb_info *sbi = ll_s2sbi(sb);
2818         char *profilenm = get_profile_name(sb);
2819         int err;
2820         __u32 read_only;
2821
2822         if ((*flags & MS_RDONLY) != (sb->s_flags & SB_RDONLY)) {
2823                 read_only = *flags & MS_RDONLY;
2824                 err = obd_set_info_async(NULL, sbi->ll_md_exp,
2825                                          sizeof(KEY_READ_ONLY),
2826                                          KEY_READ_ONLY, sizeof(read_only),
2827                                          &read_only, NULL);
2828                 if (err) {
2829                         LCONSOLE_WARN("Failed to remount %s %s (%d)\n",
2830                                       profilenm, read_only ?
2831                                       "read-only" : "read-write", err);
2832                         return err;
2833                 }
2834
2835                 if (read_only)
2836                         sb->s_flags |= SB_RDONLY;
2837                 else
2838                         sb->s_flags &= ~SB_RDONLY;
2839
2840                 if (sbi->ll_flags & LL_SBI_VERBOSE)
2841                         LCONSOLE_WARN("Remounted %s %s\n", profilenm,
2842                                       read_only ?  "read-only" : "read-write");
2843         }
2844         return 0;
2845 }
2846
2847 /**
2848  * Cleanup the open handle that is cached on MDT-side.
2849  *
2850  * For open case, the client side open handling thread may hit error
2851  * after the MDT grant the open. Under such case, the client should
2852  * send close RPC to the MDT as cleanup; otherwise, the open handle
2853  * on the MDT will be leaked there until the client umount or evicted.
2854  *
2855  * In further, if someone unlinked the file, because the open handle
2856  * holds the reference on such file/object, then it will block the
2857  * subsequent threads that want to locate such object via FID.
2858  *
2859  * \param[in] sb        super block for this file-system
2860  * \param[in] open_req  pointer to the original open request
2861  */
2862 void ll_open_cleanup(struct super_block *sb, struct ptlrpc_request *open_req)
2863 {
2864         struct mdt_body                 *body;
2865         struct md_op_data               *op_data;
2866         struct ptlrpc_request           *close_req = NULL;
2867         struct obd_export               *exp       = ll_s2sbi(sb)->ll_md_exp;
2868         ENTRY;
2869
2870         body = req_capsule_server_get(&open_req->rq_pill, &RMF_MDT_BODY);
2871         OBD_ALLOC_PTR(op_data);
2872         if (op_data == NULL) {
2873                 CWARN("%s: cannot allocate op_data to release open handle for "
2874                       DFID"\n", ll_s2sbi(sb)->ll_fsname, PFID(&body->mbo_fid1));
2875
2876                 RETURN_EXIT;
2877         }
2878
2879         op_data->op_fid1 = body->mbo_fid1;
2880         op_data->op_open_handle = body->mbo_open_handle;
2881         op_data->op_mod_time = ktime_get_real_seconds();
2882         md_close(exp, op_data, NULL, &close_req);
2883         ptlrpc_req_finished(close_req);
2884         ll_finish_md_op_data(op_data);
2885
2886         EXIT;
2887 }
2888
2889 int ll_prep_inode(struct inode **inode, struct ptlrpc_request *req,
2890                   struct super_block *sb, struct lookup_intent *it)
2891 {
2892         struct ll_sb_info *sbi = NULL;
2893         struct lustre_md md = { NULL };
2894         bool default_lmv_deleted = false;
2895         int rc;
2896
2897         ENTRY;
2898
2899         LASSERT(*inode || sb);
2900         sbi = sb ? ll_s2sbi(sb) : ll_i2sbi(*inode);
2901         rc = md_get_lustre_md(sbi->ll_md_exp, req, sbi->ll_dt_exp,
2902                               sbi->ll_md_exp, &md);
2903         if (rc != 0)
2904                 GOTO(out, rc);
2905
2906         /*
2907          * clear default_lmv only if intent_getattr reply doesn't contain it.
2908          * but it needs to be done after iget, check this early because
2909          * ll_update_lsm_md() may change md.
2910          */
2911         if (it && (it->it_op & (IT_LOOKUP | IT_GETATTR)) &&
2912             S_ISDIR(md.body->mbo_mode) && !md.default_lmv)
2913                 default_lmv_deleted = true;
2914
2915         if (*inode) {
2916                 rc = ll_update_inode(*inode, &md);
2917                 if (rc != 0)
2918                         GOTO(out, rc);
2919         } else {
2920                 LASSERT(sb != NULL);
2921
2922                 /*
2923                  * At this point server returns to client's same fid as client
2924                  * generated for creating. So using ->fid1 is okay here.
2925                  */
2926                 if (!fid_is_sane(&md.body->mbo_fid1)) {
2927                         CERROR("%s: Fid is insane "DFID"\n",
2928                                 sbi->ll_fsname,
2929                                 PFID(&md.body->mbo_fid1));
2930                         GOTO(out, rc = -EINVAL);
2931                 }
2932
2933                 *inode = ll_iget(sb, cl_fid_build_ino(&md.body->mbo_fid1,
2934                                              sbi->ll_flags & LL_SBI_32BIT_API),
2935                                  &md);
2936                 if (IS_ERR(*inode)) {
2937                         lmd_clear_acl(&md);
2938                         rc = IS_ERR(*inode) ? PTR_ERR(*inode) : -ENOMEM;
2939                         *inode = NULL;
2940                         CERROR("new_inode -fatal: rc %d\n", rc);
2941                         GOTO(out, rc);
2942                 }
2943         }
2944
2945         /* Handling piggyback layout lock.
2946          * Layout lock can be piggybacked by getattr and open request.
2947          * The lsm can be applied to inode only if it comes with a layout lock
2948          * otherwise correct layout may be overwritten, for example:
2949          * 1. proc1: mdt returns a lsm but not granting layout
2950          * 2. layout was changed by another client
2951          * 3. proc2: refresh layout and layout lock granted
2952          * 4. proc1: to apply a stale layout */
2953         if (it != NULL && it->it_lock_mode != 0) {
2954                 struct lustre_handle lockh;
2955                 struct ldlm_lock *lock;
2956
2957                 lockh.cookie = it->it_lock_handle;
2958                 lock = ldlm_handle2lock(&lockh);
2959                 LASSERT(lock != NULL);
2960                 if (ldlm_has_layout(lock)) {
2961                         struct cl_object_conf conf;
2962
2963                         memset(&conf, 0, sizeof(conf));
2964                         conf.coc_opc = OBJECT_CONF_SET;
2965                         conf.coc_inode = *inode;
2966                         conf.coc_lock = lock;
2967                         conf.u.coc_layout = md.layout;
2968                         (void)ll_layout_conf(*inode, &conf);
2969                 }
2970                 LDLM_LOCK_PUT(lock);
2971         }
2972
2973         if (default_lmv_deleted)
2974                 ll_update_default_lsm_md(*inode, &md);
2975
2976         /* we may want to apply some policy for foreign file/dir */
2977         if (ll_sbi_has_foreign_symlink(sbi)) {
2978                 rc = ll_manage_foreign(*inode, &md);
2979                 if (rc < 0)
2980                         GOTO(out, rc);
2981         }
2982
2983         GOTO(out, rc = 0);
2984
2985 out:
2986         /* cleanup will be done if necessary */
2987         md_free_lustre_md(sbi->ll_md_exp, &md);
2988
2989         if (rc != 0 && it != NULL && it->it_op & IT_OPEN) {
2990                 ll_intent_drop_lock(it);
2991                 ll_open_cleanup(sb != NULL ? sb : (*inode)->i_sb, req);
2992         }
2993
2994         return rc;
2995 }
2996
2997 int ll_obd_statfs(struct inode *inode, void __user *arg)
2998 {
2999         struct ll_sb_info *sbi = NULL;
3000         struct obd_export *exp;
3001         char *buf = NULL;
3002         struct obd_ioctl_data *data = NULL;
3003         __u32 type;
3004         int len = 0, rc;
3005
3006         if (!inode || !(sbi = ll_i2sbi(inode)))
3007                 GOTO(out_statfs, rc = -EINVAL);
3008
3009         rc = obd_ioctl_getdata(&buf, &len, arg);
3010         if (rc)
3011                 GOTO(out_statfs, rc);
3012
3013         data = (void*)buf;
3014         if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
3015             !data->ioc_pbuf1 || !data->ioc_pbuf2)
3016                 GOTO(out_statfs, rc = -EINVAL);
3017
3018         if (data->ioc_inllen1 != sizeof(__u32) ||
3019             data->ioc_inllen2 != sizeof(__u32) ||
3020             data->ioc_plen1 != sizeof(struct obd_statfs) ||
3021             data->ioc_plen2 != sizeof(struct obd_uuid))
3022                 GOTO(out_statfs, rc = -EINVAL);
3023
3024         memcpy(&type, data->ioc_inlbuf1, sizeof(__u32));
3025         if (type & LL_STATFS_LMV)
3026                 exp = sbi->ll_md_exp;
3027         else if (type & LL_STATFS_LOV)
3028                 exp = sbi->ll_dt_exp;
3029         else
3030                 GOTO(out_statfs, rc = -ENODEV);
3031
3032         rc = obd_iocontrol(IOC_OBD_STATFS, exp, len, buf, NULL);
3033         if (rc)
3034                 GOTO(out_statfs, rc);
3035 out_statfs:
3036         OBD_FREE_LARGE(buf, len);
3037         return rc;
3038 }
3039
3040 /*
3041  * this is normally called in ll_fini_md_op_data(), but sometimes it needs to
3042  * be called early to avoid deadlock.
3043  */
3044 void ll_unlock_md_op_lsm(struct md_op_data *op_data)
3045 {
3046         if (op_data->op_mea2_sem) {
3047                 up_read_non_owner(op_data->op_mea2_sem);
3048                 op_data->op_mea2_sem = NULL;
3049         }
3050
3051         if (op_data->op_mea1_sem) {
3052                 up_read_non_owner(op_data->op_mea1_sem);
3053                 op_data->op_mea1_sem = NULL;
3054         }
3055 }
3056
3057 /* this function prepares md_op_data hint for passing it down to MD stack. */
3058 struct md_op_data *ll_prep_md_op_data(struct md_op_data *op_data,
3059                                       struct inode *i1, struct inode *i2,
3060                                       const char *name, size_t namelen,
3061                                       __u32 mode, enum md_op_code opc,
3062                                       void *data)
3063 {
3064         LASSERT(i1 != NULL);
3065
3066         if (name == NULL) {
3067                 /* Do not reuse namelen for something else. */
3068                 if (namelen != 0)
3069                         return ERR_PTR(-EINVAL);
3070         } else {
3071                 if (namelen > ll_i2sbi(i1)->ll_namelen)
3072                         return ERR_PTR(-ENAMETOOLONG);
3073
3074                 /* "/" is not valid name, but it's allowed */
3075                 if (!lu_name_is_valid_2(name, namelen) &&
3076                     strncmp("/", name, namelen) != 0)
3077                         return ERR_PTR(-EINVAL);
3078         }
3079
3080         if (op_data == NULL)
3081                 OBD_ALLOC_PTR(op_data);
3082
3083         if (op_data == NULL)
3084                 return ERR_PTR(-ENOMEM);
3085
3086         ll_i2gids(op_data->op_suppgids, i1, i2);
3087         op_data->op_fid1 = *ll_inode2fid(i1);
3088         op_data->op_code = opc;
3089
3090         if (S_ISDIR(i1->i_mode)) {
3091                 down_read_non_owner(&ll_i2info(i1)->lli_lsm_sem);
3092                 op_data->op_mea1_sem = &ll_i2info(i1)->lli_lsm_sem;
3093                 op_data->op_mea1 = ll_i2info(i1)->lli_lsm_md;
3094                 op_data->op_default_mea1 = ll_i2info(i1)->lli_default_lsm_md;
3095         }
3096
3097         if (i2) {
3098                 op_data->op_fid2 = *ll_inode2fid(i2);
3099                 if (S_ISDIR(i2->i_mode)) {
3100                         if (i2 != i1) {
3101                                 /* i2 is typically a child of i1, and MUST be
3102                                  * further from the root to avoid deadlocks.
3103                                  */
3104                                 down_read_non_owner(&ll_i2info(i2)->lli_lsm_sem);
3105                                 op_data->op_mea2_sem =
3106                                                 &ll_i2info(i2)->lli_lsm_sem;
3107                         }
3108                         op_data->op_mea2 = ll_i2info(i2)->lli_lsm_md;
3109                 }
3110         } else {
3111                 fid_zero(&op_data->op_fid2);
3112         }
3113
3114         if (ll_i2sbi(i1)->ll_flags & LL_SBI_64BIT_HASH)
3115                 op_data->op_cli_flags |= CLI_HASH64;
3116
3117         if (ll_need_32bit_api(ll_i2sbi(i1)))
3118                 op_data->op_cli_flags |= CLI_API32;
3119
3120         op_data->op_name = name;
3121         op_data->op_namelen = namelen;
3122         op_data->op_mode = mode;
3123         op_data->op_mod_time = ktime_get_real_seconds();
3124         op_data->op_fsuid = from_kuid(&init_user_ns, current_fsuid());
3125         op_data->op_fsgid = from_kgid(&init_user_ns, current_fsgid());
3126         op_data->op_cap = cfs_curproc_cap_pack();
3127         op_data->op_mds = 0;
3128         if ((opc == LUSTRE_OPC_CREATE) && (name != NULL) &&
3129              filename_is_volatile(name, namelen, &op_data->op_mds)) {
3130                 op_data->op_bias |= MDS_CREATE_VOLATILE;
3131         }
3132         op_data->op_data = data;
3133
3134         return op_data;
3135 }
3136
3137 void ll_finish_md_op_data(struct md_op_data *op_data)
3138 {
3139         ll_unlock_md_op_lsm(op_data);
3140         security_release_secctx(op_data->op_file_secctx,
3141                                 op_data->op_file_secctx_size);
3142         llcrypt_free_ctx(op_data->op_file_encctx, op_data->op_file_encctx_size);
3143         OBD_FREE_PTR(op_data);
3144 }
3145
3146 int ll_show_options(struct seq_file *seq, struct dentry *dentry)
3147 {
3148         struct ll_sb_info *sbi;
3149
3150         LASSERT(seq && dentry);
3151         sbi = ll_s2sbi(dentry->d_sb);
3152
3153         if (sbi->ll_flags & LL_SBI_NOLCK)
3154                 seq_puts(seq, ",nolock");
3155
3156         /* "flock" is the default since 2.13, but it wasn't for many years,
3157          * so it is still useful to print this to show it is enabled.
3158          * Start to print "noflock" so it is now clear when flock is disabled.
3159          */
3160         if (sbi->ll_flags & LL_SBI_FLOCK)
3161                 seq_puts(seq, ",flock");
3162         else if (sbi->ll_flags & LL_SBI_LOCALFLOCK)
3163                 seq_puts(seq, ",localflock");
3164         else
3165                 seq_puts(seq, ",noflock");
3166
3167         if (sbi->ll_flags & LL_SBI_USER_XATTR)
3168                 seq_puts(seq, ",user_xattr");
3169
3170         if (sbi->ll_flags & LL_SBI_LAZYSTATFS)
3171                 seq_puts(seq, ",lazystatfs");
3172
3173         if (sbi->ll_flags & LL_SBI_USER_FID2PATH)
3174                 seq_puts(seq, ",user_fid2path");
3175
3176         if (sbi->ll_flags & LL_SBI_ALWAYS_PING)
3177                 seq_puts(seq, ",always_ping");
3178
3179         if (ll_sbi_has_test_dummy_encryption(sbi))
3180                 seq_puts(seq, ",test_dummy_encryption");
3181
3182         if (ll_sbi_has_encrypt(sbi))
3183                 seq_puts(seq, ",encrypt");
3184         else
3185                 seq_puts(seq, ",noencrypt");
3186
3187         if (sbi->ll_flags & LL_SBI_FOREIGN_SYMLINK) {
3188                 seq_puts(seq, ",foreign_symlink=");
3189                 seq_puts(seq, sbi->ll_foreign_symlink_prefix);
3190         }
3191
3192         RETURN(0);
3193 }
3194
3195 /**
3196  * Get obd name by cmd, and copy out to user space
3197  */
3198 int ll_get_obd_name(struct inode *inode, unsigned int cmd, unsigned long arg)
3199 {
3200         struct ll_sb_info *sbi = ll_i2sbi(inode);
3201         struct obd_device *obd;
3202         ENTRY;
3203
3204         if (cmd == OBD_IOC_GETDTNAME)
3205                 obd = class_exp2obd(sbi->ll_dt_exp);
3206         else if (cmd == OBD_IOC_GETMDNAME)
3207                 obd = class_exp2obd(sbi->ll_md_exp);
3208         else
3209                 RETURN(-EINVAL);
3210
3211         if (!obd)
3212                 RETURN(-ENOENT);
3213
3214         if (copy_to_user((void __user *)arg, obd->obd_name,
3215                          strlen(obd->obd_name) + 1))
3216                 RETURN(-EFAULT);
3217
3218         RETURN(0);
3219 }
3220
3221 static char* ll_d_path(struct dentry *dentry, char *buf, int bufsize)
3222 {
3223         char *path = NULL;
3224
3225         struct path p;
3226
3227         p.dentry = dentry;
3228         p.mnt = current->fs->root.mnt;
3229         path_get(&p);
3230         path = d_path(&p, buf, bufsize);
3231         path_put(&p);
3232         return path;
3233 }
3234
3235 void ll_dirty_page_discard_warn(struct page *page, int ioret)
3236 {
3237         char *buf, *path = NULL;
3238         struct dentry *dentry = NULL;
3239         struct inode *inode = page->mapping->host;
3240
3241         /* this can be called inside spin lock so use GFP_ATOMIC. */
3242         buf = (char *)__get_free_page(GFP_ATOMIC);
3243         if (buf != NULL) {
3244                 dentry = d_find_alias(page->mapping->host);
3245                 if (dentry != NULL)
3246                         path = ll_d_path(dentry, buf, PAGE_SIZE);
3247         }
3248
3249         /* The below message is checked in recovery-small.sh test_24b */
3250         CDEBUG(D_WARNING,
3251                "%s: dirty page discard: %s/fid: "DFID"/%s may get corrupted "
3252                "(rc %d)\n", ll_i2sbi(inode)->ll_fsname,
3253                s2lsi(page->mapping->host->i_sb)->lsi_lmd->lmd_dev,
3254                PFID(ll_inode2fid(inode)),
3255                (path && !IS_ERR(path)) ? path : "", ioret);
3256
3257         if (dentry != NULL)
3258                 dput(dentry);
3259
3260         if (buf != NULL)
3261                 free_page((unsigned long)buf);
3262 }
3263
3264 ssize_t ll_copy_user_md(const struct lov_user_md __user *md,
3265                         struct lov_user_md **kbuf)
3266 {
3267         struct lov_user_md      lum;
3268         ssize_t                 lum_size;
3269         ENTRY;
3270
3271         if (copy_from_user(&lum, md, sizeof(lum)))
3272                 RETURN(-EFAULT);
3273
3274         lum_size = ll_lov_user_md_size(&lum);
3275         if (lum_size < 0)
3276                 RETURN(lum_size);
3277
3278         OBD_ALLOC_LARGE(*kbuf, lum_size);
3279         if (*kbuf == NULL)
3280                 RETURN(-ENOMEM);
3281
3282         if (copy_from_user(*kbuf, md, lum_size) != 0) {
3283                 OBD_FREE_LARGE(*kbuf, lum_size);
3284                 RETURN(-EFAULT);
3285         }
3286
3287         RETURN(lum_size);
3288 }
3289
3290 /*
3291  * Compute llite root squash state after a change of root squash
3292  * configuration setting or add/remove of a lnet nid
3293  */
3294 void ll_compute_rootsquash_state(struct ll_sb_info *sbi)
3295 {
3296         struct root_squash_info *squash = &sbi->ll_squash;
3297         int i;
3298         bool matched;
3299         struct lnet_process_id id;
3300
3301         /* Update norootsquash flag */
3302         spin_lock(&squash->rsi_lock);
3303         if (list_empty(&squash->rsi_nosquash_nids))
3304                 sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3305         else {
3306                 /* Do not apply root squash as soon as one of our NIDs is
3307                  * in the nosquash_nids list */
3308                 matched = false;
3309                 i = 0;
3310                 while (LNetGetId(i++, &id) != -ENOENT) {
3311                         if (id.nid == LNET_NID_LO_0)
3312                                 continue;
3313                         if (cfs_match_nid(id.nid, &squash->rsi_nosquash_nids)) {
3314                                 matched = true;
3315                                 break;
3316                         }
3317                 }
3318                 if (matched)
3319                         sbi->ll_flags |= LL_SBI_NOROOTSQUASH;
3320                 else
3321                         sbi->ll_flags &= ~LL_SBI_NOROOTSQUASH;
3322         }
3323         spin_unlock(&squash->rsi_lock);
3324 }
3325
3326 /**
3327  * Parse linkea content to extract information about a given hardlink
3328  *
3329  * \param[in]   ldata      - Initialized linkea data
3330  * \param[in]   linkno     - Link identifier
3331  * \param[out]  parent_fid - The entry's parent FID
3332  * \param[out]  ln         - Entry name destination buffer
3333  *
3334  * \retval 0 on success
3335  * \retval Appropriate negative error code on failure
3336  */
3337 static int ll_linkea_decode(struct linkea_data *ldata, unsigned int linkno,
3338                             struct lu_fid *parent_fid, struct lu_name *ln)
3339 {
3340         unsigned int    idx;
3341         int             rc;
3342         ENTRY;
3343
3344         rc = linkea_init_with_rec(ldata);
3345         if (rc < 0)
3346                 RETURN(rc);
3347
3348         if (linkno >= ldata->ld_leh->leh_reccount)
3349                 /* beyond last link */
3350                 RETURN(-ENODATA);
3351
3352         linkea_first_entry(ldata);
3353         for (idx = 0; ldata->ld_lee != NULL; idx++) {
3354                 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, ln,
3355                                     parent_fid);
3356                 if (idx == linkno)
3357                         break;
3358
3359                 linkea_next_entry(ldata);
3360         }
3361
3362         if (idx < linkno)
3363                 RETURN(-ENODATA);
3364
3365         RETURN(0);
3366 }
3367
3368 /**
3369  * Get parent FID and name of an identified link. Operation is performed for
3370  * a given link number, letting the caller iterate over linkno to list one or
3371  * all links of an entry.
3372  *
3373  * \param[in]     file - File descriptor against which to perform the operation
3374  * \param[in,out] arg  - User-filled structure containing the linkno to operate
3375  *                       on and the available size. It is eventually filled with
3376  *                       the requested information or left untouched on error
3377  *
3378  * \retval - 0 on success
3379  * \retval - Appropriate negative error code on failure
3380  */
3381 int ll_getparent(struct file *file, struct getparent __user *arg)
3382 {
3383         struct inode            *inode = file_inode(file);
3384         struct linkea_data      *ldata;
3385         struct lu_buf            buf = LU_BUF_NULL;
3386         struct lu_name           ln;
3387         struct lu_fid            parent_fid;
3388         __u32                    linkno;
3389         __u32                    name_size;
3390         int                      rc;
3391
3392         ENTRY;
3393
3394         if (!cfs_capable(CAP_DAC_READ_SEARCH) &&
3395             !(ll_i2sbi(inode)->ll_flags & LL_SBI_USER_FID2PATH))
3396                 RETURN(-EPERM);
3397
3398         if (get_user(name_size, &arg->gp_name_size))
3399                 RETURN(-EFAULT);
3400
3401         if (get_user(linkno, &arg->gp_linkno))
3402                 RETURN(-EFAULT);
3403
3404         if (name_size > PATH_MAX)
3405                 RETURN(-EINVAL);
3406
3407         OBD_ALLOC(ldata, sizeof(*ldata));
3408         if (ldata == NULL)
3409                 RETURN(-ENOMEM);
3410
3411         rc = linkea_data_new(ldata, &buf);
3412         if (rc < 0)
3413                 GOTO(ldata_free, rc);
3414
3415         rc = ll_xattr_list(inode, XATTR_NAME_LINK, XATTR_TRUSTED_T, buf.lb_buf,
3416                            buf.lb_len, OBD_MD_FLXATTR);
3417         if (rc < 0)
3418                 GOTO(lb_free, rc);
3419
3420         rc = ll_linkea_decode(ldata, linkno, &parent_fid, &ln);
3421         if (rc < 0)
3422                 GOTO(lb_free, rc);
3423
3424         if (ln.ln_namelen >= name_size)
3425                 GOTO(lb_free, rc = -EOVERFLOW);
3426
3427         if (copy_to_user(&arg->gp_fid, &parent_fid, sizeof(arg->gp_fid)))
3428                 GOTO(lb_free, rc = -EFAULT);
3429
3430         if (copy_to_user(&arg->gp_name, ln.ln_name, ln.ln_namelen))
3431                 GOTO(lb_free, rc = -EFAULT);
3432
3433         if (put_user('\0', arg->gp_name + ln.ln_namelen))
3434                 GOTO(lb_free, rc = -EFAULT);
3435
3436 lb_free:
3437         lu_buf_free(&buf);
3438 ldata_free:
3439         OBD_FREE(ldata, sizeof(*ldata));
3440
3441         RETURN(rc);
3442 }